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

So, you are arguing that IPC is higher, but speeds are going way down, meaning that it will have lower performance because slower, rather than IPC lower, but keeping the speed. Either way, INTEL STATED ITSELF THAT PERFORMANCE (edit: of transistors) OF 10NM IS LOWER THAN 14NM++!

THIS SLIDE IS ONLY INTEL, PRESENTED AT ITS TECHNOLOGY AND MANUFACTURING DAY MONTHS AGO! INTEL ITSELF SAYS THAT 14NM++ IS BETTER ON PERFORMANCE, 10NM IS BETTER ON USING LESS POWER!

 

https://hothardware.com/news/next-gen-ddr5-to-offer-twice-the-bandwidth-and-density-of-ddr4-memory
Just a reminder that the DDR5 Demo is tomorrow at the Santa Clara Server Forum! We should know more about it then. Wondering about the system demoing the hardware...

 

i wont try to argue with you man, intel can say whatever they wanted to say, in order to benefit their business and further their own agenda. kabylake vs skylake difference was 0-1% within margin of error, which already puts the graph you quoted and debunked your statement. again you can believe what you want to believe.. i'll believe in that coffeelake will be lower ipc than cannonlake but will be able to clock higher due to better refined 14nm process vs first gen 10nm.

 

i for one can't stand to have a 6 cores laptop while desktop people could get 14-18 cores, unlocked. you know what i mean? difference is just too big. i could live with 6c laptop and 8c desktop or 8c laptop and 12c desktop, but not 6 vs 12 or 14c lol.

 

I'm not saying you are wrong here, I'm saying that Intel still says 14nm++ has better transistor performance than 10nm. Intel regularly says 15% improvement (bw to skylake, skylake to Kaby), but delivers less. Now, they are claiming coffee will have 30%, while saying the transistor performance of 10nm, which is cannonlake, will be lower than coffee. So look at which they are moving to 14nm++- mainstream and server chips. Look what isn't -mobile (needs low power) and HEDT. This means GloFo getting 40% transistor performance at 5GHz suggests if AMD uses that, TR replacement gets 25% from speed boost and a sizeable jump in IPC over their 14nm design. We'll see if that pans out.

So, we all know Intel inflates things, but admitting a process won't be beat until 10nm+ ice lake for transistor performance is very telling, especially if they don't go crazy on adding cores. That is my point, although I can admit it may not be on IPC.

Sent from my SM-G900P using Tapatalk

 

A storage company doesn't care about fastest speeds - it cares much more on total capacity.

 

From your link below:

Notice the part where it says they can change anything and everything at anytime without notice?

So... go ahead and buy AMD 14nm today just because AMD's partner stated they might (or might not) deliver 7nm, uhmm... sometime in the future.

With this kind of logic; it brings all your statements into question, huh?

 

It is only you and a few online rags that are pushing for page hits that think Intel has no clear path.

The path they are following is very clear; give definite performance improvements to their customers...

Because, if they don't? They won't have them (the customers) any more.

Other than price on a single component (CPU), AMD isn't offering anything close to what Intel has (and has had for a long time).

You can analyze and twist the info they're sharing with the world; but in the end, the only thing that matters is where is the product(s) for my productivity.

Oh; there they are - over in Intel's corner.

 

Edit: http://www.tomshardware.com/reviews/intel-core-i9-7900x-skylake-x,5092-11.html
So, tom's could not do above 4.6 on the 10-core, so went with a water chiller.

Edit 2: http://www.guru3d.com/articles_pages/intel_core_i9_7900x_processor_review,22.html

"Issues - Hardware P states
I won't lie, the past two weeks with the Core i7 7900X and the X299 boards I have tested have been an absolute challenge. Memory XMP profiles would not stick, power consumption with one BIOS was OK, the other through the roof. But most of all the processor performance was all over the place. We've seen perf differences of up-to 20% in-between merely different motherboards. MSI however with it's latest BIOS seems to have found equilibrium. Now the biggest fight the two weeks was ironically game performance, it was severely lacking. Example a platform like this should run Rise of the Tomb Raider at 140 FPS at 1080p on a GeForce GTX 1080. We'd end up at 90~100 FPS. And that problem occurred with pretty much all games. I have been discussing this with the motherboard partners (as yes it is widespread) and we all agreed, it has everything to do with 'hardware P states' that Intel recommends to leave enabled for the new platform. Intel recommends certain power states to keep the TDP in line, as otherwise they simply cannot achieve that 140W TDP. For most overall tests that worked out okay enough, but specifically the toll on game performance was abysmal. Days before this launch MSI however released and provided a new BIOS, this restored the performance to what it needs to be. But as you have been able to see, the power consumption is certainly on that high-side. In the end though, the performance is there, but we do expect several BIOS updates that will have an effect in performance overall, in gaming and on power-consumption. "

