cageymaru

Fully [H]
Joined
Apr 10, 2003
Messages
22,060
AMD has released a new live demo of its prototype 2nd generation AMD EPYC "Rome" 64 core 7nm CPU vs 2x Intel Xeon Platinum 8180M CPUs where the workload is C-Ray. C-Ray is a floating point intensive benchmark that renders images. The systems in the demo rendered a total of 3 images; a 1080p (HD), 1440p (QHD), and finally a 2160p (4K) image. The C-Ray version used in both systems during the demo is 1.2.0. The 1P AMD system was clearly faster than the 2P Intel system in this demo.

The AMD Rome Development Platform consisted of CPU: 1 x 64 core "Rome" SOC, Memory: 8 x 32GB DDR4 2667 DIMMs, Drive: 1TB SSD Samsung EVO 970, and OS version - Ubuntu 18.04.

The Intel Platform consisted of Chassis: Supermicro Model: SYS-1029U-TRTP, CPU: 2x Intel(R) Xeon(R) Platinum 8180M CPU @2.50GHz, Speed: 2500MHz, Total Cores/Threads: 56/112, Memory: 24xDIMM - 32BG = 768GB, Speed 2666 MHz, Manufacturer: Samsung, Drive: Samsung 970 EVO 1TB - NVME, OS version - Ubuntu 18.04

One 2nd Generation AMD EPYC 64 core CPU beats two Intel Xeon Platinum 8180M CPUs on the C-Ray benchmark.
 
There is a bit to think about here, such as what is the clock speed of the Rome CPU ? The Intel CPU's have a total of 56 cores and 112 threads. The AMD system has 64 cores, but how many threads ? Having 8 cores more you would expect the Rome CPU to perform better if they have similar IPC's. This benchmark doesn't really say that much.
 
There is a bit to think about here, such as what is the clock speed of the Rome CPU ? The Intel CPU's have a total of 56 cores and 112 threads. The AMD system has 64 cores, but how many threads ? Having 8 cores more you would expect the Rome CPU to perform better if they have similar IPC's. This benchmark doesn't really say that much.
64x2 would be 128
 
There is a bit to think about here, such as what is the clock speed of the Rome CPU ? The Intel CPU's have a total of 56 cores and 112 threads. The AMD system has 64 cores, but how many threads ? Having 8 cores more you would expect the Rome CPU to perform better if they have similar IPC's. This benchmark doesn't really say that much.

It says that if AMD used a 2 socket system it would absolutely obliterate Intel.
 
I think it looks great for AMD. Of course it won't be competing against the intel cpus they've shown, it'll be competing against the 48 core cascade ap. AMD looks like they'll be faster in some if not most workloads. That's pretty impressive.
 
There is a bit to think about here, such as what is the clock speed of the Rome CPU ? The Intel CPU's have a total of 56 cores and 112 threads. The AMD system has 64 cores, but how many threads ? Having 8 cores more you would expect the Rome CPU to perform better if they have similar IPC's. This benchmark doesn't really say that much.
It says Intel is getting their ass handed to them.
 
There is a bit to think about here, such as what is the clock speed of the Rome CPU ? The Intel CPU's have a total of 56 cores and 112 threads. The AMD system has 64 cores, but how many threads ? Having 8 cores more you would expect the Rome CPU to perform better if they have similar IPC's. This benchmark doesn't really say that much.
It's odd the lack of detail of the specifications for the AMD system. You would think that if you were comparing two systems you would show detail of each system on a comparable level, but that could also be to protect trade secrets. No matter, AMD could have cherry a picked bench as well as a cherry picked CPU then highly optimized for the bench.

Also they are sure to specify any aspect of the Intel system that outweigh that of the Rome, obviously to get the fans fired up, with lack of detail on their side. Marketing videos/demos are just that, marketing. Let's wait and see!
 
But Intel did have a record year with over 70 billion with a 3 year record revenue
They missed their numbers so either:
1) chip sector related
2) AMD gain

My best is BOTH. AMD guided VERY conservative, so hopefully they will be good.
 
This does look good for AMD, but... codec's matter, algorithms matter. I would LOVE to build our next SQL server refresh (a 32 core 64 thread system) on a 'Rome' platform and get that reliably out of one socket. I just don't see that quite yet.

Would love to see some performance comparisons with SQL DB's or Oracle DB's and Different I/O intensive workloads (where with all the PCIE lanes I would expect the ROME platform to shine.).
 
This does look good for AMD, but... codec's matter, algorithms matter. I would LOVE to build our next SQL server refresh (a 32 core 64 thread system) on a 'Rome' platform and get that reliably out of one socket. I just don't see that quite yet.

Would love to see some performance comparisons with SQL DB's or Oracle DB's and Different I/O intensive workloads (where with all the PCIE lanes I would expect the ROME platform to shine.).
Anandtech usually has db benchies,
 
I'd really like to see a benchmark between CPU Ray-Tracing and RTX 2080 Ti's Ray Tracing. I wanna know exactly how capable Nvidia's hardware is.
 
It says that AMD is able to squeeze more than double the number of cores than Intel into a single socket at comparable performance per core. That's huge in the server market.
intel already has 72 core 288 thread cpus (phi 7290). I have a few and they obliterate the intel platinums on some tasks.
 
intel already has 72 core 288 thread cpus (phi 7290). I have a few and they obliterate the intel platinums on some tasks.

Not every server app needs 72 cores. Dual socket and single socket Rome will undercut Intel's offering significantly on price / performance in many areas - which is really the point.
 
Intel Seconds per core: 0.54
AMD seconds per core: 0.44

So a bit less than 20% faster with about 14% more cores? So at least in this case AMD seems to have an IPC win...

So, AMD keeping the pressure on. Which is good for all of us.

Really hope new consumer CPUs close or surpass the IPC gap. With Intel passing off binned parts as "9th generation" I think they need more motivation. Like market and mind-share loss big time.
 
intel already has 72 core 288 thread cpus (phi 7290). I have a few and they obliterate the intel platinums on some tasks.

That is an interesting CPU. Interesting in this case meaning only good for very specific cases. It has zero support for virtualization at all. I didn't see mention of any sort of hyper threading (though I suppose 288 thread means 4 threads per core.) Unless you are running fairly specific ASIC like jobs I don't think this would be a very good platform.

Yes it has more raw cores and threads... to do a specific thing... slowly UNLESS it is heavily multi threaded.

I don't think it's even in the same ballpark as that of a ROME or Platinum Intel Xeon CPU for general use.

But you DID get me to look it up. So props for that! :)
 
AMD has released a new live demo of its prototype 2nd generation AMD EPYC "Rome" 64 core 7nm CPU vs 2x Intel Xeon Platinum 8180M CPUs where the workload is C-Ray. C-Ray is a floating point intensive benchmark that renders images. The systems in the demo rendered a total of 3 images; a 1080p (HD), 1440p (QHD), and finally a 2160p (4K) image. The C-Ray version used in both systems during the demo is 1.2.0. The 1P AMD system was clearly faster than the 2P Intel system in this demo.

The AMD Rome Development Platform consisted of CPU: 1 x 64 core "Rome" SOC, Memory: 8 x 32GB DDR4 2667 DIMMs, Drive: 1TB SSD Samsung EVO 970, and OS version - Ubuntu 18.04.

The Intel Platform consisted of Chassis: Supermicro Model: SYS-1029U-TRTP, CPU: 2x Intel(R) Xeon(R) Platinum 8180M CPU @2.50GHz, Speed: 2500MHz, Total Cores/Threads: 56/112, Memory: 24xDIMM - 32BG = 768GB, Speed 2666 MHz, Manufacturer: Samsung, Drive: Samsung 970 EVO 1TB - NVME, OS version - Ubuntu 18.04

One 2nd Generation AMD EPYC 64 core CPU beats two Intel Xeon Platinum 8180M CPUs on the C-Ray benchmark.
Damn, and with 256Gb vs 768Gb.. even if C-Ray won't use it all Intel has far, far more bandwidth available at the same memory speed with a few less cores. Interesting bench.
 
The devil in the details: on tha Rome runs on a total of 256GB DDR4 2667 Ram with 1TB SSD (SATA or NVMe) vs 768GB DDR4 2667 Mhz RAM with 1TB NVME SSD for the dual Xeon CPU.
 
intel already has 72 core 288 thread cpus (phi 7290). I have a few and they obliterate the intel platinums on some tasks.

That's 72 very, very slow cores, though. Not only are they clocked low, they are also Atom cores. Their IPC is terrible. The entire point of it is purely the SIMD floating point performance. That chip isn't competing against other Xeons or Epyc chips, it's competing against Nvidia's Tesla. It's basically a GPU in a CPU socket.

It has a use, but it's entirely irrelevant here. Only useful in a completely unrelated market segment.
 
It says that AMD is able to squeeze more than double the number of cores than Intel into a single socket at comparable performance per core. That's huge in the server market.

At comparable performance per core on C-Ray... And this is comparing Zen2 to Skylake-SP.
 
It's odd the lack of detail of the specifications for the AMD system. You would think that if you were comparing two systems you would show detail of each system on a comparable level, but that could also be to protect trade secrets. No matter, AMD could have cherry a picked bench as well as a cherry picked CPU then highly optimized for the bench.

Also they are sure to specify any aspect of the Intel system that outweigh that of the Rome, obviously to get the fans fired up, with lack of detail on their side. Marketing videos/demos are just that, marketing. Let's wait and see!

the only unofficial clocks for the 64/128 cpu i've seen was 1.4Ghz base 2.0Ghz max boost on an engineering sample that was posted on sisoftware.. we won't know til it's officially launched what the actual numbers end up being but it's probably some where around there.
 
And what about integer tests?

CPUs are primarily strong in integer calculations, not floating point. So while this is certainly a promising test, we need to see more.
 
That is an interesting CPU. Interesting in this case meaning only good for very specific cases. It has zero support for virtualization at all. I didn't see mention of any sort of hyper threading (though I suppose 288 thread means 4 threads per core.) Unless you are running fairly specific ASIC like jobs I don't think this would be a very good platform.

Yes it has more raw cores and threads... to do a specific thing... slowly UNLESS it is heavily multi threaded.

I don't think it's even in the same ballpark as that of a ROME or Platinum Intel Xeon CPU for general use.

But you DID get me to look it up. So props for that! :)

they also have crazy memory bandwidth due to the 16 gb of mcdram on the chip. the newer ones do have support for virtualization.
 
At comparable performance per core on C-Ray... And this is comparing Zen2 to Skylake-SP.

Sure, but is there anything to support the conclusion that they won't be competitive in other non-AVX-512 workloads? Epyc was already very competitive with Skylake-SP and that's before the architectural enhancements in Zen 2, not the least of which is full AVX2 support. Given that it seems Intel's only real response is going to be the niche product that is Skylake-AP, I'd say AMD is going to be the overall performance leader at least until Intel is able to push out a true competitor, certainly on a per-socket basis if not per-core.
 
  • Like
Reactions: N4CR
like this
There is a bit to think about here, such as what is the clock speed of the Rome CPU ? The Intel CPU's have a total of 56 cores and 112 threads. The AMD system has 64 cores, but how many threads ? Having 8 cores more you would expect the Rome CPU to perform better if they have similar IPC's. This benchmark doesn't really say that much.

It tells us tons. First, it's faster at going from start to finish o a given scene. All reasonable renderers these days are heavily multi-threaded, so this should translate to multiple rendering platofrms reasonably reliably. Second you don't care if it is faster as long as it isn't slower because....

The tweo xeons have a TDP of 205W each.

The AMD SoC has a TDP of 65W+ It's safe to say that the + means 250W or less given that has been the max thermal envelope for their previous entries and AMD still lists it for ryzen2 entires with lower core counts.

Right there It wins. Because it means for the amount of power and cooling in your render farm machine room, you can either use less and save money every month on utilities, or you can pack in more render nodes and get work done faster without paying for HVAC, power, and UPS expansion.

If you have a better understanding of the renderer it alsso tells you something about memory needs and efficiency.

I have no idea if the intel box is WAAAAY over specced to make AMD look good or if it needs that much ram. But if you can get your desired render times with 128 gigs of ram instead of 700+, those boxes will HAVE to be significantly cheaper. People with more knowledge of the software will know though and be able to tell.
 
Sauce?
Last I had not seen any confirmation. But been too busy to keep on top of things.. cheers

I'm just hoping that its fully implemented not the we will just combine two AVX 128 piplines ala how they are doing AVX 256 on Zen
 
Back
Top