Info [Toms, Anand] AMD EPYC Benchmarks

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Yotsugi

Golden Member
Oct 16, 2017
1,029
487
106
Sounds good to me.
Yeah they'll continue stretching their already comfortable perf lead over Intel further.
Rome has been running at Google/Amazon/etc for awhile now anyway. I'm sure they've figured out how to make at least their in-house software work just fine on these machines. Whether or not that data becomes available to standard VM/hypervisor providers is another matter.
Rome is very simple and straightforward so it should never cause any excessive deployment troubles.
 
Reactions: lightmanek

ultimatebob

Lifer
Jul 1, 2001
25,134
2,446
126
ahemmm..... epyc has been out since 2017 and the official info of rome being anywhere between 90-110% faster at roughly the same power and removing the biggest disadvatage of naples (numa) has been out since computex, more than a half year now.
If you really were a smart IT guy and not a nitpicking problem-presenter who has either intel or lazyness in their blood, then you'd already know by now, what kind of value and cost-effectiveness rome represents to your company.

So, are you trying to convince me that there are NO architectural differences between the 1st and 2nd generation EPYC processors that could cause compatibility issues? I'm not buying it.

I know that this might sound crazy to AMD fanboys like yourself, but in the real world we do this crazy thing called "validation testing" before moving production workloads over to a new hardware platform. Maybe you should read up about it so you can make a more intelligent argument next time.

Insulting other users, and using phrases such
as "fanboys" is not allowed in the tech forums.

AT Mod Usandthem
 
Last edited by a moderator:

jpiniero

Lifer
Oct 1, 2010
15,292
5,807
136
I know that this might sound crazy to AMD fanboys like yourself, but in the real world we do this crazy thing called "validation testing" before moving production workloads over to a new hardware platform. Maybe you should read up about it so you can make a more intelligent argument next time.

Sounds more like an argument to go to the Cloud. Let Amazon deal with that.
 
Reactions: dmens

dmens

Platinum Member
Mar 18, 2005
2,275
965
136
The man is really a jewel, no one shines like him:



Yesterday he tweeted that he was "pretending" to talk nonsense.
Why?
To attract people to point out he was wrong.
Why?
To feel better blocking "fanboys".
But because he really believes everything he is saying he was "pretending to pretend", right?

Besides the people with "Piednoel Syndrome" and the usual bribery/FUD, there's still any plausible argument to deny what AMD achieved with Rome? There's still some light left for Intel in some small niche on a forgotten corner?

Personally what I want to see is not new products, is new prices.
Will Intel finally lower it's prices? Because the're will be a time when even with a better product people may stay away because of the ridiculous prices.

I have seen Francois' resume, if you can even call it that. It wasn't even a word document or pdf. That's a bad sign LOL.

Protip: don't take him seriously.
 

ultimatebob

Lifer
Jul 1, 2001
25,134
2,446
126
Even going into the cloud involves validation testing the specific instances available. Different hardware is different hardware.

Yeah, AMD did a presentation about their new EPYC instance types available in EC2 at AWS Summit in NYC, and even they mentioned that certain workloads (Like SAP HANA, I believe) still performed better on Intel hardware. In most cases, you can use an AMD instance type and get a 20% discount on your hosting charges for similar performance, though. When the 2nd gen EPYC instances are available on AWS, the performance will probably get even better.
 

Abwx

Lifer
Apr 2, 2011
11,591
4,408
136
https://www.storagereview.com/supermicro_now_offers_amd_epyc_7002_servers

For both the TPCx-IoT, and TPC-DS benchmark categories, Supermicro was able to use its new H12 A+ Servers to set world records. According to the company, for TPCx-IoT, performance of 472200.88 IoTps was established on Supermicro's H12 TwinPro 2U 4-node server where faster IoT gateway data analytics are critical for the coming explosion of IoT device numbers. Using a Supermicro H12 A+ BigTwin system the company was able to set another world record in TPC-DS.
 