"Two things I find to be significantly bothersome, this launch is clouded by too many architectures and processors that really haven't been released. You can choose from 4 to 10 cores, which is fine. However Intel went big with announcement on 12, 14, 16 and even a 2000 USD 18-core part. The 12-core version should see the light next month, however the last three are nowhere to be found and I am starting to doubt that you'll see them anytime soon. The latest indication is October. Sure perhaps a review here and there to show off that Intel can do it. But retail availability? I don't know man. So this all is cloudy and confusing really."

 

https://www.pcper.com/reviews/Processors/Intel-Core-i9-7900X-10-core-Skylake-X-Processor-Review

Those that complained about the CCX latency!

 

I'll be honest, I thought Intel would have issues keeping TDP but not this early in the game. I had expected to see issues with the 12 core variant. I guess I started to believe their hype and all was going to be ok. I knew OC'ng would get it out and above TDP but not just stock clocks.

 

Here is the review roundup (just collection of all review links) from WCCFTech:
http://wccftech.com/intel-core-x-skylake-x-kaby-lake-x-cpu-review-roundup-x299-platform/

Edit:

Anandtech Intel Core i9-7900X
Intel Core i7-7820X
Intel Core i7-7800X Skylake-X ASRock X299 Taichi
MSI X299 Gaming Pro Carbon
GIGABYTE X299 Gaming 9
Hothardware Intel Core i9-7900X
Intel Core i7-7740X Skylake-X
Kaby Lake-X ASUS Prime-X299 Deluxe
Coolaler Intel Core i9-7900X
Intel Core i7-7740X Skylake-X
Kaby Lake-X AORUS X299 Gaming 7
Techbang Intel Core i9-7900X
Intel Core i7-7820X Skylake-X ASUS Prime-X299 Deluxe
Overclock3D Intel Core i7-7820X Skylake-X ASUS X299-A Prime
PCOnline Intel Core i7-7820X Skylake-X ASUS Prime-X299 Deluxe
PCPer Intel Core i9-7900X Skylake-X ASUS Prime-X299 Deluxe
Tomshardware Intel Core i9-7900X Skylake-X MSI X299 Gaming Pro Carbon AC
Tech Report Intel Core i9-7900X ASUS Prime-X299 Deluxe
TweakTown Intel Core i9-7900X Skylake-X AORUS X299 Gaming 9
Guru3D Intel Core i9-7900X Skylake-X MSI X299 GAMING Pro Carbon
PCWorld Intel Core i9-7900X Skylake-X ASUS Prime-X299 Deluxe
Hardware.info Intel Core i9-7900X Skylake-X ASUS Prime-X299 Deluxe
4Gamer Intel Core i9-7900X Skylake-X ASUS ROG STRIX X299-E GAMING
Mobile01 Intel Core i9-7900X Skylake-X ASUS Prime-X299 Deluxe
HardwareCanucks Intel Core i7-7740X Kaby Lake-X ASUS ROG STRIX X299-E GAMING
Vortez Intel Core i7-7740X Kaby Lake-X AORUS X299 Gaming 3
Expreview Intel Core i7-7740X Kaby Lake-X AORUS X299 Gaming 7
Bit-Tech Intel Core i9-7900X Skylake-X Not Mentioned
Hexus Intel Core i9-7900X Skylake-X Not Mentioned

 

WOW, I'm very surprised by the 7900X's TDP under full load. I'm sure the mobo vendors will release new BIOS to fix performance and voltage but damn.

 

You have to understand, the bios they released for this weekend, which caused reviewers headaches, was to remove the tdp limit on some of these boards to allow them to get the performance. Before that, the performance was lower, especially on gaming. So, it looks like you go without the tdp limit to get full use, then worry about the heat generated, followed up with toothpaste instead of solder, etc. It will need refinement, for sure...

