Intel gets EPYC-ally annihilated - Benchmarks & Testing (2x 64 "ROME" cores)

[The Benjamins]

Here is a video of Cinebench R15 and R20 runs

 

https://hwbot.org/submission/4216395_sth_patrick_cinebench___r15_2x_epyc_7742_11080_cb

 

 

[leadeater]

7 hours ago, VegetableStu said:

@leadeater what do you make of this? o_o There's a chip purported to be a 32-core Zen 2 threadripper with 16MB x8 L3

(is it a misdivision, or is it possible to weld off half the cache? or two CCXes are sharing one L3 section? whaaaaaaat)

Probably fake, but also each CCX is technically 2x 16MB rather than 1x 32MB so that may be how they are representing it so it's just 4 CCDs. Check an EPYC2 result to see. Still though, probably fake.

[Flying Sausages]

GO AMD!

[leadeater]

27 minutes ago, VegetableStu said:

i looked up a 7452 result (4x8 core 128MB L3), and it says "16MB x8" o_o so that checks out, but it doesn't rule out spoofing by imitation though ._.

https://browser.geekbench.com/v4/cpu/14236174

 

also if that's the case, yeah I'd agree with that sentiment that it's a 4 CCD part. no point making threadripper needlessly expensive to begin with ._.

 

EDIT: yeah, geekbench explicitly states the L3 count rather than the per-CCX count. a 8x 8 core CPU is being shown with 16x 16MB L3

https://browser.geekbench.com/v4/cpu/14236448

A higher clock higher TDP TR part with 32 cores and up to 4.6Ghz single core boost is going to be amazing though, if they allow that higher single core (kind of no point tbh on that type of part). All 32 cores at 4.1Ghz will be about 300W just for the cores and another 25W to 30W for the IOD. 2990WX was a 250W TDP part, AMD could do the same again with a PPT of 330W.

[leadeater]

Just now, VegetableStu said:

holy crap I don't know if I even have the power budget for that O_O (not just the motherboard, but like buying a suitable PSU, or even wanting to own a system with that high of a power target)

Check my edit, with the way AMD has PPT now the TDP likely doesn't have to increase and just allow configurable PPT as it is now with Ryzen 3000.

[leadeater]

13 minutes ago, VegetableStu said:

derped. pls ignore ,_,

Nah, make me 

[lonewolf5460]

I hope intel keeps the fight above the waist man that 56 core part and the asking price just ouch. I really hope the don't pull the same stunts they did so long ago. Hopefully AMD can stockpile enough cash to keep the R&D going so they stay competitive because we all know intel is lighting money on fire to get back #1 and what ever comes out of their fabs WILL be good. Bet money after talking all that smack they go to a chiplet design, monolithic dies are not feasible anymore.

[BiG StroOnZ]

Somewhat related (maybe a new thread is justifiable) -  

 

New AMD Ryzen Threadripper shows off, hugely faster than the old 32-core: 

 

Quote

 

 

You will have noticed that benchmarks of an unpublished processor have surfaced on Twitter, it holds a Geekbench entry named AMD WhiteHavenOC-CP with 32 cores, and that likely is Threadripper 3000 ( "Castle Peak") with 32-cores. The entry indicates a base clock of 3.6 GHz for the 32-cores.

 

Identified as Whitehaven in the benchmark, the processor positions itself at a veru healthy single-core score of 5,677 points and a multi-core score of 94,772. Compared that to a system with an AMD Ryzen Threadripper 2990WX, this results in an increase of 6 percent for the single-core score and over 35 percent for the multi-core score. Actually, two entries in the Geekbench database can be spotted. Rumored indicate ThreadRipper 3000 could be released in October close to Core i10-10000 Intel Core ( "Cascade Lake X").

