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

i agree with your entire post but Intel has a good record on clockspeeds.

I wouldn't brush this news away.

 

You don't have to, but I already have

I don't care if Intel is a little faster, I'm not putting more cash in their hands for a long time.

It's much less stressful to decide early on to support AMD, keep them in business, and make Intel work for a very long time for their next money from me.

AMD would need to have a massive failure to deliver for me to give up on them.

So far their releases with Ryzen and RX GPU's have been more than close enough in price and performance to win me over.

Not to mention AMD's massive efforts to keep improving performance after release, that's gold.

 

Actually, Intel regularly has a variance of 400-500MHz on best bin and worst bin chip. Yes, they click faster than AMD, but can be looser on where they fall on that spectrum at times.

With that said, they are great at IPC. Toothpaste use, though, gives part of that variance mentioned above. So, mixed. But yes, overall, they still have good products, even if incremental.

Sent from my SM-G900P using Tapatalk

 

Ryzen at 14nm+ might be able to reach Intel's clocks as its a refined process.
Wonder what's stopping current Ryzen from clocking to 4.5 Ghz though.
Some have said its lack of quad channel support or low memory speeds (mobo bios issues on that last one is being corrected).

As for quad channel support... could AMD bring that about with microcode upgrades, 14nm+ Ryzen, or not gonna show that until Ryzen 2?

Oh and in regards to those supposed benchmarks... I'd be skeptical of any benchmarks released before any official releases and wait for the real deal.
Either way, I'm getting myself that Asus ROG Ryzen 1700 with RX 580 (any word on when it will be available for purchase?)

 

So, I need to decompress all of that for the truths and falsities in it.

First, we have to remember, this is GloFo 14nm+ process, in part based on Samsung's 14nm process. If it has the improvements like Samsung's, then it will be quite good. But, 14nm+ is not likely to appear in processors until late 2017 or early 2018. Ryzen 2 is 7nm. There are arguments on when or how they will sandwich the refresh in, which Vega will also see a process refresh before Navi. Lisa Su said they are going to be taping out 7nm in Q4 2017 and suggested seeing 7nm as early as Q3 2018. For most lines, don't expect to see it until Q4 2018 or H1 2019. They are promising Zen 3 with 7nm+ by 2020. But, since it will all be on the same socket, not that big of a deal.

Then comes the reason for the cap on speed. Quad channel support would not be the cause on Ryzen. That is a matter of whether they have the IMC on chip to support it. Now, Microcode update MAY have some legs. Before now, they locked the secondary and tertiary timings. That can affect trying to overclock and balance all errors or mobo issues. But, I wouldn't hold much hope on this process. But, it does allow .25 multiplier, something that is new, which implementing that may be part of the reason. It is a new architecture and refinements will go further. But that is it. But, RYZEN WILL NEVER HAVE QUAD CHANNEL THIS GENERATION. The controller is not on the chips, so nothing will change that. That also means that Ryzen on 14nm+ will likely not have it. AM4+ may add it, which may happen with the jump to Ryzen 2, but that is pure speculation. TR is what is needed for Quad.

I agree on the benchmark, but it needed out there, presented with Xeon info and the BM of the 1800X to show it doesn't make sense, unless it only ran on one die. But, as the video posted by TAN mentioned, it looks a bit like a b350, which would be someone spoofing the CPU ID, etc. Can be done, but... So, we continue to wait.

And that laptop looks like a great DTR. Waiting to see Clevo answer...

 

AMD is putting some interesting thermal limits on Ryzen.

I am suspecting it to be related to it being a new platform, which doesn't clock well when above a certain grade of power, however, as software develops, I suspect Ryzen's OC will too.


Again, keep in mind that the Ryzen architecture is brand new. Time will only improve it. Just as Intel has "perfected" their binning and clocking process'.

