Techpowerup：Intel Says AMD EPYC Processors "Glued-together" in Official Slide Deck

[Topweasel]

AMD didn't accuse Intel of pasting the C2D's together. They just noted that theirs was a "true quadcore CPU". There is a distinction they were noting that theirs was a complete 4 core die and the competitors was not.

Intel's slide doesn't just state that theirs are a single die. They are suggesting A.) that by using Zeppelin dies its not a real server CPU and B.) with the gluing (hammering, stitching, or stapling are other derogatory terms) that it is an unprofessional pairing of dies.

Which is a bad move PR wise. Specially when the vocal public tends to root for the underdog. It's one thing to tell people why your stuff is good. But it's bad to attack or insult the competitor. Or in this case "Ours is a single die" is fine. "Theirs is just a bunch of desktop CPU's glued together" isn't"

 

[tential]

Press is having a field day in bashing Intel for this insanity.. http://www.pakpg.com/amd-epyc-processors-glued-together-intel-losing-epyc-cpu-performances-test/

So Intel makes fun of AMD's architecture when its faster in many apps, cheaper, and more power efficient. Unbelievable.

Please tell me what is linked is fake... Lol....

 

[Markfw]

There is another way to look at this. AMD figured out a way to have a very universal chip, that could be expanded in servers with little redesign(2 CCX, 4 CCX, 8 CCX on one physical socket, etc), and then FIRST implemented the base core in desktop, to weed out any problems (Ryzen) before going with the enterprise versions. (EPYC and threadripper). Not to mention the efficiency is WAY better than the Intel desktop, HEDT or Xeon lines.

And why do you think Ryzen is ECC compatible (if the motherboard supports it). Its a server CORE !

 

[nwr]

Of course it is a server core!
But it is also... glued together!
Intel: "Hey, our chip may be slower, run hotter & is more expensive. But at least it is not glued together. "

Sorry for joking around, but this time Intel seems a bit... glueless

 

[Cookie Monster]

There is another way to look at this. AMD figured out a way to have a very universal chip, that could be expanded in servers with little redesign(2 CCX, 4 CCX, 8 CCX on one physical socket, etc), and then FIRST implemented the base core in desktop, to weed out any problems (Ryzen) before going with the enterprise versions. (EPYC and threadripper). Not to mention the efficiency is WAY better than the Intel desktop, HEDT or Xeon lines.

And why do you think Ryzen is ECC compatible (if the motherboard supports it). Its a server CORE !

Its funny because a multi billion dollar company like Intel would kill to have a CPU architecture as modular (in terms of the MCM scheme) as Zen due to how flexible it is in terms of configuration and its associated manufacturing cost vs a monolithic approach that Intel has right now.

Just imagine once Zen 2 and 3 comes out with better CCX inter latency, high core clocks etc its going get a whole lot more competitive than ever.

 

[Markfw]

Oh, and an afterthought.... Even though Ryzen is not QUITE as fast as Intel for gaming, its probably because that was not designed as its primary function ! SERVERS. But it does a gang good job anyway. Certainly very much in the ballpark, a few percent lower.

 

[StefanR5R]

I am pleased with Intel. AMD did this to Clovertown with Barcelona. The shots have finally been returned.

having a hard time finding the actually super super aggressive slides...
ewh.ieee.org/r5/denver/sscs/Presentations/2007_08_Searles.pdf
//Not above slide:
(Monolithic, Native, True, etc) ((There was one that pointed towards TDP being a huge issue with any MCM approach as well. Implying, that it was worse than NetBurst TDP.))

Quote will have to do instead I guess:
“We came to the conclusion that, given the capabilities and performance with the monolithic design, it was clearly the right answer.”
- Randy Allen, AMD’s corporate vice president for server and workstation products. (2007)

AMD did this first, so it is fair game.

Aggressive undertones aside: These were quad core CPUs ten years ago with their set of problems at the time. But from today's perspective, for a low core count CPU, a monolithic die surely is the right way to do it. A high core count CPU is another matter.

 

[itsmydamnation]

Intel are doing this because unlike last time there its massive market growth, any increase in AMD market share will result in intel selling less processors. This wasn't the case in K7/K8 days.

intels biggest problem isn't right now, its tomorrow and it isn't Xeon performance, its Xeon performance vs EPYC performance relative to the number of employees (10k vs 100-110k). If EPYC takes server market share its probably going to hurt GP numbers more then revenue numbers which is going to put a lot of pressure on in terms of cost cutting.

