Intel Xeon Scalable Sapphire Rapids Leaps Forward

erek

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
11,156
Interesting as always!

"Intel knows what AMD launched with Genoa. Some of the list pricing of Sapphire Rapids looks almost like it was designed to discount. The market will sort that out. While Intel does not have a direct socket-to-socket top-bin competitor to AMD, what it does have is a range of products under 200W TDP, almost as many 32-core SKUs as AMD has in its entire EPYC 9004 SKU stack, and scale. These lower power and core count SKUs move volume, ensuring that Intel has volume for its Sapphire Rapids parts and for its server OEMs."

ntel-Vision-2022-Sapphire-Rapids-HBM-Top-1-800x623.jpg


h-Generation-Intel-Xeon-Scalable-Sapphire-Rapids-2.jpg


Source: https://www.servethehome.com/4th-gen-intel-xeon-scalable-sapphire-rapids-leaps-forward/

https://www.phoronix.com/review/intel-xeon-platinum-8490h
 
Last edited:
Some additional interpretations of these readings:

"The new Intel Sapphire Rapids processors represent a significant upgrade to Xeon, with 50% more cores than the previous generation, along with an increase in PCIe lanes, support for PCIe Gen5, CXL 1.1, DDR5, and onboard acceleration capabilities. These upgrades make the new processors game-changing for many organizations.

It's worth noting that Intel does not have a direct top-bin competitor to AMD's Genoa processors, but it does offer a range of products under 200W TDP, almost as many 32-core SKUs as AMD has in its entire EPYC 9004 SKU stack, and scale, which will move volume, this guarantees Intel volume for its Sapphire Rapids parts and for its server OEMs.

The launch of Sapphire Rapids also includes some high-priced SKUs, such as the $17,000 Platinum 8490H, which are designed for massive scale-up systems where the TCO is measured in hundreds of thousands of dollars per system and list pricing is often heavily discounted at a system level.

Moreover, Intel has two different performance stories; one with accelerators and one without. It can be competitive in many segments without them, but with the help of them, Intel has the ability to get outsized performance per core gains.

The Sapphire Rapids launch can be seen as a direct impact of competition in the market, as it matches AMD's 50% generational core count improvement for 2023 servers. With more CPUs expected to be launched, it remains to be seen how the market will evolve. But the most important factor for server buyers with this launch is that the market is now more competitive and dynamic, which will benefit buyers.

The Intel Xeon Platinum 8940H processor showed strong performance in AI benchmarks such as oneDNN, DeepSparse, and OpenVINO, as it was able to fully leverage the Sapphire Rapids capabilities with AMX. Other 4th Gen Xeon processors also demonstrated good performance in certain workloads such as Open Image Denoise for image denoising and OSPRay, GraphicsMagick, Python, PHP, OpenJDK Java, etc. However, 4th Gen Xeon Scalable performance struggled in HPC/server workloads that scale well to high thread counts, where the 96-core / 192 thread AMD EPYC 9654 could benefit, even in 2P configurations up to 384 threads. Some of the memory-intensive workloads also performed better thanks to the AMD EPYC "Genoa" supporting 12 channels of DDR5 memory. But in memory-intensive workloads it's also where the Intel Xeon CPU Max Series with HBM2e should perform very well too. It is worth mentioning that no hands-on access to the Sapphire Rapids HBM2e SKUs yet for benchmarking them.

It will take some time for the software ecosystem to catch-up and fully embrace the new accelerators in the Sapphire Rapids processors. But the Linux kernel driver support is in place for Linux 5.19+ and now it's up to Linux user-space software to catch up. Availability of Sapphire Rapids in public clouds will soon allow more developers to try out the 4th Gen Xeon Scalable and ideally begin adapting their software for leveraging the accelerators. However, It should be noticed that for the lower-end Xeon Sapphire Rapids SKUs, the accelerators are limited or outright disabled unless engaging the Intel On Demand licensing model.

The pricing of 4th Gen Xeon Scalable, particularly the $17,000 Xeon Platinum 8490H, is steep unless the workloads can make optimal use of the accelerators and new Sapphire Rapids capabilities. The AMD EPYC 9654 meanwhile has a list price of $11,805 which, even when factoring in the costs of going for 12 DIMMs per socket rather than 8 DIMMs to populate all available memory channels, will still come out ahead of the 8940H pricing. On the other hand, the pricing for Intel's HPC-optimized Xeon CPU Max Series is better, if not planning to make use of the accelerators aside from DSA. The Xeon CPU Max Series 9480, with 56 cores and the same base/turbo frequency as the 8490H, same 112.5MB of cache, same 350 Watt TDP rating, eight channels of DDR5-4800, and 4 DSA devices but no QAT/DLB/IAA accelerators, is priced at $12,980 USD, and is closer to the AMD EPYC 9654, while the 64GB of HBM2e should be interesting for HPC benchmarks.

In summary, this is Intel's most exciting server lineup in many years, and it has many new significant features that will only see more positive returns as software is adapted to make use of these new capabilities, in addition to the competition push that Sapphire Rapids is pushing forward with other vendor."
 
The server parts are nice and all, but what about Sapphire Rapids HEDT workstations with gobs of PCIe lanes to slot in extra add-in cards that the current mainstream DT platforms can't offer?

I got wind of some new systems being built up at my workplace on the now-dated X299 platform specifically because it has the PCIe lanes to handle several capture cards soaking up about 16K worth of pixels altogether in real-time, and Threadripper is now far from cost-effective to build up.

I've been itching for Intel to get the ball rolling on Sapphire Rapids precisely so that AMD doesn't continue to neglect Threadripper in favor of EPYC, but it certainly looks like that's not happening. I suppose when your chip production is maxed out and you're selling absolutely everything, might as well make most of that the most profitable, highest-end product lines you can.

EDIT: Oh, that's apparently next month. W-2400 and W-3400 CPUs, along with the W790 chipset, are what I want to keep my eyes on.
https://www.hwcooling.net/en/return-of-intels-hedt-w790-xeon-w-2400-and-w-3400-processors/
 
Last edited:
erek said:

NVIDIA Pairs 4th Gen Intel Xeon Scalable Processors with H100 GPUs​

Wonder how this competes with the Mi300 ?
Click to expand...
I can’t get a read on the Mi300, it appears to be a 900 watt monster that supposedly delivers 8x the “AI” performance of the 250x.

The problem is they aren’t really defining what calculations they are basing that on. The 250x doesn’t support 4 or 8 bit calculations. Most of the new AI models rely on 8bit precision or 16bit half precision. Using those calculations as an example the H100 is between 9 and 30x faster than the 250x, if you are using the Nvidia Transformer Engine.

The biggest hurdle for AMD is their ROCm platform, it is a solid decade behind CUDA.
OpenAI has moved itself over to the Triton open source libraries which could be made to work for the Mi300 of course but that is still a very small use case overall.
 
NamelessPFG said:
The server parts are nice and all, but what about Sapphire Rapids HEDT workstations with gobs of PCIe lanes to slot in extra add-in cards that the current mainstream DT platforms can't offer?

I got wind of some new systems being built up at my workplace on the now-dated X299 platform specifically because it has the PCIe lanes to handle several capture cards soaking up about 16K worth of pixels altogether in real-time, and Threadripper is now far from cost-effective to build up.

I've been itching for Intel to get the ball rolling on Sapphire Rapids precisely so that AMD doesn't continue to neglect Threadripper in favor of EPYC, but it certainly looks like that's not happening. I suppose when your chip production is maxed out and you're selling absolutely everything, might as well make most of that the most profitable, highest-end product lines you can.

EDIT: Oh, that's apparently next month. W-2400 and W-3400 CPUs, along with the W790 chipset, are what I want to keep my eyes on.
https://www.hwcooling.net/en/return-of-intels-hedt-w790-xeon-w-2400-and-w-3400-processors/
Click to expand...
AMD essentially abandoned the Threadripper lineup and they keep it there specifically for us who need PCIe lanes or the memory channels. They know we are stuck needing it and they are charging us for it.

I’ve gone back to Intel and the Ice Lake stuff to fill that need, it costs the same but there’s better support and less vendor lock-in.
 
You must log in or register to reply here.
Back
Top