Ryzen vs i7 (Mainstream); Threadripper vs i9 (HEDT); X299 vs X399/TRX40; Xeon vs Epyc

Oh'well, this means keep last years TDP is more important than offer best possible performance. Forced by the OEM's customers (they, I mean the OEM's are probably happy their buyers won't get more than this). Or rather say AMD needed to this because they offer upgrading to newer chips. Oh well not nice with suddenly changes in specifications due AMD couldn't increase higher clocks within old specifications on 12nm Zen. I hope AMD can keep same spesifications on their coming 7nm chips

And nice for the consumers that AMD increased the TDP for ordinary Ryzen 2nd gen consumer chips. If not they would be very dissapointed. Aka they would get same performance as the professional corporate machines(max 4.1GH boost clocks). Saved by TDP

 

Are you being absurd? All core is still better than last year's and Intel's vPRO doesn't really touch it. Also with security and these going into a mass of PCs, this outperforms Intel's locked offerings significantly, with awesome guarantee of the same chip being available, or drop in 7nm upgrades also on these platforms.

As to 7nm, if you compared density of TSMC to Intel's, then realize the last comparison on density is from before it came out that Intel was using a less dense node to get 10nm out the door, plus Intel ceding EUV lead to TSMC and Samsung, Intel has said TSMC is a superior process. That is confirmed with both Apple and Qualcomm selecting TSMC 7nm over Intel 10nm. Do you even read the news?

Sent from my SM-G900P using Tapatalk

 

Apple choosing AMD over Intel is not very flattering for AMD. That is like a womanizer bragging about his STDs. AMD should not want crApple's endorsement because they are idiots.

 

WTH? I said Apple chose TSMC for production of their ARM chips, just like Qualcomm and AMD, not that they chose AMD CPUs. WTF are you talking about. Here is what I am talking about:
https://www.eetimes.com/document.asp?doc_id=1333657

and this, which is densiity calculations before Intel making their node less dense was announced at the end of July to the beginning of Aug. This is from June 25, 2018

https://www.semiwiki.com/forum/content/7544-7nm-5nm-3nm-logic-current-projected-processes.html

https://wccftech.com/analysis-about-intels-10nm-process/
https://www.semiaccurate.com/2018/08/02/intel-guts-10nm-to-get-it-out-the-door/


Now, the fact Intel is ceding EUV lead to samsung and TSMC shows that the statements hedging to benefit Intel in the WCCFTech article will not occur. Then you have Apple taking their ARM chips, the A-12, to TSMC, same as qualcomm, and Intel scrapping their 5G modem plans after both rejected Intel.
https://www.macrumors.com/2018/07/05/apple-informs-intel-5g-modems-not-needed-iphones/

Then you have ARM attacking Intel's i5-U series:


And you have Intel saying they are trying to hold AMD to 15% market share next year in servers:
https://www.tomshardware.com/news/intel-ceo-amd-server-market,37273.html

That is straight from Intel's former CEO while he was CEO in June.

Edit: and if you didn't hear, AMD is fabbing ALL 7nm designs with TSMC and GF is abandoning their 7nm node and future node shrinks for the moment. But, GF and Intel may be working on 12nm FD-SOI together. The roadmap image was removed.

 

IIRC no one uses Intel node except for themselves since they aren't Pureplay foundry or IDM like Samsung. Except for one news about the 10nm LG fabbing for ARM only cores (not custom uarch like A, M or Kryo) which also failed perhaps 10nm issues. So its already known that Qcomm/Apple use only TSMC or Samsung, plus these Foundries mostly focus on Low power parts like SoCs.

If we see news Samsung is on track for 7nmLPP but EUV is later this year (Risk production) and HVM is not fixed. Also TSMC 7nm is not EUV like GF, Samsung and Intel. 7nm+ is though. And we know how nm marketing works esp 7nm now. Apple ditched Samsung due to high volume of Samsung components in iPhone their prized bread and butter business for TSMC and pour Billions/Millions into them for First Production volume (80% of TSMC volume this time is for Apple) and R&D.

Its a shame that GF left the industry, all 5nm, 3nm is history for them and slowly they will fade over like all other Fabs. Just like Andy Groove mentioned long back the brightest that Intel got, there will be only one true Foundry left (correct me if I'm wrong) in the world as we move towards the end of Moore's Law era. We were already to 3 and now 2 (Not counting Intel) perhaps another license from Samsung in future ? I don't think its possible since CEO change and Mubdala. Its bad for us and a shame that GF failed after absorbing IBM foundries as well along with 14nm /12nm LP (renamed, improved 14nm, just like Intel) of Samsung.

The thing that most people forget is how Intel clocks their CPUs esp at 14nm++ or ++++++ whatever fits. That 9900K at 5.3+ GHz on 8C/16T + 4000MHz DRAM speed is literally the meaning of "Insane" and peak of Semiconductor technology. Well at least to me since the choice and liberty we have today. Unfortunately that will be past as well due to Capitalism beast. Ryzen is already using TR chips and MCM due to costs and clock them at peak already leaving zero room for tweaking or OCing sadly. I doubt we will ever be able to see this again given the current situations from Intel's CEO change, Meltdown/Spectre issues PC market changing landscape, Apple own ARM IP, silicon limits. Hope for the best tho.

And I don't take any sides for any corporate. IMO Its stupid to have feeling towards them. Any sane smart person would rather look at the liberty, choice and geeks in us for Technology. Ryzen and TR are fantastic in core game, PCIE lanes, Price to Perf ratio, kicking Intel's lazy arse but have flaws just like everything, nothing's perfect.

Finally I'm happy to see the progress, at least the IHS is soldered, hopefully it allows Clevo to build better QC Heatsinks, so that before 2020, a.k.a Win7 + BIOS EOL, I can buy a Prema machine which respects freedom and choice for us.

 

First point, you are correct on. I mistakenly said contract for manufacture and was wrong. Instead, it was solely for 5G radios that they lost the contract on, which is still a sizeable contract. But, in April, Apple announced it is looking at replacing all Intel chips in their products. If this is the case, considering Apple holds 8% market share globally, that is still a sizeable loss, but will occur, at the earliest, by 2020.

This year is not the benchmark. Intel said EUV is targeted for 2021, whereas TSMC is doing volume 7nm EUV next year, with Samsung shortly following. Here are more articles, but had you read the articles I already attached, you would know this. Here is Samsung saying EUV is on track for next year for HVM, meaning both TSMC and Samsung will have EUV HVM next year.
https://www.anandtech.com/show/13329/samsung-foundry-updates-8lpu-for-2019
Here is Intel not having it until 2021 and talk of TSMC being first and starting volume next year.
https://www.extremetech.com/computing/276376-intel-reportedly-wont-deploy-euv-lithography-until-2021
Here is GF saying they have dumped 7nm. Also, GF had both DUV and EUV designs.
https://semiengineering.com/gf-puts-7nm-on-hold/

And, I am directly comparing Intel 10nm to other fabs 7nm. So that point is moot if you actually review the materials presented.