Sent from my SM-G900P using Tapatalk

 

Oh, I understand perfectly. Some of the reviews I've read, show power consumption hitting 147 but that is still higher than even my 6950X at load.

If I were to buy the 7900X, I wouldn't be OC'ing it. I'm ready to see what TR has to offer now.

Side Note: People called me crazy for buying 1200W+ PSU's.

 

I've got an evga 1600 titanium, so I hear ya. Now I need to find out more about TR and epyc, as well as Vega and Volta!

Sent from my SM-G900P using Tapatalk

 

The more I delve into the 7900X, the less I like about it. This excerpt from Tomshardware sums up what I'm feeling.

" Enthusiasts might hope for similar improvements from Intel. After all, AMD is overcoming its roadblocks with a fraction of the R&D budget. We asked Intel if it expects software-based optimizations to fix what disappointed us, and company representatives responded that software tuning for the new architectural enhancements and cache hierarchy could improve performance. Remember, though, Core i9-7900X is based on the same micro-architecture as older Core CPUs. It's improbable that mere code updates will rectify issues introduced by Skylake-X's layout when Skylake-S and its derivatives are already well-supported."

I truly didn't think the 7900X was going to smash my 6950X in single or multi-core IPC, but I expected more overall. The safest OC I'm seeing is 4.6, otherwise you will require a delid and a custom cooler. Now I question anyone that will spend a $1000 or more on a CPU and then delid it.

On the other hand, @hmscott you're right. The 7820X represents the best value if you are not going multi-GPU and I think that is going to be Intel's best seller this gen.

 

This is why I said the 12-core presents a better value. If you are hitting 4.4-4.5 on that, you will be doing alright on performance (although it will suck down power).

Personally, I'm interested if the TR can hit 3.9-4.0 on all cores. You have more core count and, depending on heat and TDP, more performance on some multithreaded tasks. The latency fear about AMD has now been shown in Intel's chips, so that is now a wash. But, TR/EPYC/Ryzen are new, and as this article points out, Skylake has been around for awhile. So wait and see seems like a safe bet. But, tomorrow EPYC goes up for sale, so hopefully we can get some info on that.

 

Again; you missed the very next graph.



AMD is still 40% slower when jumping between CCX's. Intel doesn't have that issue.

Whether or not that higher latency is a problem for Intel will depend on how well their cache algorithms are designed - a few cache misses and it won't be to much of an issue on a fully loaded HCC workload - more than a few; yeah it will be... (just, what is 'a few misses' though?).

 

I get what you are saying here, that's a lot of money to risk delidding, and if I hadn't already it myself - delidding $1k CPU's - I'd agree, it's too risky for a 1st timer; practice on dead CPU's if possible or on something cheap enough that you can take the hit.

In this case I would question anyone that would buy any of these "toothpaste" IHS CPU's without expecting to delid

You lose too much headroom by keeping the "toothpaste" IHS material in operation, you must replace that before the CPU is worth the cost.

Price / value wise particularly for the core count it makes the most sense, but only if you can live with the limited PCIe lanes offered.

I wonder how many x299 board features are unusable unless you get the 7900x, and at each step down in PCIe lanes for the lessor CPU's?

 

I will address two things from you, since I got an email telling me you responded.
1) you point to regular standard language on the GloFo materials as some sort of proof that what is said means nothing. That language is standard and is included on Intel material as well, so your pointing to it MEANS NOTHING, except you cannot understand standard disclaimers. It's ok, you can hire me to interpret that for you, or to look over your contracts if you like.

2) if you look at all comparative data on bench scores from pcper, it was conducted BEFORE the platform improved, including BEFORE AGESA 1.0.0.6, meaning the infinity fabric was lower than the speeds currently seen by the majority of users. Considering infinity fabric is geared to ram speeds and infinity fabric is what connects the CCX to the other CCX, which is where the higher latency is found, that is not representative of current intercore latency. But, moreover, it closes the gap in latency significantly.

3) Intel's applies to any movement from core to core, AMD's can be designed around to look for lower latency paths, then group it to promote lower latency paths being used between cores. Not all programs can park in L2, even with the larger cache, and are not designed to use the cache. Sometimes that would be a larger redesign of software structure. Meanwhile the actual architecture will not see improvements on optimization for software usage. So, it isn't as simple as you suggest.

4) people laughed and derided at AMD using L3 as a victim cache. Now Intel designs the same, following AMD's lead. Now not a peep and saying it just needs optimized for. Duplicity!

Edit: "According to a new revelation from AMD, the speed Infinity Fabric runs at is linked to memory speeds. That is the IMC speed controls the speed of the 256-bit wide bi-directional crossbar. Faster DDR4 memory will lead to faster Infinity Fabric and hence faster communication between the 2 CCX making up Ryzen. This means to unlock the full performance of Ryzen 7 when stressing more than 4 cores, one must obtain the fastest DDR4 possible and OC the IMC like crazy. Perhaps AMD should have considered sending out faster DDR4 sticks with their review kits. This may also mean Ryzen can perform better than we’ve been led to believe with many reviews and perhaps even in gaming."
https://www.eteknix.com/amd-ryzen-ccx-interconnect-infinity-fabric-tied-to-memory-speeds/

PCPer did not directly describe the rig, instead it ran the CPU in testing at 3.5GHz (all were at that speed), but they mention it supported 2400MHz ram, which means they likely tested latency when ram was at 2400, meaning the Infinity Fabric interconnect was running at 1200MHz. Considering just about everyone is running 3200MHz now (1600MHz on infinity fabric), the speed of the interconnect is running at 33% for the average user OVER that of the latency test. Latency is connected to clock rates, but they measured it in ns. As such, since we do not know how the latency is geared to the speed, as the speed goes up, the latency should reduce in raw time, but not the amount connected per clock. So, considering EPYC and TR have an improved I/O interconnect, and depending on what speeds people are able to achieve on ram speed, as well as tying the channels to increase the speed of the Infinity Fabric interconnect, we may see different results, although I admit that proof is in the pudding.
https://www.pcper.com/reviews/Processors/AMD-Ryzen-7-1800X-Review-Now-and-Zen
https://www.pcper.com/reviews/Processors/AMD-Ryzen-and-Windows-10-Scheduler-No-Silver-Bullet


Sent from my SM-G900P using Tapatalk

 

intel's cpu is known of great quality and they last. i have had maybe 15-20 cpus in the past couple of years, aside from the bad ones that i buy from ebay people arent telling me, all of the cpu lasted years. i'd go ahead and delid this little sucker and overclock it for best temperature and highest frequency.

on the other note, im very disappointed in all of these reviews not testing about turbo boost max 3.0 while overclocked. my understand is that turbo boost 2.0 becomes pointless as people clock higher than the boosted speed, which in terms would seem that making turbo boost 3.0 pointless too however there were 2 reviews from broadwell E i remembered clearly was that they clocked all cores to say 4.2ghz, and 1 core at 4.6ghz.

thats something i wish to see, 10 cores at 4.5ghz frequency, and 1-2 cores at say 5ghz, yet nobody tested it.

 

To be fair to the reviewers, they had to deal with getting a last minute BIOS/UEFI update to remove the TDP limit, which increased performance a LOT in many benchmarks. So they were so busy rerunning EVERY benchmark, they may not have had time. Another possibility is that it wasn't ready software side. If I remember correctly, the boost 3.0 used to be software controlled in windows (but I could be wrong here). That is likely why it wasn't tested.

 

The latency is not as bad as TR but it is worse than it was. This is the reason some are blaming on the lackluster gaming performance of the 7900x compared to prior generations per clock. Being as then overall clock though is still higher than Ryzen and again not as bad latency it still is a more potent gaming platform.

I think if they would just switch back to solder quite a few more people would jump back on their bandwagon. As it is right now it seems the biggest universal complaint is having to delid a $1000 CPU. This was fine for them to do to consumers when essentially where else could you go for a CPU, but times are a changing.

 

Boost 3.0 for the new processors requires a separate driver but Intel is working on getting it natively supported in Win10.

 

I've delidded many CPU's, but they were mainstream and all of those cost below $500. If one were to die, I could afford a new one. Spending over 1K on a CPU, it comes with the expectation that I shouldn't need to delid it to extract the maximum performance. I think this was Intel's intent with the new processors, more delids equals less warranty woes.

For this new generation of processors, I'm officially skipping the X299 platform unless Intel starts producing soldered HEDT processors.

 

That is certainly a reasonable position, and I agree.

 

New information on that:
https://www.pcper.com/reviews/Proce...Core-i5/CCX-Latency-Testing-Pinging-between-t

It should be noted that the Alienware Area 51 is built with 2933 ram. So, this means the gap is practically neutral moving between cores, with Intel having a slight advantage. Home built rigs may see higher ram speeds than even 3200MHz AND TR and EPYC are supposed to have an improved interconnect over Ryzen!

 

I've responded many times to you; yet you've ignored me before.

Look, I know it is standard lang and all manufacturers state it. Yet, that doesn't detract from the point I made. You're betting on future 'ifs/whens' for something that you'll spend money on today. Seems silly, eh?

I'm a little above your calibre, but thank you. I will interpret any and all written materials myself.

You are so wrong with Intel 'following' AMD. Sigh...

See:
http://www.anandtech.com/show/11550...-core-i9-7900x-i7-7820x-and-i7-7800x-tested/5

Yeah; March 2014 over 3 years ago... with initial silicon models from 2013 or so...

The L1/L2/L3 Intel cache is also not AMD comparable:

See:
http://www.anandtech.com/show/11550...-core-i9-7900x-i7-7820x-and-i7-7800x-tested/4

We're all in agreement (finally) that until actual testing and comparisons are done on all announce hardware/platforms with working BIOS's, etc. - we're just wasting time here...

But running Ryzen on 3200 MHz RAM is overclocking (by a lot). The AnandTech review shows Intel beating up the theoretically superior Ryzens easily (even with your overclocking being taken into account vs. stock Intel).

The whole is greater than the sum of it's parts. (i.e. the platform is more than the CPU).

In the end; we've moved vastly beyond 'gaming' performance here. These are grown up tools and should be used accordingly in the appropriate workflows they're meant for.

Right now; gaming is still the i7-7700K's domain.

 

Sorry, but your logic is flawed. The price of a CPU has nothing to do with what you should need to do (or not) to extract maximum performance from it.

If you could afford to kill a $500 CPU previously; you can now only do so every second time.

For the record: yeah; I've played with O/C'ing in the past. Not impressed (with speeds gained nor heat/noise/power and reliability of the platform).

I now drive stock - but that is why I buy at the higher end of the spectrum.

 

Boost 3 is processor controlled - drivers will be available via WU - but processor vs. O/S controlled makes for a very fast switching platform (from min/low power to max/high power).

See:
http://www.anandtech.com/show/11550...-core-i9-7900x-i7-7820x-and-i7-7800x-tested/7

 

You are comfortable buying at the higher end of the spectrum where Intel used to give their best. This is not Intel's best. You run an unlocked processor stock but many HEDT consumers do not. The point is, Intel released a family of high end CPU's that are hindered by simple paste and there is no excuse for it at this price point. I should not have to void my warranty to curtail a cost savings metric by Intel to achieve maximum performance.

OC'ing IS a major part of the HEDT market, thus the reason Intel markets these chips as having the capability and capacity to do so. I am personally not impressed by Intel's lowered product standards.

 

I may be wrong and you may be right; but O/C'ing died out for my immediate circle of acquaintances at least half a decade ago.

Intel is not marketing these chips with O/C'ing in mind though? They actually state the opposite.

No one is forcing you to void your warranty. Just the need to repeat what you've done in the past is.

Does O/C'ing any current platform today give slightly better performance? Yeah, obviously.

Does it give more productivity with no loss of reliability, dependability and availability of the platform in question? Far from it, ime.

O/C'ing is going the way of RAID0 arrays. And I stopped playing with that before I joined nbr (thank you vRaptors ).

 

Intel launch info, opening to 06:27

Spoiler: Full show index

Published on Jun 19, 2017
Article URL's mentioned in the show:
https://www.one-tab.com/page/YQHccTVeTliv-UJT5FuExw