Vattila

Senior member
Oct 22, 2004
809
1,412
136
Joel Hruska shares his view on the EPYC 2 launch and AMD's competitive position, reflecting on AMD's previous successful era with Opteron back in 2005:

"I don’t use phrases like “golden age” lightly. I’m using it now. While I make no projections on how long it will last, 7nm Epyc’s debut has made it official, as far as I’m concerned: Welcome to the second golden age of AMD."

extremetech.com

Interestingly, there is a fear in the article about history repeating itself — that AMD's success is like "hitting a rubber wall with a sledgehammer" — and that Intel will push AMD back into irrelevance. However, his look back at history also illuminates the differences between now and then.

Top of all, AMD is now not in dire need of making a $5B acquisition just to ensure a future. Getting graphics IP was vital back then. Nvidia was the courted partner, but that fell apart, and the forced marriage with ATI took place and hurt for years.

While ex-CEO Ruiz navigated the ATI acquisition and the factory spin-off, securing a deal with Abu Dhabi on the creation of GlobalFoundries, and deserves credit for that, his tenure was preoccupied with fighting Intel in the courts over their dirty business tactics. His successor, Dirk Meyer, quickly settled the dispute, although some argue he was too quick to do so and only got a paltry payment out of Intel.

Also, in 2005 the industry woke to the fact that "the free lunch was over", that single-core performance improvement was stagnating, and that parallel processing was the future. My understanding is that Dirk Meyer, who was in charge of the technical vision for AMD, struggled with integrating ATI while transitioning the engineers to fabless design methodology and SoC design — the use of PDKs, EDA and standard cell libraries, and moving away from manual circuit layout (something Intel may be facing now, under the direction of Jim Keller).

Dirk Meyer's vision for SoC design, "Fusion", HSA and the "APU" made longterm sense. Unfortunately, it took longer than expected (the first APU, Llano, was announced in 2011), and wasn't as profitable as hoped. As the new short-lived CEO he didn't manage to find a profitable way, the stepping stones above water, to get to the desired destination. And unfortunately, his bets on longterm CPU microarchitecture didn't pan out (while the chief architect on Bulldozer, Chuck Moore, got cancer and died).

This was all happening in a quickly changing market where the growth of mobile exploded, while the PC segment slowed down and stagnated. The AMD board reportedly wasn't pleased that, despite AMD's great IP, the company saw no share in this revolution. Many reports claim this is the reason Meyer got the boot (although I suspect it has more to do with lack of a profitable roadmap and execution).

Throughout this period, Intel was steaming ahead with a strong CPU architecture (in large parts built on AMD innovations and system design ideas, such as AMD64, integrated memory controller and coherent interconnect). Intel also had a huge lead in manufacture. They introduced the first "3D" transistor, the FinFET in 2011 to great fanfare, and leapt even further ahead.

Today the situation is quite different. After resetting the company under CEO Rory Read in 2011 with the hire of CTO Mark Papermaster, and subsequently Lisa Su and then Jim Keller in 2012, with Lisa Su executing the new vision as CEO since 2014, AMD again has a superior CPU architecture, but this time with a solid multi-year roadmap. They are also strong in GPU architecture, which is now a vital part of the mix in high-performance compute, and where Intel is the laggard. In manufacture, TSMC has overtaken Intel and has an unrelenting and competitive roadmap stretching far into the future. It is now uncertain, some say unlikely, Intel will ever get back in front.

And, AMD now has Lisa Su, Mark Papermaster and a strong management team. They have proven to have the technical insight — the ability to make the right bets — and a laser focus on roadmap execution and financial stability.
 
Last edited:

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
And, AMD now has Lisa Su, Mark Papermaster and a strong management team. They have proven to have the technical insight — the ability to make the right bets — and the laser focus on roadmap execution and financial stability.

The stars aligned for AMD as well. First their incompetent foundry partner GF got carbon copy of Samsung process, and then completely failed to execute on shrink.
You can be as much laser focused as you want, if you are contractually stuck on 20nm and no FINFETTs. I think it is very safe to say that AMD would be on some GF "11" nm node with FF, one that would be inferior to TSMC 16nm if those two things did not happen.