 

[JoeRambo]

intels biggest problem isn't right now, its tomorrow and it isn't Xeon performance, its Xeon performance vs EPYC performance relative to the number of employees (10k vs 100-110k). If EPYC takes server market share its probably going to hurt GP numbers more then revenue numbers which is going to put a lot of pressure on in terms of cost cutting.

One can only hope the marketing and MBA types will go first. You know the ones that forced nasty AVX/AVX2/FMA segmentation on us for years (ironically Intel would probably be much better off if devs that skipped AVX2 due to all those Celerons/Pentiums not included, would have added support). The swan song of these guys is Skylake-SP both on HEDT and server, achievement in segmenting by everything. The guy who who asked for those "uber memory" SKUs can sure be proud of himself now that Intel's CPUs are getting rekt by epic disadvantage in memory capacity and pricing.

One can only dream, we all know who will go first

 

[maddie]

Intel are doing this because unlike last time there its massive market growth, any increase in AMD market share will result in intel selling less processors. This wasn't the case in K7/K8 days.

intels biggest problem isn't right now, its tomorrow and it isn't Xeon performance, its Xeon performance vs EPYC performance relative to the number of employees (10k vs 100-110k). If EPYC takes server market share its probably going to hurt GP numbers more then revenue numbers which is going to put a lot of pressure on in terms of cost cutting.

Agreed. Although I will say when, not if, EPYC takes server market share.

Panic is never good, but it's a very human response to the sudden realization that jobs are going to be lost. I wonder how hard the board will intervene against top management.

 

[TheGiant]

One can only hope the marketing and MBA types will go first. You know the ones that forced nasty AVX/AVX2/FMA segmentation on us for years (ironically Intel would probably be much better off if devs that skipped AVX2 due to all those Celerons/Pentiums not included, would have added support). The swan song of these guys is Skylake-SP both on HEDT and server, achievement in segmenting by everything. The guy who who asked for those "uber memory" SKUs can sure be proud of himself now that Intel's CPUs are getting rekt by epic disadvantage in memory capacity and pricing.

One can only dream, we all know who will go first

Exactly. Intel shows signals that there is majority of "when it will be done managers" just without knowing content. AMD on the other hand is totally driven by technical excellence now.

Man has to switch them, because we need to say, the MBAs do know how to get the money but they ruin the product. I think for intel the time to switch was yesterday.

 

[moinmoin]

I am pleased with Intel. AMD did this to Clovertown with Barcelona. The shots have finally been returned.

That's quite the revenge plan, 10 years in the making...

 

[Mopetar]

Oh, and an afterthought.... Even though Ryzen is not QUITE as fast as Intel for gaming, its probably because that was not designed as its primary function ! SERVERS. But it does a gang good job anyway. Certainly very much in the ballpark, a few percent lower.

Ryzen does fairly well in newer games and at high resolutions where the game is GPU-bound it truly doesn't matter what you use (even an i3 will hang with the best gaming CPUs if you're running 4k with a Titan or 1080 Ti) but I suspect that the biggest thing holding it back is just clock speed and the wall that Ryzen runs into a 4 GHz.

Intel still has a process advantage (and probably always will to some degree) but the IBM 7nm node that GF will be using is supposed to be designed for performance parts instead of mobile SoCs like the 14 nm Samsung process being used with Ryzen now, so the next iteration of Zen may see a nice clock boost along with any architectural improvements which will go a long way towards closing the gap.

There isn't any technical reason that Ryzen can't or shouldn't run games well, but anything using DX11 is more likely to ultimately be bound by single (or double) threaded performance, where Intel can win on clock speeds alone. I suspect that the move to 7 nm will give AMD a decent leg up in that department, though I have no doubt that Intel will still have an edge in their fab tech because its practically inconceivable that they would bung that up as well.

What I'm most interested in though is how much improvement AMD can get out of tuning and tweaking Zen. As it's a new architecture, there's no doubt significant room for improvement. Looking back at the first several iterations of the Core architecture, Intel was able to get double-digit gains clock-for-clock, so I think AMD will be able to close the gap with Intel here as well as Core seems to be closer to tapped out.

 

[Jan Olšan]

Look at the slides in question.. Easily visible BOGUS slides!

Again.. Show me an intel link..

Not to mention that Intel has probably quickly learned of the backlash from communities like ours and deleted/won't upload the slides to its website now. Only problem.. they can't force the tech press to redact the slides now that they are in the public domain!

It is completely normal for journalists to get different slides and other supporting resources in pdf than the companies put up for public download in their press/newsroom website section (I have some experience in this). Many materials are only sent to journalists via the company contacts. These slides are probably the case, clutching at conspiracies is silly here, IMHO.
Stuff like reviewers guides doesn't get released for public either.

 

[Shamrock]

Apparently, AMD is using some excellent glue...

Sorry, but it had to be said...

It's Gorilla Glue!

 

[DrMrLordX]

Are many server customers that ignorant, naive, stupid?
If not, and knowing a bit of human nature, will Intel's apparent assumption that they are, and using this argument backfire?

People who actually work with the servers from day to day? No, not really. Procurement people? Eh well some of them can be ignorant or are easily bribed.

 

[JimmiG]

AMD didn't accuse Intel of pasting the C2D's together. They just noted that theirs was a "true quadcore CPU". There is a distinction they were noting that theirs was a complete 4 core die and the competitors was not.

Intel's slide doesn't just state that theirs are a single die. They are suggesting A.) that by using Zeppelin dies its not a real server CPU and B.) with the gluing (hammering, stitching, or stapling are other derogatory terms) that it is an unprofessional pairing of dies.

Which is a bad move PR wise. Specially when the vocal public tends to root for the underdog. It's one thing to tell people why your stuff is good. But it's bad to attack or insult the competitor. Or in this case "Ours is a single die" is fine. "Theirs is just a bunch of desktop CPU's glued together" isn't"

Also, suggesting (well screaming from the rooftops, actually) that Epyc is using 4 "desktop" die is just incorrect. Zen was designed primarily for the server market. If anything, Ryzen consists of 1 "server" die.

AMD's "glue" in the form of Infinity Glue Fabric is pretty good compared to Intel's DMI...

 

[Ancalagon44]

Its funny because a multi billion dollar company like Intel would kill to have a CPU architecture as modular (in terms of the MCM scheme) as Zen due to how flexible it is in terms of configuration and its associated manufacturing cost vs a monolithic approach that Intel has right now.

Just imagine once Zen 2 and 3 comes out with better CCX inter latency, high core clocks etc its going get a whole lot more competitive than ever.

The irony is that you can bet Intel is investing into a modular architecture right now.

 

[NTMBK]

Intel's latest optimization guide recommends enabling Sub-NUMA Clustering on Skylake-SP for optimal performance. In other words- a giant unified L3 doesn't actually scale very well, and you're better off splitting it into local clusters...

https://software.intel.com/sites/de...4-ia-32-architectures-optimization-manual.pdf

This is all going to look very silly when Intel rolls out multi-die Xeons using EMIB.

 

[redzo]

Its funny because a multi billion dollar company like Intel would kill to have a CPU architecture as modular (in terms of the MCM scheme) as Zen due to how flexible it is in terms of configuration and its associated manufacturing cost vs a monolithic approach that Intel has right now.

I believe that intel would've never ended up pursuing a design like this. I know that Intel invests a lot in fabs. They have to put all those fabs at hard work in order to make up for their investment. I imagine that a foundry likes manufacturing different dies and sizes more than one simple die over and over again. A foundry probably makes a ton more money by manufacturing a big variety of chips. Fabs must be quite happy with intel's approach.

 

[Rngwn]

At least, gluing 4 "mainstream" dies to form a server CPU is still better than gluing 1 mainstream die to a bigger socket and call it a HEDT.

Sorry, I couldn't resist posting this.

 

[3DVagabond]

https://www.computerbase.de/2017-07/intel-xeon-skylake-sp-purley/3/#

Not all press agreed publishing what is basically FUD at its worst, an exemple below :

If AMD and Intel were members here on AT this would get an infraction for a personal attack and off topic.

 

[Insert_Nickname]

AMD's "glue" in the form of Infinity Glue Fabric is pretty good compared to Intel's DMI...

Infinity Fabric is a masterstroke. I'm really impressed how well it performs. Both on and off die.

Now if we can only persuade AMD to change the branding to Infinity Glue...

 

[raghu78]

The irony is that you can bet Intel is investing into a modular architecture right now.

When Intel launches Tigerlake server CPUs with Embedded Multi die Interconnect bridge in 2020 they will claim they have the superior glue.

Sent from my SM-G935V using Tapatalk

 

[CatMerc]

As someone on reddit nicely put:

The competition uses FLOOR TILES and GROUT in their bathroom floor installations. WE use a SOLID ****ING PIECE of GRANITE.