 

Actually I might have a solution as to why ryzen is clocking to current levels. High density library design. They used it in carrizo and the API couldn't clock too high because otherwise the thermals went through the roof. Obviously, ryzen is incorporating this design if I'm not mistaken, and it could be contributing to lower clock speeds Vs Intel. 7nm+ might help AMD address this though and raise both the clocks and efficiency, but right now it's just a theory.

Can't say I'm too bothered to be honest about clock speeds. Ryzen already can be overclocked to 3.9 GHz which should be enough right now. It offers comparable performance in everything while giving you more cores at a much lower cost.

 

I'm very skeptical of those supposed TR geekbench scores.
I went to search for 1800x geekbench scores, and clocked at 3.6 GHz, it still scores mich higher than this supposed TR in both single threaded and multithreading.

https://browser.primatelabs.com/v4/cpu/search?dir=desc&q=ryzen+1800x&sort=score

Someone is either telling fibs (lies) on purpose, or the geekbench results that come up for 1800x are all falsified.

 

I suspect bad optimization.

 

Well,............. At least 24816/3.4*4 is 29915

TBH this is a skew we are unfamiliar with. it is possible it was designed to keep TDP down and could have some features cut, like Cache etc. (but it doesn't appear too). maybe we just have not seen the 1998x is why. Would be nice for an official explanation.

Edit; and lastly if this is someone spoofing the system they did it a few times.

 

An unreleased product from AmD that scores on 16 cores and 32 threads lower than 1800x which is clocked mere 200 MHz higher? I don't think so. Something big is amiss. The scores should be close to double for multithreading given near perfect scaling due to infinity fabric. As I said something is very wrong with these supposed scores.

 

Geekbench is famous for favoring Intel CPU's. I just want an official explanation from AMD. Be whatever that is, at least we will then know.

 

That's not hard to do. Once you spoof it, you can run it as much as you want. That means nothing that there were multiple occurrences. But, why dual channel ram? Why is it running at the Ryzen 8 core levels? Etc.

Sent from my SM-G900P using Tapatalk

 

Favoring doesn't explain a score matching an 8 core on a supposed 16 core.

Sent from my SM-G900P using Tapatalk

 

http://www.tweaktown.com/news/58013/intels-skylake-use-integrated-voltage-regulator/index.html
So, IVR is used...

Sent from my SM-G900P using Tapatalk

 

From the link quoted below:

I'm not surprised at all...

(You don't just 'bolt on' cpu parts and create a new architecture - like some peeps here think. Intel is much more thorough than that).

 

actually thats how i use to see it too, some benchmark does favor a company and their specific design more than other. however the other way of saying it also is true, that intel's cpu under a specified workload (something similar to geekbench, whatever it maybe) will be much faster than AMD's cpu.

now as to how realistic of that workload translate to real life workload is another story, could be 1% could be 0%

just to add another comment here, that 18c intel cpu using possibly using HCC better be soldered to it's IHS..

 

Agreed, synthetics are nothing to go by real world. The concern is not of the overall score but the scaling of it going from 8 to 16 cores. It is highly possible as the unique features of the cpu not every test uses all 16 cores or uses them properly. If that is what it is AMD should just come out and say so.

 

Does Intel use solder on IHS for any new CPU generations these days?

Intel probably sold off that equipment years ago.

AMD still has that equipment hanging around and maybe can't afford to switch to toothpaste.

A nice side benefit to AMD costing savings

 

if we back date when they start ditching soldering, mainstream came first.

3770k was the first i believe to use non-soldering, however the HEDT up until broadwell E are still soldered. I think theres a clear reason why intel separate xeon socket and HEDT socket and its much obvious now, to save money and make more money off server side of things, which made sense because had they use same socket and same LCC for xeons as for their HEDT they would have no reason to NOT solder as it'll be quicker. simply pick up any low end xeons thats 4-10 cores and then unlock them and sell them for HEDT, when these xeons originally are soldered it makes no sense to desolder them as thats more work and cost extra.

now starting skylake-X HEDT is LGA 2066 while xeon is LGA 3647 i think, so clearly seperated and no longer used on the same board/chipset. hence intel can actually make them separately and not solder these chips.

 

I am almost sure I have seen where the intent was paste. One of the reasons you see delidding as a common practice with Intel CPU's. I would never do that too a $2,000 CPU. Nor would I expect another to do so as well. This is why if you consider the format the CPU is what it i and do not expect, especially those high core count ones, to get something for nothing.

Edit; I think they delided a 7900x to get the 5GHz, I do not believe it was originally soldered and it is considered the current HEDT.

 

thats definitely a cherry chip, consumer will not be able to get that for sure. we will be lucky to see 8c with 4.6ghz with a decent voltage, which is still very ver ygood dont get me wrong.

 

They have switched the entire HEDT to toothpaste, according to reports. The 10 core did only hit 5GHz with being delidded. So, 4.8 on a cherry chip, we'll see how much less otherwise.

Sent from my SM-G900P using Tapatalk

 

Just another thing irking me, in the submissions page they list the processor code name as "Summit Ridge" but it should be "Whitehaven" for the 16c,32t. Another link was pulled from a video.

https://www.bing.com/videos/search?q=threadripper&view=detail&mid=CBC8E46C2ADCE33CD741CBC8E46C2ADCE33CD741&FORM=EWREVC&mmscn=wnmp&ru=/search?q=threadripper&qs=n&form=QBRE&sp=-1&pq=threadripper&sc=8-12&sk=&cvid=7BBF8498C1D54156AB2EE28F07B517D1

 

One more stone as to why it is a fake...

Edit: When you spoof some things, you might not think to change it all. The other explanation is the program filled in the blanks, but...

Sent from my SM-G900P using Tapatalk

 

But below is why AMD needs to get on this.
https://www.techpowerup.com/234342/flagship-amd-ryzen-threadripper-16-core-chip-appears-on-geekbench

 

When delidding became popular. From Ekwb.com

"Delidding got really popular in late 2011 with the launch of the Ivy Bridge microarchitecture which was the third generation of processors based on the 22nm manufacturing process developed by Intel®. The new CPU was the upgrade of the Sandy Bridge architecture and should have brought better power efficiency, lower TDP, and lower temperatures. However, users noticed that their Ivy Bridge processors spiked higher temperatures than the old Sandy Bridge CPUs. This made overclocking a bit of an issue, and overclocked processors were hitting 100°C very easily."

 

http://www.cpu-monkey.com/site.php?...annels=&memory_ecc=&pcie_version=&pcie_lanes=

I post this here too as besides the benchmarks being competitive with Intel, all skews support ECC as well, will Intel and x299 do the same for its business consumers?

 

http://www.tweaktown.com/image.php?...-epyc-32c-64t-2ghz-2tb-ram-128-pcie-lanes.jpg
http://www.tweaktown.com/news/58052/amd-epyc-32c-64t-3-2ghz-2tb-ram-128-pcie-lanes/index.html

Sent from my SM-G900P using Tapatalk

Edit: Epyc speed and pricing!

 

So, from those links;

At the $800 USD level; Epyc offers 70% more performance with 60% more cores and almost 90% more power (TDP). At least with this 'cloud' bm 'score' they're offering...

But no, I'm not impressed. I would be surprised if any data center that needs multiple servers would be too.

What is even more interesting to me though; they're comparing future CPU's to what Intel released a year ago.

I love the spec's that the platform offers right now. The overall package? Not quite prime time, imho.

Again; (this should go without saying by this point...) I'll be interested to see actual products from both competitors go head to head in a real world shoot out (please - whoever does the comparisons... leave the synthetic bm 'scores' to the kids...).

 

So, sells start on Epyc June 20th. I'll post the competitive date for Intel's HEDT in the appropriate thread.

Edit: so, got the threads confused for a second. Intel takes preorders for the KB-x and SL-X 6-10 starting June 19, actual sell a week after.

On June 19th, DDR5 will be shown off with a convention this fall. Make what you will with that. But, servers based on deep learning and heavy duty data storage, AMD has positioned themselves well for. I just wonder/wish that they have one or two chips unlocked like Intel did. Imagine a 16 core with 8 channel memory driving a home server on a water chiller...

Sent from my SM-G900P using Tapatalk

 

Where did you get that info (June 20)?

And this is the appropriate thread...

 

Those links do not say Epyc pricing but the price is supposedly what the Intel variant listed will cost. Most of the other prices were generally right but I founf the 2699a v4 for 2175 at Amazon.

https://www.amazon.com/INTEL-PROCES...F8&qid=1497593714&sr=8-1&keywords=E5-2699a+v4

Edit, it is the right thread for Epyc vs. Xeon but pricing (if we truly had it) as specific it could go to the Ryzen thread if not being used for comparison sake.

 

I'm assuming that they put Intel's chips that are at that price against the Epyc chips around that price with performance to show the improvement with this line.

Sent from my SM-G900P using Tapatalk

 

LOL, never happened and you know it too. One thing management is in love with is statistics. The more data you can flow to them the better, synthetics or otherwise.

 

So that means AMD thinks it's Epyc 7601 is worth $4K USD? lol...

Info from AMD is like finding chicken teeth... and if the June 20 launch date is correct... Why AMD Why? (why no info...).

 

I have to agree with the pricing I myself found, they may need to revisit that one again. Remember this is all very early too. They actually are not released yet and maybe 3-4 months ago when this was planed those prices were more relevant then.

Edit; Myself for competitive pricing I was thinking for the top end Epyc more $2,200 to $2,500 if they were being aggressive but even this could go down if Intel counters.
If they want to compare at the highest end well use a e7-8890 v4, that is their top end right now. and a TDP of 165w.

http://ark.intel.com/products/93790/Intel-Xeon-Processor-E7-8890-v4-60M-Cache-2_20-GHz

 

Well, four days to find out if true on Epyc.
https://www.extremetech.com/computi...h-june-20-consumer-vega-gpus-follow-late-july


Sent from my SM-G900P using Tapatalk

 

That is what they should be comparing, imo. TDP to TDP and then let's see where the performance falls.

That 15C/30T e7-8890 v4 CPU is also a year past launch/availability too. Guess AMD doesn't value efficiency as much as it claims it does?

I can't find any PM comparison 'scores' to their E5-2680 v3, but with ~18% increase in clock speed to the E7... it looks like Intel has eclipsed or at least equalled the top Epyc from a year ago... and with 2C/4T less... (at the same/similar TDP).

'Performance' on it's own means little in today's world. There are always many other factors to consider. Price being the least of them when absolute productivity is the goal.

I've mentioned this many times in the past w/regards to SSD's: if we had an unlimited power budget on our notebooks; we'd have storage subsystem performance that would equal and possibly exceed DT systems...

Instead, the storage subsystem options that make sense for notebooks (still, and going forward...) are the models that balance that raw high performance along with high efficiency.

Processors for servers are no different. Very few shops run a single platform (no backup plan!) and when considering multiple systems, using platforms with ~90% more power usage to gain multicore performance in the low 20% range isn't what one does if they still want a job next Monday morning...

 

So, first, do you believe AMD did this comparison or a third party. If the latter, no one should be discussing that except to tell the person that made the comparison to choose other hardware. Second, the 8xxx series is made for 8 socket connections. Although AMDs two socket is 8 dies, it is not comparable, as you don't often have the 8xxx ran by itself or in a pair of twos. So, should it be done, yes. Will it, open question. Blame them for third parties, now you are being absurd.

As to pricing, you are saying charge so low it may not make sense, especially as Intel has done little on that front. Also, with server chips, the suggested retail isn't what is paid for either company's silicon. So what are you talking about?

Sent from my SM-G900P using Tapatalk

 

Not too sure who did the testing but even if a third party it probably was at the bequest of AMD. As far as connections I am not saying any Xeon should be directly compared but being as the tech exists for the cores it can be imported to a degree of success most likely to x299. FYI you contract someone to do a job and they blow it, it always comes back to you. If wrong you can try and blame the third party but in the end you are responsible.

FYI, in Cinebench r15 the TR 1998x most likely as clocked will beat the 8890 v4, see below. And I was going by prices I found on either Amazon or newegg except the 8890 v4 as that is a much more expensive chip. So I do research before I post, do you?

http://hwbot.org/hardware/processor/xeon_e7_8890_v4/

Edit; Oh, there is a e7-8894 v4 but it seems announced in Feb. but there is little to no info out there, anyone have a single one of these running somewhere?

 

First, we have our wires crossed. You are comparing the TR numbers, an HEDT platform limited to 16 cores, to a XEON e7. Besides how wrong I believe that comparison to be (my opinion) (edit: On later reflection, I do see your point on the comparison, but still must add the caveat that if the speed is slowed down too much due to additional cores, then a lower core count will overtake the higher core counts, but that will not matter as much in certain highly multi-threaded applications specifically built for better core and node usage, but point taken), I was comparing EPYC to the Xeon line, as both are SERVER chips. Now, I'll agree HEDT chips you don't get much discount on, but server chips, it is rare for customers to pay full price for these chips as Intel and AMD offer deals to their customers. It is also why it is harder to find e7s floating around than the e5 2xxx series, which is used for workstations as well as servers. So, it was going by the MSRP of the chips that I took issue with, as that means nothing on what is charged.

Next is to address the numbers. CPU-monkey is a reseller of chips, dealing in a lot of ES and QS. So, their numbers likely are from in-house testing, not done at someone's behest. The leak to videocardz is harder to say, as they have both been fed numbers from manufacturers, but they also are good at scouring for tidbits of information. Considering you are focused on the TR numbers, we are not talking about the same thing, hence the loss of sense and reference.

 

Ok, the reason for the comparison is All tech being brought down to beat TR and Epyc are Xeon tech. As far as the 8890 v4 if it can't beat a TR how can it possibly beat an Epyc. as far as the numbers I am talking about the Epyc and all price range comparisons on the list. This is the problem having all these as one thread and tech being mixed between platforms like Xeon.

as far as the contract someone, I am referring to the Epyc list. The top tier chip they use is way cheaper on amazon. That is the one they should have replaced with the 8890 v4, it would have been more fair and correct and it still would have been spanked. Now again I know they are not using R15 for their comparisons but I still think Epyc would be the winner.

I am sure there might have been one chip and one test on a TR but I highly suspect they extrapolated it to all skews for the benchmarks. Not sure if people remeber this from pre-computex.

http://wccftech.com/amd-ryzen-9-lineup-threadripper/

 

OK, thank you for the clarification and position, as that I agree with. But, once again, access to higher Xeons (e7 8xxx) is restricted unless you find them secondarily on the market (used) or you are purchasing a direct order. So prices, even speculative ones currently (on Ryzen, although I agree you can find Xeons cheaper than what is listed there as well) bandied about, shouldn't be given full weight, as Intel can offer steep discounts on the chips (and I'm betting similar pricing will be negotiated on the Epyc chips). I more used the Epyc approximated pricing to lend credence to the TR expected pricing, which will heavily undercut Intel. So, we weren't too far off, we both just needed more information from each other.

As for that assumption, quite possible. What we have seen from rumors is cannardPC got hold of an Epyc chip, which the 2.x base and 3.x boost matches one of the skus in the videocardz release extremely closely. Meanwhile, the other leak on the 16 core back in March matched the TR specs pretty closely as well. Then we have the 12C/24T leak a week or two ago, whose clock speeds are roughly in line, but other issues were found with that leak. The 1950X is an outlier and looks spoofed for many reasons. So, we are looking at being able to confirm some rumors, ignore others, etc.

Now, the Epyc series being released in a couple days means we'll have a lot more on scaling ability real soon. Meanwhile, the Xeons for SL-X are a no show and should properly be called out, except for the one benchmark score in GB4 from months ago:
http://wccftech.com/intel-skylake-e5-2699-v5-32-core-geekbench-score-leaked/
(edit: remember that the linux distro of GB4 scores higher than the Windows version, as is seen with Ryzen scores)
(edit 2: also note - this was before the change in cache scheme and since this, I do not remember if Intel announced as the top a 28-core chip or 32-core chip, so salt...)

But, we only have GB3 scores for TR so far (which scores higher than GB4, so we must wait). But, if these scores and speeds are true, Intel may have trouble on performance competition for the 32 Epyc vs Xeon. But, calling for more info from AMD when Intel has given nothing is improper, or saying comparing it to v4 is wrong when Intel has given nothing else to compare it to.

But, back to the point, I'd say that if they did get hold of an ES, it would have been the 12 or the 16. Also, with the stepping on core count for Epyc, it makes less sense to argue for the existence of the 10 and 14 core chips. But we shall see.

Edit: I was wrong, either missed someone else posting the new generations benchmarks (took my nephew to a game yesterday and to other things). Here are the new Intel Xeon benches:
http://wccftech.com/intel-28-core-xeon-platinum-8176-8168-xeon-gold-6161-6142-benchmarks/
https://hothardware.com/news/intel-...um-8176-server-cpu-rocks-cinebench-benchmarks
Making sure to point out that those benches were using dual P setups, not single P.

 

This is more inline with my original thoughts, hopefully the Epyc can do the 6,000 or so in R15 and since the reseller is $3,222 for that CPU it puts more in line with AMD asking $2200-$2500. Agreed Intel can cut prices on Xeon and that is exactly what we want. Intel ups the capabilities of their chips and makes the cost more realistic.

Good times to look forward too.

 

https://www.bit-tech.net/news/hardware/2017/06/16/intel-mesh-architecture/1
On Intel mesh, it has been used for awhile now in Xeon Phi and goes together with the recent report of components at different nodes on a CPU. As such, it should be ignored that this is new or special (edit: meaning this is a natural progression needed to solve a problem which has been in the works for years, not something recent). It should work with the EMIB for designs moving forward (which would utilize a mesh over a ring better).
http://www.pcworld.com/article/3185...e-cobbled-together-using-different-parts.html
https://hothardware.com/news/intel-kaby-lake-g-series-leaks-with-heterogenous-multi-die-design

Edit:
https://insidehpc.com/2017/03/intel...ogramming-interactions-memory-modes-nutshell/
Discussion on programming for the mesh...

 

Intel i9 7900X benched! Performance numbers, etc.:
http://hexus.net/tech/reviews/cpu/107017-intel-core-i9-7900x-14nm-skylake-x/
This is not in violation of NDA and does not seem to be pushed by Intel (indicia of truthfulness).
http://www.guru3d.com/news-story/intel-core-i9-7900x-review-published.html

Edit: One thing I do not like is they didn't show the overclocked score against the OC scores of the other chips, which could effect placement in the lineup on both sides, Intel and AMD.

 

Hexus 7900k 10c/20t - skylake-x review out early lol breaking NDA.

http://hexus.net/tech/reviews/cpu/107017-intel-core-i9-7900x-14nm-skylake-x/?page=3

 

Yeah, those are about where I figured the chip would lie. Not overclocked the supposed 1955x is not too far behind it. That would only be a $550 chip. If they can get all skews of TR out it should be a winning combination and not answerd until October by Intel.

Edit; I loved the one comment they made "can I hear your wallet opening". NOT.

 

I ninja'd you. Also, they cannot break NDA because they were not given the chip subject to NDA (they got it through a source, not directly from Intel). http://www.guru3d.com/news-story/intel-core-i9-7900x-review-published.html