0:22 - Intel Announces Skylake-X: Bringing 18-Core HCC Silicon to Consumers for $1999
6:28 - CIA Created Toolkit for Hacking Hundreds of Routers Models
9:36 - Face-recognition system at Dubai airport soon
11:44 - Industroyer: Biggest threat to industrial control systems since Stuxnet
14:05 - Revealed: Facebook exposed identities of moderators to suspected terrorists
16:09 - Facebook (FB) built an AI system that learned to lie to get what it wants
18:07 - Pakistan: man sentenced to death for blasphemy on Facebook
19:58 - Germany Plans to Fingerprint Children and Spy on Personal Messages
21:07 - New law in Japan lets police arrest and surveil those merely planning or discussing certain acts, like copyright violation
22:55 - Linux Malware Infects Raspberry Pi Devices Making Them CryptoCurrency Mining Zombies
23:46 - Samsung Left Millions Vulnerable to Hackers Because It Forgot to Renew a Domain, Researchers Say
25:26 - Swapping Linux for Windows in Munich too risky after WannaCry attacks, warn Greens
27:20 - Pavel Durov on Twitter: During our team's 1-week visit to the US last year we had two attempts to bribe our devs by US agencies + pressure on me from the FBI.
30:00 - Firm That Made Mirai-Infected Webcams Gets Security Religion
31:36 - The NSA has linked the WannaCry computer worm to North Korea
33:09 - Kim Dotcom loses latest battle to recover seized assets
35:11 - Apple Mac computers targeted by ransomware and spyware
39:40 - Netflix has more American subscribers than cable TV
41:12 - Microsoft to Disable SMBv1 in Windows Starting This Fall
41:55 - You Can't Open the Microsoft Surface Laptop Without Literally Destroying It
43:24 - You Can Hack Some Mazda Cars with a USB Flash Drive
45:19 - A Roomba for Your Garden
48:14 - Amazon granted a patent that prevents in-store shoppers from online price checking
50:05 - Tesla Model X the First SUV Ever to Achieve 5-Star Crash Rating in Every Category
51:22 - Hello's sleep-tracking Kickstarter hit collapses
53:43 - Uber CEO to Take Leave, Have Diminished Role After Scandals
55:25 - Verizon closes $4.5B acquisition of Yahoo, Marissa Mayer resigns
58:14 - How to Track What Congress Is Doing on the Internet
1:01:45 - Popular GTA V Modding Tool Shuts Down, Community Explodes
1:05:41 - Atari CEO confirms the company is working on a new game console
1:06:33 - Konami reportedly blacklisting ex-employees across Japanese video game industry
1:08:37 - IISc Bangalore scientists are doing seed bombing with drones to plant a forest
1:10:43 - China's quantum satellite achieves 'spooky action' at record distance

Live Tech Q&A on Intel Launch Day (We suck at Streaming)


It looks like both Bitwit and Paul's Hardware got stuck with only getting Kaby Lake X for release day

What does a $250 7640X CPU do to a $490 X299 Motherboard?


Kaby Lake X i7-7740X DELIDDED Benchmarks!


The coverage for this Intel release all seems pretty unenthusiastic, and almost depressing...apparently that's what a rushed poorly planned release looks like, thanks Intel.

@Papusan Hey man!! The Kaby Lake X is OC'able to 5.2ghz and beyond on air!! That lame duck CPU in the X range would be great if back ported to the 1152 platform, maybe Intel will do that in the future!!
Starts at 01:37


Core i9-7900X Review: Fact-Checking Intel
Note: there is something wrong with their test configuration that is shown but not noticed / mentioned - The Ryzen 1800x CPU utilization was 11% while the Intel CPU utilization was 100% (see task manager at 09:58), so the test results are invalid...

 

from what ive seen the review, changing the L2 cache and L3 cache didn't change much in performance tbh, some software benefits them some doesn't. however from going the trend that legacy software and single threaded software doesn't benefit as much, i can see why its meant for the future as future software are more than likely to be coded for multi processing.

well unless 14-18c uses soldering, i will be buying an 8 or 10c and hope i can hit 5ghz. all eyes on silicon lottery BOYS

 

Intel 7820X i7 X299 8C16T CPU Review

 

hmscott, you don't seriously think anyone will view almost 4 hours of video to get 2 minutes worth of info?

Have you actually seen these yourself, in full?

Any written transcriptions of these links you keep posting?

If you have seen these, why don't you post your short points on them for us? (PP!).

 

Great that Intel started to complain+warned that people should not delidde and overclock their mainstream chips Just before the news of zero solder on their new HEDT Random?

I think he Watching 2 or 3 vid at same time

 

Really more mayonnaise thermal compound? In 112W+ TDP CPUs with bigger dies?!??!

 

https://twitter.com/AMD/status/877167059448233984
https://twitter.com/AMDServer

 

I hope not. I'm happy for RAID to die but I'd miss OCing if it was gone.

 

When single drives could offer what RAID0 promised just a few years before - and - when multiple drives (not in RAID) were used optimally gave even better performance than a single RAID0 volume, the dream of getting more performance for 'free' just disappeared from my 'wants'.

With O/C'ing; the same thing happened for me. When the newer CPU's/platforms were allowed to run as intended; the benefits were obvious, even if the bragging rights were not.

Just like with the yesteryear muscle cars that you could make go faster with a bigger carburetor and a screwdriver, today's cars are self tweaking with regards to performance, fuel efficiency and how much power they make; depending on how they are being driven. Overall; a much better experience than the fire breathing muscle cars of yore and with much more performance and driver control for the average driver too.

With the right car; go ahead and change the brains (i.e. 'chip') of your car to make it as fast as you like.

With processors: if the performance isn't where you need? Upgrade as needed and sell/donate/repurpose your old platforms as you see fit.

In my time with O/C'ing; there were very few benefits of staying with an older platform with O/C'ing applied vs. the benefits a new platform brought. And that is ignoring the stability/reliability aspects of it too.

Change is not to be feared; explore it and become familiar with it (for a while, ) until it is needed to be done again.

 

I do kinda miss those V8s that you could feel in your chest when they idled. But I get your point.

 

Yeah; I kinda miss them too. But when they're beside me now on a random road somewhere; they're just annoying.

A blip of the right pedal makes them the distant memories they're supposed to be. lol...

The truth is, we don't miss the actual 'thing' - we miss what we were when those 'things' were around (i.e. 'young and the world was ours for the taking').

 

An interesting point. And true, they're a lot slower than I remember.

 

Funny, an analogy to not overclocking as Intel says don't delid, puts bad toothpaste on HEDT, bad numbers on heat, and a reminder of Intel telling 7700K purchasers not to overclock or delid, while having high temp spikes... Hmmm. Especially from the cheerleader for Intel!

Sent from my SM-G900P using Tapatalk

 

Bad toothpaste? Why? It doesn't work to spec's?

Again; not a cheerleader for Intel - they just happen to be delivering what I want and need right now.

 

http://www.anandtech.com/show/11551...7000-series-cpus-launched-and-epyc-analysis/4

Sent from my SM-G900P using Tapatalk

 

"The connectivity here is set at a bidirectional 42.6 GB/sec per link, at around an average energy of ~2 pJ per bit (or 0.672W per link, 0.336W per die per link, totaling 4.032W for the chip). It is worth noting that Intel’s eDRAM for Broadwell was set as a 50 GB/s bidirectional link, so in essence moving off die in EPYC has a slightly slower bandwidth than Crystalwell. With a total of six links within the silicon, that provides a total of 2 terabits per second of data movement, although AMD didn’t state what the bottlenecks or latency values were."

Edit:
"Socket-to-socket communication is designed at the die level, rather than going through a singular interface. One die in each processor is linked to the same die in the other processor, meaning that for the worst-case scenario data has to make two hops to reach a core or memory controller on the other side of the system. Each link has a bidirectional 37.9 GB/s bandwidth, which is only slightly less than the intra-socket communication bandwidth, although we would expect socket-to-socket to have a slightly higher latency based on distance. AMD has not shared latency numbers at this time."

Edit 2:
"It is worth noting that the 42.6 GB/s die-to-die bandwidth is identical to the dual-channel memory bandwidth quoted per die:

http://images.anandtech.com/doci/11...or_press_and_analysts_06_19_2017-page-077.jpg

Time will tell if these become bottlenecks. Latency numbers please, I’d love to fill in that table above."

"As part of the launch today, AMD is announcing partners working with them to optimize the platform for various workloads. Sources say that this includes all the major cloud providers, as well as all the major OEMs. We saw several demo systems at the launch event with partners as well, such as HPE and Dell."

 

Sigh...

See:
http://www.tomshardware.com/news/amd-epyc-processor-models-pricing,34833.html

For people that think SPEC bm data mimics their usage... AMD just lost all credibility with them...

If the compiler does provide Intel users an advantage; then using an AMD platform is a disadvantage to them.