Even if GF did not fail 7nm, AMD would be in much much worse position, being capacity limited and probably stuck with poor initial yields from GF 7nm. Now only sky is the limit, Apple has proved that TSMC can produce hundreds of millions on 7nm chips that ~same size as chiplets, so AMD can supply whatever demand the market will generate for Rome.
 

Vattila

Senior member
Oct 22, 2004
809
1,412
136
Even if GF did not fail 7nm, AMD would be in much much worse position

I don't have clear evidence, but my understanding is that Lisa Su (and probably Papermaster) deserves much credit for not putting all eggs in the GF 7nm basket. From early on, after deciding to make their big bet on 7nm, they worked closely, I suspect, with both GF and TSMC. When it was clear that TSMC would be first to market, perhaps even with a better process, Lisa Su decided to commit to TSMC, even if leaving GF out of the game (and AMD possibly incurring penalties from the wafer-supply agreement). I am pretty sure that contributed to GF's decision to then pull out of 7nm development altogether.
 
Reactions: lightmanek

DrMrLordX

Lifer
Apr 27, 2000
22,117
11,785
136
AMD planned from the start to put Rome at TSMC. Also Vega20. Those were the products most likely to make a big profit for AMD with the current gen of tech . . . though they probably couldn't have guessed that Matisse would sell at 3x the anticipated rate. I think the only problem AMD would have had is with Matisse.
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,947
3,623
136
.

Even if GF did not fail 7nm, AMD would be in much much worse position, being capacity limited and probably stuck with poor initial yields from GF 7nm. Now only sky is the limit, Apple has proved that TSMC can produce hundreds of millions on 7nm chips that ~same size as chiplets, so AMD can supply whatever demand the market will generate for Rome.
You assume gf won't be 14/12nm limited for the io dies...... Lol
 
Reactions: lightmanek

DrMrLordX

Lifer
Apr 27, 2000
22,117
11,785
136
You assume gf won't be 14/12nm limited for the io dies...... Lol

GF is darn lucky to be getting that business. AMD just pulled most of their CPU dice off GF nodes (all they have left is Picasso). Polaris should go out of production soon if it isn't already. There isn't much left that AMD needs from GF.
 

krumme

Diamond Member
Oct 9, 2009
5,956
1,595
136
A new point here in relation to the wsa. With the extreme success of rome and zen2 amd have a good opportunity to load off more than anticipated of their agreement of this otherwise not competitive gf product. They thereby have an incentive to set rome and zen prices lower than what a normal io die would cost on the open market. The irony here is that in combination with the extraordinary rome performance the wsa then becomes more a pain for Intel shareholders than amd shareholders.
 
Reactions: senseamp

Vattila

Senior member
Oct 22, 2004
809
1,412
136
Regarding ExtremeTech's envisioned "2nd golden age" for AMD, it is interesting to look back on those competitive years at the start of the century. Due to the harsh competition, AMD's innovation and products never translated into solid earnings — except in year 2000, after the launch of the hugely successful K7 (Athlon/Athlon XP), which development was lead by Dirk Meyer. K7 replaced K6, a competitive processor that helped bring PC prices below the $1000 mark in the late 1990s. Fred Weber's (and Jim Keller's) K8 generation (Athlon 64/X2, Opteron) in 2003-2005, never saw much profit, despite making huge inroads in the server space by redefining standard system architecture (64-bit, integrated memory controller, multi-core and coherent direct-connect). In short, AMD never had much money to reinvest in their business.

It will be interesting to see what AMD can achieve over the next 10 years. Now that the longterm headwind in manufacture has turned into a tailwind — with close engagement with all the leading foundries — it seems that their roadmap has better chances than ever.