 

Source: https://browser.geekbench.com/v4/cpu/search?q= AMD+Sharkstooth

Source 2: https://www.guru3d.com/news-story/new-amd-ryzen-threadripper-shows-off-hugely-faster-than-the-old-32-core.html

[leadeater]

6 hours ago, VegetableStu said:

someone brought up that a few early EPYC servers didn't have enough BIOS space for the new 7002 CPUs, which is posited to affect TR3K on X399 as a whole because almost none of the boards out there has the space for a TR BIOS that's double its current size (if we're extrapolating the size increase from Zen 2 Ryzen) 

Ryzen has more of a problem with that than TR/X399 does, its due to the number of SKUs a motherboard has to support and X399 is a very limited platform in comparison to AM4. AM4 has compiled binaries for multiple generations of Ryzen desktop and APU processors which requires a lot of space, TR on the other hand has like 7 or 8 in total currently. 

[GOTSpectrum]

@LinusTech you guys doing any videos with these things, that would be awesome

[Jito463]

1 hour ago, leadeater said:

AM4 has compiled binaries for multiple generations of Ryzen desktop and APU processors which requires a lot of space

Given the issue with dropping Bristol Ridge support on AM4, I'd say it's high time the BIOS chips get an update in size by default.  It can't add more than a few cents per board to go from 16MB to 32MB of flashable ROM.

[Fnige]

On 8/15/2019 at 12:20 AM, valdyrgramr said:

Just tell your doctor you have ADHD.

Probably do, idk

[Results45]

So my dad works in Intel's Data Center Group and here are his thoughts (in summmary): 

 

"Yeah we look at how competitors are doing from time to time, but we keep our heads down and work as hard as we can with partners on future products and hardware solutions. Yes Intel as a company has gotten kind of complacent in the recent past [insert AMD as living proof], but the majority of customers will likely stick with Xeons for now due to advantages in platform support for things like Optane DC, Agilex FPGAs, and CXL interconnect interface."

 

So yeah, AMD has the lead in terms of performance per dollar, but Intel's playing the long game too so I guess we'll see how well AMD does with introducing competitive support for more features and technologies in coming years. May the best chip designer win!  

[mr moose]

8 hours ago, Results45 said:

So my dad works in Intel's Data Center Group and here are his thoughts (in summmary): 

 

"Yeah we look at how competitors are doing from time to time, but we keep our heads down and work as hard as we can with partners on future products and hardware solutions. Yes Intel as a company has gotten kind of complacent in the recent past [insert AMD as living proof], but the majority of customers will likely stick with Xeons for now due to advantages in platform support for things like Optane DC, Agilex FPGAs, and CXL interconnect interface."

 

So yeah, AMD has the lead in terms of performance per dollar, but Intel's playing the long game too so I guess we'll see how well AMD does with introducing competitive support for more features and technologies in coming years. May the best chip designer win!  

 

The only thing about that that doesn't sit well with me is that it infers AMD only got in front because Intel got complacent,  What if AMD got in front because they worked their ass off developing a product that was good?   I personally don't think Intel got complacent, I think they just had a few issues and AMD caught up.

[GoldenLag]

32 minutes ago, mr moose said:

 

The only thing about that that doesn't sit well with me is that it infers AMD only got in front because Intel got complacent,  What if AMD got in front because they worked their ass off developing a product that was good?   I personally don't think Intel got complacent, I think they just had a few issues and AMD caught up.

And unlike the consumerspace. Intel hasnt exactly been sitting still in the datacenter sector. 

 

The 28 core Xeon is not cheap, and certainly not a product someone complacent would release. 

There was talk of 35% yields on the 28 core iirc.

[AngryBeaver]

When he says complacent I have to agree. Intel was only releasing small improvements by and large. If they don't need to drop a ton into r&d or can release those new products slowly over a few years then it work out better for them when competition is low.

 

Now I think the big problem for them now was all of the 10nm issues and the fact they are still hanging on to the monolithic design which does hurt yields for the higher core count chips.

 

Now this isn't to say they won't be back on top in short order... it just means they weren't expecting the major wins AMD made with Ryzen. In the end this works out better for the consumers.

[mr moose]

On 10/1/2019 at 10:50 PM, AngryBeaver said:

When he says complacent I have to agree. Intel was only releasing small improvements by and large. If they don't need to drop a ton into r&d or can release those new products slowly over a few years then it work out better for them when competition is low.

this assumes that the reason Intel were only releasing minor increments was intentional and not because it was the limit of their abilities.  Remember Intel got caught with their pants down two or three times now and haven't been able to answer.  x299 was a shamozzle response to Ryzen, the 9000 series barely held their lead then they couldn't even match AMD so the rumor is the 10 series will be cheaper.  They are all signs that Intel are trying really hard to keep in front but just can't. This tells me they are not being complacent at all.   maybe when they crack 10nm things will change, who knows (not I), but until then the writing on the wall says they are pushing their current tech as hard as they can.

 

On 10/1/2019 at 10:50 PM, AngryBeaver said:

Now I think the big problem for them now was all of the 10nm issues and the fact they are still hanging on to the monolithic design which does hurt yields for the higher core count chips.

I agree, 10nm has caused them grief, but they have been playing with 3d stacking and gluing for a while now.  AMD's gluing seems to have yielded results Intel didn't think were possible.

 

On 10/1/2019 at 10:50 PM, AngryBeaver said:

Now this isn't to say they won't be back on top in short order... it just means they weren't expecting the major wins AMD made with Ryzen. In the end this works out better for the consumers.

 

I actually think they were expecting it, but just couldn't keep up. I think both Intel and AMD have a very good idea where each other is at.

 

 

[porina]

18 minutes ago, mr moose said:

x299 was a shamozzle response to Ryzen

Don't know what a shamozzle is, but I assume it isn't good. X299 I felt was a generally consistent release in line with past HEDT launches. The two different things were the lower Kaby Lake processors (I still want these for OC fun), and more cores at top end than previously to go against Threadripper. The mid of the range was solid enough. The Threadripper of the time might offer more cores for less, but they were weaker cores. So you end up having two HEDT platforms that didn't directly go against each other and that was the bigger decider of which platform you might go if you were wanting more than consumer products.

[mr moose]

49 minutes ago, porina said:

Don't know what a shamozzle is, but I assume it isn't good.

messy, disjointed or generally unorganized.  Having to buy keys for raid, the seemingly rushed state of release, lack of clear information etc. 

49 minutes ago, porina said:

X299 I felt was a generally consistent release in line with past HEDT launches. The two different things were the lower Kaby Lake processors (I still want these for OC fun), and more cores at top end than previously to go against Threadripper. The mid of the range was solid enough. The Threadripper of the time might offer more cores for less, but they were weaker cores. So you end up having two HEDT platforms that didn't directly go against each other and that was the bigger decider of which platform you might go if you were wanting more than consumer products.

Yes the product was fine in the end and performed well, but it really seemed it wasn't ready for launch when it did hence my conclusion Intel weren't relaxing or sandbagging. Had they been doing that I think it would have been a much smoother release.

[Unixsystem]

On 8/29/2019 at 3:18 AM, mr moose said:

 

The only thing about that that doesn't sit well with me is that it infers AMD only got in front because Intel got complacent,  What if AMD got in front because they worked their ass off developing a product that was good?   I personally don't think Intel got complacent, I think they just had a few issues and AMD caught up.

I absolutely love the AMD chiplet strategy. It's an incredibly smart way to sidestep both their limited funds from R&D, the yields of new nodes and the inherent limits we've been hitting with regards to Moore's Law. 

 

That said, if Intel had their original vision for 10nm out in 2016 with a substantial increase in both power efficiency and IPC, Ryzen would have been a pretty minor success. A huge step up from what they had, but the shocker with Ryzen 1 was that they were slightly behind but within striking distance of anything Intel had. If Intel could have pivoted around and 3-6 months later slapped down Cannonlake with +20% IPC and 8+ cores due to the increased density, Ryzen would have immediately been relegated back to the budget bin. Still great CPUs, but nothing they had would be on top of any benchmarks, even the multithreaded ones. 

 

That's not to say that it's complacency per se that led to Zen's success, but Intel had a few bad calls (and some that were totally outside their control) that left the door wide open for AMD to swoop in and start taking marketshare.