As to TSMC and Apple, sure, this year amounts to 75%, I believe it is now, with Q4 being a ramp up of Qualcomm and some on AMD, which AMD production is likely going to ramp Q1 and Q2 for volume, depending on products.

As to GF, they didn't fail, it was a purely fiscal decision. AMD inked with TSMC, GF would have needed like $2B, IIRC, to get 7nm going, and too few were looking to get on 7nm due to the costs for design and mfr. So, the new CEO, appointed this year, pivoted to do 12nm FD-SOI, which is considered like other fabs 10nm (not counting Intel's). So, that is the problem of it being all about the money.

As to AMD having flaws, I've pointed many out. I think you give too much to an unreleased product whose speeds are likely less. Common sense says you add 2 more cores, and the fact Coffee 8700K, 8086K, and 9900K are all on the same 14nm++ process (they only have 3 14nm processes, first was used with broadwell and Skylake, 14nm+ was used with Kaby and SK-X, 14nm++ is Coffee, Whiskey, Cascade, and Cooper). You have coffee and the coffee refresh, which is the 9900K. If using the 8086K as a reference, you are looking at 5.2-5.3GHz. Two extra cores, or 33% more cores, means more heat, and likely clocking 100-200MHz less. But, it is in the ballpark, just a matter of a couple hundred MHz.

And Clevo changing things and the MXM thing, I've given up on power laptops.

 

And you are 100% sure they won't use a possible Intel's 14nm+++ fabrication process for Coffee Lake S refresh ? I mean 100% proof. Has Intel officially said this?

 

https://www.notebookcheck.net/Leake...cade-Lake-X-to-debut-in-Q4-2018.320114.0.html
https://en.wikichip.org/wiki/intel/cores/coffee_lake_r

https://www.anandtech.com/show/1319...oadmap-cooper-lakesp-and-ice-lakesp-confirmed
https://www.extremetech.com/computi...-new-14nm-chips-in-2019-10nm-ice-lake-in-2020

So, I'm going with what these outlets have put out, that Intel doesn't have a 14nm+++ planned. Just a refresh on the 14nm++ process node, which is Coffee Refresh, Cascade (which is coffee) in December, and Cooper (cascade refresh) sometime next year.

 

Rumors (these outlets) ain't 100% proof bruh. Neither the rumors I posted below... The final and trustworthy proof will come from Intel.

Everything We Know About Intel 9000-Series CPUs So Far
Tomshardware.com September 5, 2018 at 7:34 AM

The Intel 9000-series, aka Coffee Lake Refresh, is the last wave of processors to come out of the oven with the chipmaker's aging 14nm process node. The chips are still based on the Coffee Lake architecture but produced under the third iteration (14nm+++) of the 14nm process. As a result, they feature higher operating clocks and more cores.


Intel roadmap leak points to Core i9-9900K release in September-theinquirer.net
A Coffee Lake S refresh of the rest of the Core i5 chips is expected in piecemeal form, with the Core i5-9600K and Core i5-9400 debuting in September, followed by a full refresh throughout the first and second quarters of 2019. The Core i3 range won't get the Coffee Lake S refresh until 2019.

So that's the roadmap, and it would appear that the chips will use Intel's 14nm+++ fabrication process; Coffee Lake S chips use the 14nm++ process and Kaby Lake CPUs has a 14nm+ fabrication. This again suggests Intel is still some way off from creating its first 10nm chips, with it instead eking out as much as possible of out the 14nm process.


As a curisity... Whiskey Lake - Microarchitectures - Intel (wikichip.org)

 

https://en.wikichip.org/w/index.php?title=14_nm+++_process&action=edit&redlink=1
Also, if you say they are not 100%, you cannot believe that it exists either! Otherwise, you are committing a logical fallacy.

Like this contradiction here: https://www.fudzilla.com/news/pc-hardware/46938-intel-about-to-release-a-lake-of-whiskey

third gen = 14nm++, not 14nm+++. Wikichips, which says 14nm+++, also shows ZERO information on it. Why? Because Intel has said NOTHING about a new process. If there was a new process, wouldn't we know about it, especially considering we are entering into the final month or so before release? Has Intel ever kept a new process node secret up to that point before release?

Edit:

https://sports.yahoo.com/intel-corp-reveals-detail-upcoming-010700626.html
This image is from last April. You can see, word of 14nm++ was there since then, 6months ahead of Coffee Lake. Now, you have articles saying will they or won't they on a 14nm+++. Kinda funny to be so tight lipped, especially when they clearly show they have a 10nm+ and 10nm++ planned. And yet, you are suggesting they have 14nm+++ up their sleeve. Think about that for a minute.

Edit 2: Sorry, 14nm++ was shown in an image for the March 30, 2017 briefing. https://techreport.com/review/31660/intel-defends-its-process-technology-leadership-at-14nm-and-10nm
The 10nm+ and 10nm++ was not shown until may or september shows. So, same slide, just without the 10+ and 10++, as seen in this techreport link.

 

I don't know what to say but the websites write a lot. From the link you posted above...
Intel’s Entire 2018, 2019 CPU Roadmap May Have Just Leaked-extremetech.com
"If Kaby Lake was built on 14nm+ and Coffee Lake on 14nm++, the 9th Generation family would be built on 14nm+++, assuming Intel even has a defined process node extension in the first place. Any chips that launch next year would be built on 14nm++++, which is the point at which I start asking if we should be using superscripts or exponential notation"

Intel roadmap leak points to Core i9-9900K release in September-theinquirer.net
"So that's the roadmap, and it would appear that the chips will use Intel's 14nm+++ fabrication process; Coffee Lake S chips use the 14nm++ process and Kaby Lake CPUs has a 14nm+ fabrication. This again suggests Intel is still some way off from creating its first 10nm chips, with it instead eking out as much as possible of out the 14nm process"

"As such, we doubt we'll see 10nm based Intel processors much earlier than 2020. We can expect either a mild performance hike or improvement in power consumption for the Coffee Lake S refresh, but we don't expect any serious increases in processing power or core counts unless Intel knocks out some new special edition chips."

And from tomshardware.com who posted this 3 days ago.. Everything We Know About Intel 9000-Series CPUs So Far
"The Intel 9000-series, aka Coffee Lake Refresh, is the last wave of processors to come out of the oven with the chipmaker's aging 14nm process node. The chips are still based on the Coffee Lake architecture but produced under the third iteration (14nm+++) of the 14nm process. As a result, they feature higher operating clocks and more cores"

Not me who have written those articles. Take it for what it is.

But anyway/anyhow... The facts or shall we say rumors say 33% more cores with Coffee Lake S refresh over Coffee and within Intels "so called" 95W TDP design. This with a lot higher boost clocks. Even with use of solder under the IHS... I mean 33% more cores will/should push the chips well above Intel's own defined, high-complexity workload. Maybe saved by the decrease of base clock down from 3.7 to 3.6GHz ? I don't know. For the records... The binned i7-8086K have 4.0GHz as base clock.

 

Now, for the first article, can you tell sarcasm while reading? Read the sentence after it. Also, notice the image in the roadmap, saying Q1 2019. If the chip is coming out in September or October, do you see red flags on that roadmap?

As to Tom's, I think that they heard third iteration and assumed 14+++, instead of the first iteration being 14nm, second iteration being 14nm+, and the third iteration being 14nm++, which considering they said "of the 14nm process," 14nm is usually included. Considering Intel announced 14nm++ over six months before first use, and Intel still has said nothing now, these authors are making assumptions, just like you and I are both doing.

On the performance, I already have given my opinion on it. It will be good, and at minimum the top mainstream chip until at least April. If AMD doesn't deliver on 7nm, then it will be there until after the mainstream chip 2019 (yes, from what is known of 10nm from rumors, Intel's 14nm++ outperforms even 10nm+ according to Intel's own slide from last year). That would be either AMD or Intel's mainstream chip in 2020.

As to clocks, not surprising because soldered. That and they are clocking every chip much higher at stock (likely because AMD did that, leaving less room on the table, as Intel is this time, which is fine in my opinion). But, the clocks are impressive, and as I said previously, I do not believe you will get more than 5.3GHz other than validation on air. I'm actually guessing 5.0-5.1GHz all core for OC daily driver. Now, with the 8086K pushing 100-200MHz higher, and that being more likely a Coffee refresh, I could be on the low side by 100-200MHz. That would be 5.1GHz, or up to 5.3GHz even (although really ****ing skeptical of that for the midway point, instead thinking it is top 20% or higher that get 5.3 or 5.4 when binned). And, the well above thing, I doubt. Solder I equivocate to using LM, at least performance within a couple degrees. I think Intel just took more of what they normally leave on the table, and that performance is as I just laid out.

As to the drop in base clock, I say that is the TDP crap they pull sometime, like TDP at base clock = 95W or something like that. That one video I posted awhile back from AdoredTV was actually talking about that for why the Intel scores ranged from 1200 up to 1500 for the coffeelake chips. Setting aside presentation, what he pointed to was that ODMs that designed on spec had lower boost times before dropping frequency at stock settings. That is why some reviewers came up with the really low scores I referenced. It wasn't the reviewer, it was the cooling solution in the computer being reviewed with the chip. Better cooled custom built PCs by the other reviewers all got scores like yours did. But, that is why the maker of the video said Intel wasn't being quite straight with consumers and used "con job" in the title of the video.

But, still, with better IPC than the current Zen+ second gen ryzen chips (about 4% IPC, depending on task, whereas it was 7-10% against first gen, depending on task), plus 25% more all core overclock (4.1 on Ryzen vs 5.1 on Intel), if my estimate is correct, and you wind up with approximately 30% more performance over an AMD 2700X, you pay 33% more for that CPU, give or take a bit, and you have stronger single core performance for games and certain single thread programs or lightly threaded tasks. Even with SMT scaling better than HT, it really doesn't change the equation, the 8-core Intel mainstream will be better than a 2700X, and if correct on performance, the price per performance may suggest that if you have the money, get the Intel chip on that basis.

Now, with the 1920X on sale so low, and AMD trying to clear inventory, if you need more PCIe (mainly storage here), and need something with more cores, the 12 core, with 50 percent more cores, may outperform the Intel 8-core at heavily multi threaded tasks. But, it loses in many categories. So unless buying to get on HEDT (which then buy the MSI MEG Creation board for better power delivery for future chips, for what is currently offered) or because you have a need, majority of people will get more benefit from the 9900K.

 

I know you can use a hell lot more power than stated 95w TDP.

Here is the review of Coffee from Anandtech.com

Power Consumption
For our regular readers, the topic of power consumption has been an interesting one as of late. For the most part, Intel’s consumer processors have been under their expected power consumption, but the recent Skylake-X processors seem to have put that notion out to sea, with numbers almost 20% above what is expected at full load.

The Thermal Design Power (TDP) of a processor is the capability required to adequately cool that processor - while it is not the exact power consumption, is a rough indication of how much power a processor is likely to consume. Higher cooling requirements give to a higher TDP, which naturally fit into a chip that consumes more power. Our last review of consumer processors, the Kaby Lake 7th Generation chips, showed that the Core i7-7700K consumed pretty much exactly the TDP of the chip, while the Core i5 processors came in under their TDP rating by a large margin. The Coffee Lake processors follow this trend.

The Core i7-8700K has a TDP of 95W, but consumes 86.2W at full load, of which the cores account for 78.6W. The rest of the power is consumed mostly by the uncore and the memory controller.

The Core i5-8400 is rated at 65W, and consumes only 49.3W at full load, of which 41.7W is from the cores. That leaves 7.6W on the table for the uncore and memory controller, which is almost identical to that of the Core i7-8700K, showing the similarity in design.

 

Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc
http://forum.notebookreview.com/thr...eons-hpe-directly-recommends-amd-epyc.822772/

 

How the 32-core Ryzen Threadripper is suddenly a lot better at gaming Pcworld.com | Sept 12,2018
A bug in Nvidia's drivers drastically hurt the 32-core Threadripper 2990WX's performance...

When we reviewed AMD's CPU Colossus, the 32-core Threadripper 2990WX, it had one weakness: gaming. But it looks like the fault lay with Nvidia, not AMD, as a driver update appears to fix the problem.

PCPer.com testing broke the news, finding significant increases in performance in several games with the updated drivers. “Almost every title we tested saw some sort of frame rate increase, up to 78% for the most improved title, F1 2017,” wrote PCPer’s Ken Addison on the new drivers.

 

Intel is Serving Major Xeon Discounts to Combat AMD EPYC
By Patrick Kennedy - September 14, 2018
https://www.servethehome.com/intel-is-serving-major-xeon-discounts-to-combat-amd-epyc/

"For the past few years, Intel has not been known for giving large discounts unless you are big enough to buy direct.

Intel may have a massive amount of Xeon revenue, but unless you are one of the large hyper-scalers, or a major OEM, or are building a large cluster, discounts on CPUs have been relatively small.

We have been hearing through a number of customers that Intel is finally engaging in significant discounts to hold AMD EPYC at bay.

Spoiler

Getting Intel Xeon Discounts
From what we have been hearing from a number of customers, discounts are happening at quantity levels *well* below 1000 CPUs. Discounts or incentives are reaching well into the double-digit percentages. For some customers, if they are evaluating an AMD EPYC purchase versus a competitive Intel Xeon part, Intel is more or less matching the price.

This is not happening on frequency optimized SKUs as much since the licensing savings can be hundreds of thousands of dollars per machine. Intel knows that AMD EPYC is targeting this segment as heavily in this generation so the discounts are happening more often in the mainstream higher core count parts.

At VMworld this year, we heard this from a number of our readers and vendors at the show. The behavior shift has been dramatic in that it was a topic of a number of discussions.

A key trigger seems to be an organization’s willingness to adopt AMD EPYC. Our advice, if you are buying even as few as 50-100 servers, is to get an AMD EPYC system quote from your reseller. Doing so seems to be the trigger for Intel’s discount approvals.

Impacts of the Intel Xeon Discounts
File this one under a “no duh” bit. One of the most common forms of dealing with a new market entrant is competing on price. What this also means is that if you are buying, even 50-100 servers for a virtualization cluster, you may also want to price out AMD EPYC servers. AMD EPYC has such aggressive pricing that Intel may decide to discount accordingly to get a deal even in the low hundreds of server ranges.

Taking a step back, this makes sense. There is likely a low single-digit percentage of swing purchases (e.g. ones that will consider Xeon and EPYC) in this generation. If Intel needs to discount on say 1-2% of Xeon sales, but it can put more pressure on AMD EPYC, then this is textbook competitive pricing strategy.

The other major impact is that as you read future reviews, do not take AMD EPYC versus Intel Xeon Scalable pricing at face value. Intel already has discounts for its largest customers. It also can afford to take a mid-list price/ mid-discount or high list price/ high discount competitive pricing and discounting strategy. Unlike what we saw a year ago, Intel Xeon RCP is not as firm.

For buyers, Intel was not as willing to discount in Q1 2018. As systems like the AMD EPYC Powered Dell EMC PowerEdge Servers Are Here and our Dell EMC PowerEdge R7415 Review, AMD has only been available from most Top 5 vendors over the past few months. At the same time, it may pay to shop Intel versus AMD as we are hearing EPYC pricing may help customers at risk of switching get better pricing from Intel even if they do not buy EPYC.

Final Words
Price is not the only factor when determining a server purchase. Indeed, servers with a lot of RAM, many drives, or a number of accelerators will see CPU pricing be relatively small in the overall TCO picture. We highlighted this as part of our DeepLearning10 and DeepLearning11 builds. At the same time, this market behavior is both classic and significant. It shows that Intel is feeling the AMD EPYC threat. For customers, leaning into this competitive environment and getting a competitive quote may be a small step to save tens of thousands (or more) on your next server cluster purchase.

Intel orders vendors to offer Major Xeon Discounts when clients ask for AMD EPYC
https://www.reddit.com/r/Amd/comments/9fwb7j/intel_orders_vendors_to_offer_major_xeon/

SavageAvidLentil 841 points 13 hours ago
"AMD should launch a flash campaign "Ask for EPYC" plastered everywhere. Even unbranded ads like 'What is EPYC? ask your vendor'. A few articles titled "Why you should ask for EPYC - even if you're going Intel" etc...

If Intel is willing to burn money for 3 years until they catch up with process in hopes of derailing AMD's revenue projections - let's build a bonfire"

 

Usually offering steep discounts while in a shortage amplifies that shortage. So, Intel has to increase Xeon production now while also getting ready for a paper launch of mainstream CPUs AND getting ready for the Cascade-SP/X launch around December to Q1. Damn! That is a lot of being hit by all sides.

 

So, did the math on that. At 4GHz on all core per 2P 7601 in the recent Der8auer build, with 2667MHz memory, Roman got 10460 in Cinebench R15. If this score is correct, it is 20% faster than a 2P solution with a single processor, which suggests at least 10-15% of that 20% is IPC lift, part is not going off socket, and a CPU with boost clocks around 4GHz, which for 64 cores is AWESOME. Just throwing that out there.

 

Yup, but it's cloudy, early, IDK if it's a total fake, but when they used a shot of a previously published AMD CPU shot it took it's credibility down quite a bit - they should have left out that shot - if it's real at all.

Kinda early to be getting leaks, if anything the performance might go up / down between now and release.

It's possible that the first 7nm CPU's / GPU's on a new process might be a mixed bag, depending on AMD / TSMC taking the time for a few turn around's before production - likely as not production will be rushed and forced into a strict schedule that precludes multiple tuning turns - due to GF removing 7nm production potential from the market.

It may be everyone (chip makers) are going to need to scale back tuning and therefore real performance uplifts might be scarce initially.

Intel wasn't getting the expected higher performance from 10nm, we should expect similar for 7nm / 5nm initially.

The good news is AMD has been doing tuning turn around's for quite a while already, well ahead of the curve, so maybe production performance uplifts will be significant.

Buy now and enjoy AMD Ryzen 2 CPU's / Vega /RX GPU's until the new 7nm process tuning meets your expectations, or wait and jump in on the first 7nm hoping for more.

If you are waiting, you are playing the waiting game. Buy now and game now.

 

GF doesn't matter in it at all. Should come right on time. They started sampling in Q2, ahead of the H2 sampling that was planned. That means that the first runs, they plugged the chips in and they ran! That is no small feat as usually you do not get something you are willing to sample on the first runs!

So, they already had planned refinements, feedback periods, and sampling throughout Q3 and Q4 to hone it. The only question is exactly when fab time is open at TSMC, Now, the fab time had to be inked a quarter or two ago, but considering radios for 5G production, etc., and any Nvidia cards needing produced (I'm not betting on large runs at all on RTX), having release in early 2019 and volume by computex may be reasonable.

I think GF dumping 7nm simplifies the equation. They already had it ready for TSMC. Now, they don't need to tweak it to try to run mainstream over on GF, so, they can just focus on refining the TSMC design from here on.

And I'm still solid with my 1950X, and may be getting a free 1700X (90% sure chip is alive and just the board died).

Meanwhile, Intel stalled, and even Intel said that 10nm+ was a side-grade from 14nm++ back in 2017. It is a mix of issues on manufacture, none of which have I heard coming from the Apple chip, or any other 7nm process, other than Samsung using 7nm EUV this year will only be on the lowest levels, not entire chip. Considering TSMC EUV volume is supposedly coming sometime during H1 2019, they should be fairly solid by this point.

But, there is still a lot in the air. And, there is a chance the chip tested, if true, isn't coming at the start, instead only getting a 48 core to start on the server side. Either way, I'm waiting for 7nm and learning linux, including how to get around some of the errors I am getting with Linux that they haven't resolved yet.

 

Learn more than Linux.

Linux is the tool between you and a working implementation, you need to have a data collected and a design - a plan to implement - Linux / language are the tools between you and your goal.

GF leaving the market pushes all 7nm business to TSMC and Samsung, AMD may be ahead now, but they will need to find a way to stay in front of everyone else queuing up for production, for years ahead.

When Intel finally gives up they will be looking for production space too...

 

Alleged AMD EPYC ‘Rome’ 7nm Based 64 Core Processor Performance Leaks Out – Scores an Incredible 12,500 Points in Cinebench Multi-Tasking Benchmark
By Hassan Mujtaba, 6 hours ago
https://wccftech.com/amd-epyc-rome-7nm-64-core-cpu-performance-benchmark-leak/

The images pulled are here...

 

Here is the thing, GF pulled out of the horse race because there was not enough people seeking 7nm to go around to recoup the extra $2B that would be needed to go live. The node is too costly, especially without EUV, but even designing on EUV will have its own challenges and costs. GF's 12nm FD-SOI chips are the equivalent of 10nm chips (Intel 14nm), which are really good, but also carry a voltage of 0.40V. By pivoting to RF, auto, and ASICs, they can still provide the performance, but recoup more costs and focus on the balance sheet. Considering 12/14nm and 12FDX have strong interest, I believe we will see a slowing on rush to most recent node. That leaves the primary designs and chipmakers to fight for fab time, which is Qualcomm, Apple, AMD, Samsung, and certain chipmakers like Amlogic, Rockchip, etc., which make up a smaller portion of the market. You also have the server ARM chips that will compete for fab time at 7nm as well, but looking at the current allotment, GF didn't provide much to those players, except for AMD. Also, if the rumor from Adored is true, then the uncore chip will be built on 12nm or 14nm, which makes sense. Between that and providing Ryzen Pro and legacy chips, we may see AMD fulfill the contractual obligations to GF. And, with some of the components taken off of the 7nm chips, they could, for Epyc, reduce the size even further than half, which would further increase yields with relative defect densities.

But, because of that, I see AMD staying on the cutting edge, and possibly splitting the amount of fab time dominance in years to come with Apple at TSMC. Also, AMD could feed the 7nm EUV line while Apple tries out the 5nm for 2020. So, I think there is room for the players at the top between TSMC and Samsung fabs. Especially with the diversity of nodes proposed at Samsung. But we shall see.

As to Linux, I'm using this more to understand that and programming, which is a jump off to other aspects of what I'd like to learn. Everything is a prelude to something greater.

 

It's really all speculation, except what the bankroll said, they want profits - they were / are tired of constantly investing to the future payoff.

There was no quote about not enough 7nm customers, they didn't care to make the investment needed to outfit a 7nm fab(s) while going further into debt. Instead they decided to take the money on the table, 12nm / 14nm production.

All the specifics were about 14nm / 12nm production being profitable and that's where they wanted to make a stand to get a pay out for a decade of investment, and recover the investment from the many mis-steps they've taken along the way - of which I think this is another mistake.

What happens when everyone is wanting to produce 7nm / 5nm instead of 12nm / 14nm?

Will GF simply sell off their investment?, take their 14nm / 12nm profits and close up shop?

 

Here is the quote on demand for 7nm:
“Bleeding-edge is becoming increasingly challenged. The number of players going into these advanced nodes has dropped significantly as result of the dramatic increasing costs to design in these leading-edge technologies,” GlobalFoundries’ Patton said. “The revenue story is a bit challenging. And then you look at the R&D cost piece. The R&D cost of these leading-edge nodes has been going up exponentially.”

And to answer your question on what happens when everyone wants to go to a smaller node:
“We are parking it,” Gary Patton, CTO of GlobalFoundries, said in an interview with Semiconductor Engineering. “I can’t say that we wouldn’t came back and look at it, especially if there are some opportunities for partnering. At the moment, we have no plans to pursue it."

Analysts agree. “While the leading edge gets most of the headlines, fewer customers can afford the transition to 7nm and finer geometries,” said Samuel Wang, an analyst at Gartner. “14nm and above technologies will continue to be the important demand driver for the foundry business for many years to come. There is significant room for innovation on these nodes to fuel the next wave of technology.”
https://semiengineering.com/gf-puts-7nm-on-hold/

 

I think that's not a true statement in the long run, instead it is a flimsy reason to ignore the 7nm / 5nm opportunity - all because they don't want to keep going into debt to continue to build out the fab for 7nm, or 5nm - which GF plainly stated as being the reason to stop the build out of 7nm / 5nm.

The demand for 7nm / 5nm will be the same as previous nodes, it is a matter of time, which is why they don't want to continue to stretch their investment further, and want to "cash out" now.

And, yes, they could "change their mind" later and start 7nm / 5nm production servicing the tail end of each process node; it's not a bad strategy as long as they can get production hardware inexpensively enough.

Either way is a gamble, and the lure of "cash now" is greater than the worry about failure to re-engage later.

Time will tell, but it's hard to imagine GF having the resources to re-capitalize their fab build out later - in enough time to catch the wave of profitability for 7nm / 5nm production starting late in the game.

 

This is why I believe them - even Intel said they will not have EUV until 2021 at the earliest. Intel is the fourth largest fab (not talking capitalization). GF is the third. So, stepping back and letting the top two largest fabs take it, especially with all the recent lost contracts, etc., makes a certain amount of sense. It is like going to another bar instead of trying to flex to pick up women standing next to two body builders. You will still pick up women, just from a different place.

Also, this is after the new CEO was put in place this spring, and likely after firmly losing 7nm production of AMDs main chips to TSMC. That would have amounted to the main fab time on 7nm for GF, more than any other contract they would have gotten for 7nm. So I think it was AMD choosing TSMC that led to the choice to pivot for GF, in part.

Some of the difficulties with planar chips is similar to the complexities of 3nm/5nm designs with nanosheets as well. Specifically, you are going to have to worry about uniformity with the horizontal nanosheet construction, which comes with benefits, but brings back issues from pre-finFET. That means going 12nm FD-SOI may not be abandoning the race at all, because some of what they learn could be applied to different tech and node. I wouldn't be surprised if Intel's Sapphire Rapids was an allusion to Silicon on Sapphire designs, a type of silicon on insulator.

Overall, I do agree some of the comments were BS fluff you feed to the market to keep them happy. Plus, GF said they are doing an ASIC sub that can license 7nm for production. That means this isn't a full write-off of 7nm, just that they will have so little demand, trying to do it for a main production facility without the volume AMD would have brought didn't make a lot of sense.

But that is how I see this move.

 

GF / AMD have (had?) an agreement for current node process production, and AMD / GF already had planned on production through GF for 7nm to lag TSMC.

If you want to give the impression that GF leaving the market, taking away production capacity for 7nm will *not* affect AMD in the long run, that it doesn't matter, that would be your point.

I think it does matter that 7nm GF capacity is no longer publicly planned to be available, and that AMD / TSMC are now AMD / TSMC + the 7nm demand not served by Samsung, which will grow over time.

It will take some time to sort out, but when it happens I hope GF is ready to jump back in with piles of cash.

 

I have the 1950x and am more than happy for now. At 7nm my next interest will be a 3950x or whatever top end that will have direct CCX die access to memory channels.

 

If the rumors from AdoredTV are true, we are looking at the cores all being separate from an uncore chip, so NUMA will no longer be a consideration, just the latency to get to the uncore chip. If an interposer is used, or Intel's AIB which was given to DARPA royalty free, or something similar, then there may be reduction overall and more equal latency to whichever core is scheduled. Also, there is info coming out that between the AGESA and PSP errors, among others (the new chipset driver for X399 on windows includes updates PSP, released early Sept.), that it greatly speeds up the chip, or using processlasso to control what is processing (literally, removing one core, two threads, has shown to nearly double performance of the 2990WX, meaning that part of the problem is recognition and use of the cores properly and something is causing the system to slow or halt, something I've recognizes as well on the recent AGESA 1.1.0.1 on the beta taichi bios). So, lots of misinformation about the cause, and the problem with larger core counts was originally discovered on a 2P intel server recently which also had 32 cores like the TR2 2990WX, and had a huge reduction in score. Just, in digging on the AMD side, they found more at play than the OS not liking high core counts, but also showing that the direct memory access is NOT the handicap everyone thinks it is. So, keep an open mind to what is found between now and next year, it may surprise you!

 

I always keep an open mind where my next options may lie. If there is an active interposer of 10% it may totally change everything. Right now I am basing my next upgrade path based on current situation assumptions. No matter what I am also assuming 7nm will hold some type of viable upgrade path. The current 2990wx is just too much give to get for me though.

 

News, pretty quiet out there;
https://www.kitguru.net/components/...ccess-is-paving-the-way-for-7nm-server-chips/

 

As Intel IMPLODES, AMD Begins to Rise!
The Good Old Gamer
Published on Sep 24, 2018
With Intel's inability to supply OEMs with 14nm CPUs Market Analysts Believe AMD will be MORE than Able to Help with Supply. In turn, recapturing MUCH of the Market share, and Sizable Bump to their Stock Price.

 

@Robbo99999 - here is what I posted about 7nm in OCN forums:

So, what is it about 7nm that you want to know. The best time to buy an Nvidia GPU was used 1080 Tis leading up to the release of the 2080 Ti, when you could find them from $400-500 regularly on ebay. Picking up two of those for the price of a 2080 or a brand new 1080 Ti, after buying a hybrid block for it, was a no brainer. On AMD side for cards, prices keep coming down, but Vega costs too much atm, IMO.

But let's get back to the 7nm node, generally. AMD will be going with the TSMC 7nm node and 7nm+ node, as GF has bowed out from the 7nm horse race. Qualcomm seems to have contracted with Samsung, Apple with TSMC, and Nvidia is rumored to have designed Turing on 10nm Samsung but went with 12nm TSMC due to delays, but also means Nvidia could do the optical shrink from 10nm to 8nm at Samsung and release another card next year.

TSMC 7nm is comparable to 10nm Intel process. Just because they have the smaller number means nothing. Densities are roughly the same, depending on how they are calculated. It should be noted Intel did not achieve the maximum state of theoretical limit for them, which was 106 MTx/mm2, rather they reported 100.8 MTx/mm2, IIRC. That was from 2017 March press release. Since then, they had to gut a feature of 10nm responsible for a large portion of the die shrink, but rumors have since come back suggesting mitigations are in place, so it won't be so drastic a hit. When looking at Intel's 10nm dual core mobile chip available to Lenovo, it was really unimpressive, had lower boost than the chip it replaced, etc.

Meanwhile, the theoretical max on the TSMC 7nm was 96.49 MTx/mm2. https://www.semiwiki.com/forum/conte...processes.html. These numbers are relative and based on SRAM cells. On the Apple chip, TSMC achieved 83.3 MTx/mm2. That isn't theory, that is production silicon in the market. Because Intel keeps certain info close, we cannot calculate the 10nm cannon lake mobile chip with that much certainty. So, AMD will be using this process for 7nm and is meeting their timelines. Intel, on the other hand, bowed out of EUV until 7nm Intel process, estimated for 2021, comparable to industry 3nm or 5nm processes. That means TSMC going volume production with EUV in 2019 and Samsung currently using EUV on the bottom most layers of their production already exceeds what Intel is doing, that TSMC currently is in the lead on process node development, and that both TSMC and Samsung will have used EUV for at minimum of two years before Intel's adoption of the new lithography. Also, after 7nm, the industry is losing the ability to go finFET, instead looking at Gate All Around solutions, some of which use nanowire or nanosheets. But that is a peak into the future.

So, now that density has been discussed, let's talk about the rumors surrounding AMD's 7nm process specifically. These include 5GHz CPUs, 10-15% IPC jump, changes to the interconnect (2 rumors here, one involving interposers, the other as refinement of IF for the interconnect), an uncore chip containing the IMC and I/O, etc. First, recently, a fairly trustworthy source posted in a forum about the early silicon being 4GHz base, 4.5GHz boost, using 3600MT ram, and on a nondescript internal AMD branded board. If we look to the first Zen release, we saw those chips arrive in March and April, and the refreshed mainstream chips arriving in April of 2018. Following this pattern, we will likely see an April release next year, which puts the release 6-7 months out. The rumor was posted on Sept. 13, meaning that two weeks have passed, so considering it may be the second week in April for the release, following the past two releases as examples, that means from the time of the rumor, they had 7 solid months to get the chips ready. That means the part about it crashing a lot means nothing. In December of 2016 and January of 2017, the first Zen processors early silicon clocked to 3.1 with a boost to 3.4GHz. We all know the final silicon clocked higher. As such, we could see increases making the speed comparable to the new Intel chips being released this month (if it isn't a paper release due to supply issues again). In that same thread, the reason RTG got the CPU sample was to work on graphics driver support related to the interconnect changes on the chip.

So let's talk interconnects. It seems like the current interconnect levels off around 1.55GHz, according to what is read on Sisoft Sandra. It is tied to memory speed, which is why 3200MHz ram seems to be a sweet spot considering price. There could just be tweaked enhancements that go with it to help speed it up, including allowing for faster transfer on IF. Then, there is AdoredTV's take. He dug through papers researched by AMD in 2014 and 2015 about using an active interposer of 1-10% active components which showed a drastic reduction in latency. AMD followed this with a cost analysis for implementation in 2017. This leans toward AMD exploring putting the chips on an interposer with a custom topology at some point when the cost benefits shift in favor of adoption. If this is adopted, then the rumor on an uncore makes sense, as having the IMC and I/O on a separate chip has many benefits, like not having to shrink the hard to shrink CPU components, being able to bin IMCs so that higher ram speeds can be achieved, lower production costs by producing uncore chips on the more mature node, thereby helping with yields, etc. Also, instead of using an interposer or IF, they could also use a tech like AIB or EMIB, to note known Intel techs in this. Although those solutions are not as elegant as IF or an interposer, it gets the job done. I find that the least likely possibility, but it is out there considering Intel gave AIB patents to DARPA to try to force that as an interconnect standard in order to undercut AMD using the custom topology interposer in the future. Instead, AMD has more of a protocol based approach to control data movements which is design agnostic, thereby allowing it to be used across multiple design implementations. I know which I would prefer (it isn't Intel's solution which stifles innovation).

But, either way, Zen 2 was a larger redesign. If the I/O and IMC are moved off die, then the dies would be smaller, which would increase yields for AMD considering the normal defect densities of new nodes. This is why that rumor could have legs. It also would have other changes which could increase IPC a fair amount after learning the issues from Zen and Zen+. Even changes to the interconnect can reduce latency or increase speed potentially fixing bottlenecks on data delivery to caches and cores, which could increase IPC. That means the 10-15% improvement is quite possible. With that said, Zen+ was around 4% slower IPC than Intel's current offerings (including coffee lake), and Zen was around 7%. So, depending on starting point, that could be as little as 3% IPC over Intel's current offerings up to 11% IPC over Intel's current offerings (the 9900K and 9000 series generally are Coffee Lake refreshes on 14nm++ node. There is no reason to expect much change in IPC, rather the large boost will come from better heat dissipation through using solder between the die and IHS which allows for the increased speeds (note- AMD uses Intel's patent to solder all of their chips except for the APUs, IIRC). Also, AMD still enjoys their SMT being better implemented than Intel's HT, and does not have the same security issues that Intel chips do with HT and memory vulnerabilities.

Putting this all together, AMD will be using an equally dense, but arguably better, node from TSMC as compared to Intel's 10nm (although more dense absolutely than Intel's 14nm), is looking at an IPC increase that should exceed Intel's offerings, may reach speeds in the mid to high 4GHz range on all cores (early silicon so hard to tell, but compelling reasons to think this, stopping short of saying the 5GHz number) all at a lower price point and without supply concerns. That makes AMD 7nm really compelling from both a performance standpoint and a price to performance standpoint. Also, if the uncore rumor is true, it could also mean that an uncore chip has to be used with mainstream chips, which means that the CCX and Zepplin complex can remain otherwise unchanged with 4 cores per CCX and 8 cores per die (2 CCX complexes). Instead, the rumor started by or around the time of MSI's ad piece saying supports over 8 cores could have referenced a mainstream configuration of two 7nm dies with an uncore chip for mainstream CPUs. That would comply with the rumors without having to do the odd BS of making it 6 cores per CCX or dropping in another CCX to make three CCX per die. At worst, AMD would design 2-3 uncore chip designs to be produced on 12nm/14nm GF while the cores die is on 7nm TSMC, which may have costed less than fully shrinking the I/O and IMC, allows potentially hiding NUMA altogether since the IMCs are all on the uncore chip together, etc. I find that to be the most compelling explanation considering the rumors currently in the wild.

Now, with all this comes the rumors surrounding the motherboards for the server and HEDT market. Let's start with the Server platform. First is that PCIe 4.0 will be supported by Epyc 2. That is pretty likely and Epyc 2 support for PCIe 4.0 is confirmed. With that, IF will also be used to connect the PCIe slots for a bridgeless multi-card solution (crossfire with IF connecting PCIe). Now for the rumor side. Supposedly, with the uncore chip, the support for DIMMs will increase drastically on Epyc 2, with AdoredTV mentioning him hearing a rumor for 32 dimms supported on a 1P unit. That is a bold rumor, but if true, is phenomenal.

The rumors surrounding HEDT is that X499 boards are in development and may be ready for CES. Considering MSI is the only board with the VRM built for the 32 core behemoth, it isn't hard to imagine the other manufacturers are busy at work making their own boards, just that they are waiting for the new chipset and 7nm to release.

Now, Asrock Taichi uses the same exact VRM as the Zenith Extreme from Asus, just with a worse VRM cooling solution (literally, same design down to the chips used). So, if you cannot wait for the new X499 boards, I would recommend the MSI MEG Creation X399 board, followed by either the ZE or Taichi (Fatality is the same, just with the 10Gbps port integrated) but with active VRM cooling. I currently use Koolance universal VRM blocks and have temps under load in the 40C range. But, no one knows what, if any, new features may be supported with X499, so....

I hope this helps.

 

Herein lies the issue, without silicon to test out we can tech quote all we want but little else. There is no way to truly know what we will end up with. The track record of AMD leaves me to still doubt 5+ GHz on 7nm but we will see.

 

You need to get outside of the frequency chase. Coffee is about the same IPC as all skylake chips. That is around 4% higher than a 2700X for Intel chips, and the 9700K and 9900K should be the same. One rumor I omitted came frame redtechgaming which put IPC at 15-20%, rather than 10-15%. If the IPC is large enough over Intel, it doesn't matter if Zen2 winds up with a little lower frequency, such as high 4ghz range, as it processes more per cycle. And considering how new Zen is compared to iCore, there are plenty of ways to optimize for more IPC, whereas after a decade, how much is left in the IPC tank for Intel?

Also, if being honest, why do you primarily say such things toward only AMD considering the news about Intel over the past couple months?

Sent from my SM-G900P using Tapatalk

 

I am only in the frequency class as their people have mentioned it. Also frequency sells IPC only when thee is a severe difference in it.

 

Frequency is not king. Let me put this in perspective. If AMD can achieve a 4.7GHz all core clock, like Intel has stock, but has a 10% IPC advantage over Intel, then Intel would need 10% higher clocks, such as 5.2GHz, to equal the same performance. Considering the early silicon on Ryzen showed 3.4 boost, but the all core on final silicon was either 3.7 or 3.8GHz, depending on product line, seeing 4.5 boost on early silicon suggests 4.7GHz all core isn't unreasonable. Then there is the IPC estimates which are fairly reasonable. Remember, Zen is only on its third version with Zen 2, if you consider Zen+ its own version. Meanwhile, Intel hasn't seen much IPC change since skylake dropped, and IPC gains overall between generations has gone down as iCore gets older, meaning there are fewer tweaks that can happen to squeeze out more IPC. As such, I expect IPC to be largely unchanged since Coffee and these upcoming chips being coffee refresh. Taking that as adding 2 extra cores, I expect about 200MHz lower than what was seen with the 8700K or 8086K. That puts it at around 5.2GHz as the upper amount I expect to see from the 9700K and 9900K. So, if AMD gets 4.7GHz all core AND gets 10% IPC over Intel's current offerings, then AMD stock would perform the same as Intel OC, all other things being equal.

There is more analysis that needs done, like HT vs SMT, any changes to mem latency, etc., or whether the mem latency corrections were compounded into the expected IPC increases. But you get my point.

 

I know it is not king, it is though what sells. Just like cores are not king but what sells.

 

A take on it all from a less serious perspective.

PCWorld: Core i9-9900K vs. Ryzen 2800X: Totally fake benchmarks!.
https://www.pcworld.com/article/330...k-vs-ryzen-2800x-totally-fake-benchmarks.html

 

AMD is announcing a keynote speech on 7nm for CES 2019 to arrest the recent stock price slide of AMD in the market. Intel put out a fluff announcement on keeping up with demand along with some annalists saying without hard information that Intel will get 10nm together, will R&D new chipset and generally get its act together and at the same time degrading AMD to less than a buy but upping the target prices. This smells of a major PR hit job on AMD.

IMHO the CES 2019 announcement is not enough of a response.

 

https://videocardz.com/78523/intel-to-announce-core-i9-9900xe-basin-falls-refresh-cpu

https://twitter.com/IanCutress/status/1049293039070580736 -- Now that is a motherboard.

https://www.cowcotland.com/news/643...nus-extreme-se-montre-attention-les-yeux.html --Moaaarr Power

 

https://newsroom.intel.com/news/intels-fall-desktop-launch-event-livestream/

Intel Launch Event Live Stream starting in 9 minutes.

That 28 Core chip is 4.3Ghz out of the box. Looks like no water chiller this time just a standard AIO. Nice. Also shipping in December.

Still hilarious.



New X series chips are soldered as well.

 

My first problem with that MB is simple: WTF are they covering up the VRM cooling with a piece of plastic? That would be the first thing thrown in the box hopefully never to be seen again. Next is asking why the fan on the slot shroud? Most likely, from position, cooling for an m.2 drive, which I have no issue with. The second complaint is, from what is known, Intel will still have only 48 PCIe lanes, so people need aware that the PCIe x16 slots are likely not fully wired on all slots. Now, that doesn't mean I don't like the inclusion of the full slot to support PCIe x16 cards, just running at x8, as that is fine and allows more configurations, generally, on the board.

Finally, it is Asus, so I fully expect that even with the impressive VRM design (likely similar or same to what was seen at computex for the platform), buggy firmware may kneecap an otherwise solid motherboard.

Edit: And to be fair, the X399 boards suffer from some of these issues, like not all of the PCIe slots being fully wired for PCIe x16. AMD chips have the higher count, but it seems handcuffed because they ported designs, in part, from X299 boards because they thought the X399 boards wouldn't get market penetration, which turned out to be false. I could complain about each X399 boards for different reasons. So, want to point that out.

 

Looks aluminum to me and is covering what looks like 4 fans to cool the VRM. This motherboard looks beast mode and over engineered.

 

Looked like 3.7GHz all core load.

Edit: Also, that does NOT mean I think that is bad. For 28 cores reaching 3.7, similar to AMD's 32 core offerings, we could see similar overclocks to AMD, so 4.2-4.4 seems about max without extreme cooling (estimate BEFORE seeing reviewers get their hands on it). But that is what needs mentioned, all core, not just peak single core boost.

 

That looks aluminum? Also, if this side is closed, depending on how the other side looks, how is the air moving? you suck in to blow down on something completely closed. The VRM, if like what was at CES, is a beast. Not arguing with that. I'm saying the fans and design to try to style it is CRAP! Please don't mistake that for me not liking the engineering on the VRM itself.

Also, you cannot say that on other components until it effects your use cases or you have all information. A lot of boards look good on the surface. Also, read the edit I made on that above post. It will put my comment in context of things I was pissed about with the X399 boards. So this isn't to just bash on Intel, but critique the state of HEDT boards generally.

Edit: I also forgot to applaud the hexa-channel ram on HEDT, which is a wonderful advance on the platform and hopefully pushes AMD to offer 6 or 8 channels with Zen 2 and X499 boards. (just hopeful wishing on AMD side, but the hexa-channel and memory bandwidth would be awesome!)

 

Price lists..

https://benchlife.info/intel-9th-co...announce-with-28-cores-xeon-w-3175x-10082018/

 

So, no change per price on core count for HEDT. But, needs reminded, this is 14nm++, so has the extra speed seen from Kaby to coffee, which allowed better clocks.

Meanwhile, on mainstream, at $488, plus the premium from the shortage on chips being likely until next year, then the 9900K isn't that great, IMO.

But that 8c/8T, I'd recommend that all day for gaming and over the 8700K. If you need SMT/HT, then I'd likely point toward a 2700X, so long as you do not need the high single threaded performance. And if I see anyone reference the 8700K hardware acceleration, Puget Sound showed it was snake oil. But, there are other reasons you would have mixed workloads and the Intel 9900K would work best for the mixed workloads of that user. To them, I say get it. But generally, the 9900K at $488 MSRP doesn't seem worth the value. That is 50% more cost while having 16% performance according to leaks. We will know more on performance soon to see if the 16% holds.

Also, with MB costs also increasing on mainstream, although in part justified due to additional engineering being claimed for them, it makes looking at the HEDT platforms a bit more applicable compared to the mainstream 9900K.