 
Last edited:
Reactions: DarthKyrie

krumme

Diamond Member
Oct 9, 2009
5,956
1,595
136
Regarding ExtremeTech's envisioned "2nd golden age" for AMD, it is interesting to look back on those competitive years at the start of the century. Due to the harsh competition, AMD's innovation and products never translated into solid earnings — except in year 2000, after the launch of the hugely successful K7 (Athlon), which development was lead by Dirk Meyer. K7 replaced K6, a competitive processor that helped bring PC prices below the $1000 mark in the late 1990s. Fred Weber's (and Jim Keller's) K8 generation (Athlon 64/XP/X2, Opteron) in 2003-2005, never saw much profit, despite making huge inroads in the server space by redefining standard system architecture (64-bit, integrated memory controller, multi-core and coherent direct-connect). In short, AMD never had much money to reinvest in their business.

It will be interesting to see what AMD can achieve over the next 10 years. Now that the longterm headwind in manufacture has turned into a tailwind — with close engagement with all the leading foundries — it seems that their roadmap has better chances than ever.

I have said this for the last decade. Amd have never made any money only used them. Lol. I got a lot of flak during the Dirk Meyers years with the small profit saying it was lipstick on a pig and in reality was a deficit. It clearly was just shareholder show pushing crap in front of the company until it nearly collapsed financially.

Now. Intel bribed their way through Athlon years. Otellini complaining about Michael from Dell "calling for more money" and such fun stuff but perhaps they lost more here than they realized...

Most people think the fine they got was to small. It was 1B or so as I recall and was a minor part of AMD debt but paid the most expensive.
What really was important was that as part of the settlement amd could produce cpu outside of own fabs. Now think about it. Without that deal amd would not exist in the future. And bobcat / zakate followed right away from tsmc in 50M# and together with jaguar and the consoles and some perhaps to smart financial diluting of shares made amd survive to make right tools for automated libraries and then zen.
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
So, are you trying to convince me that there are NO architectural differences between the 1st and 2nd generation EPYC processors that could cause compatibility issues? I'm not buying it.

I know that this might sound crazy to AMD fanboys like yourself, but in the real world we do this crazy thing called "validation testing" before moving production workloads over to a new hardware platform. Maybe you should read up about it so you can make a more intelligent argument next time.

Insulting other users, and using phrases such
as "fanboys" is not allowed in the tech forums.

AT Mod Usandthem
I don't think you understand me. So I'll leave it to you to be the smart IT guy. Calling me names or things, however, does not make you smart. Also don't forget to bring up the same argument ever so loud when Ice Lake comes out with an even more different architecture Peace out!
 

zinfamous

No Lifer
Jul 12, 2006
111,207
30,161
146
You assume gf won't be 14/12nm limited for the io dies...... Lol

yes, but it's only one 14/12 IO die to the 4 7nm chiplets required for a single Rome. ....so they need roughly 1/4th the capacity from GloFo as compared to TSMC to supply Rome to all of their vendors.
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,250
136
yes, but it's only one 14/12 IO die to the 4 7nm chiplets required for a single Rome. ....so they need roughly 1/4th the capacity from GloFo as compared to TSMC to supply Rome to all of their vendors.

Wouldn't that be 8 chiplets?
 
Mar 11, 2004
23,320
5,756
146
yes, but it's only one 14/12 IO die to the 4 7nm chiplets required for a single Rome. ....so they need roughly 1/4th the capacity from GloFo as compared to TSMC to supply Rome to all of their vendors.

That's not how that works. Capacity would be based on wafers which would be dictated by die size and number of die per wafers that you'd get. The large size of the I/O die means they probably need quite a good bit of wafers from GF, and so the ratio likely isn't nearly that lopsided. Looks like the I/O die is about the size of 5 Zen 2 chiplets. But the size also means they'll get inherently fewer functional dice from GF. I'm not sure if they're both using the same sized wafers though which could impact things as well.

On Ryzen it looks closer to 1:1 or 2:1 (when doing 2 chiplets). For 4 chiplet Rome its probably about 1:1 as well.

Although its moot with regards to it causing GF to be capacity constrained as there should be almost no way that would be the case unless GF pretty drastically cut production output.
 
Reactions: Schmide
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |