A well known blogger has shared maybe the primary efficiency numbers of Intel’s upcoming Xeon Platinum Sapphire Rapids (opens in new tab) processors with 60 cores and HBM2E reminiscence. Apparently, neither a 60-core Intel CPU nor the Xeon CPU with on-package HBM2E reminiscence can beat AMD’s high-end EPYCs obtainable right this moment.
A few of Intel’s clients already get PRQ variations (opens in new tab) of choose 4th Technology Xeon Scalable Sapphire Rapids processors. Nonetheless, the launch of high-volume fashions has suffered a delay to late 2022, with a full-scale ramp now scheduled for 2023 as the corporate needed to appropriate sure bugs in silicon after which re-qualify these CPUs with companions. Intel retains specs of its upcoming merchandise below wraps. Since there are lots of Sapphire Rapids processors within the wild already, not solely do we all know a few of their alleged specs, however their efficiency numbers have surfaced forward of launch.
YuuKi_AnS (opens in new tab), a outstanding {hardware} leaker with an excellent observe report and entry to unreleased {hardware}, has laid his palms on Intel’s 60-core Xeon Platinum 8490H CPU and a 52-core Xeon Platinum 8472C processor, presumably geared up with 64GB of HBM2E cache. The {hardware} blogger not solely disclosed the specs of the stated merchandise but in addition printed their benchmark end result. All the data is confirmed by screenshots (see the gallery beneath), so every thing appears to be like considerably credible, whereas there could also be some bumps. In the meantime, since we’re coping with pre-production {hardware}, take the outcomes with a grain of salt.
Cores/Threads | Cache L2+L3 | Frequency | TDP – PBP – MTP | |
---|---|---|---|---|
Xeon Platinum 8472C* | 52/112 | 112MB + 97.5MB | 2.50 GHz – 3.80 GHz | 350W – 420W – 764W |
Xeon Platinum 8490H* | 60/120 | 120MB + 112.5MB | 1.90 GHz – 3.50 GHz | 350W – 420W – 764W |
*Specs are unconfirmed.
Final spring, Intel solely deliberate to launch Sapphire Rapids CPUs with as much as 56 cores, however now the corporate is testing 60-core Xeon Platinum 8490H processors operating at 1.90 GHz – 3.50 GHz. Moreover, the corporate appears to be like to be testing 52-core Xeon Platinum 8472C CPUs with on-package 64GB of HBM2E reminiscence and operating at significantly greater 2.50 GHz – 3.80 GHz.
CPU-Z Single-Thread | CPU-Z Multi-Thread | |
---|---|---|
Xeon Platinum 8472C | 542.5 | 31,954.3 |
Xeon Platinum 8490H | 508 | 32,747.6 |
As a consequence of greater frequency, Intel’s Xeon Platinum 8472C gives barely greater single-thread efficiency than the corporate’s Xeon Platinum 8490H CPU within the CPU-Z 2.02.0 benchmark. But the 60-core mannequin scores marginally greater in multi-threaded workloads.
Whereas core rely and frequencies have an effect on efficiency in just about all workloads, including HBM2E reminiscence (with a 1.22 TBps peak bandwidth) advantages choose reminiscence bandwidth-bound workloads, resembling computational fluid dynamics and rendering.
Threads and Frequency | V-Ray Rating (Ok samples) | |
---|---|---|
2x Xeon Platinum 8472C | 224 @ 2.50 – 3.80 GHz | 95,014 |
2x Xeon Platinum 8490H | 240 @ 1.90 – 3.50 GHz | 71,830 |
2x AMD EPYC 7773X | 256 @ 2.20 – 3.50 GHz | 102,843 |
2x AMD EPYC 7763 | 256 @ 2.45 – 3.50 GHz | 109,248 |
It’s exactly what we see in Chaos’s V-Ray 4.10.03 benchmark, which might benefit from each many cores and better per-core efficiency. 64GB of HBM2E reminiscence performs a extra vital function right here, which is why the Xeon Platinum 8472C is about 32% quicker when in comparison with its Xeon Platinum 8490H sibling.
Apparently, AMD’s EPYC 7773X (Milan-X with 3D V-Cache) doesn’t have any edge over the EPYC 7763 processor, which runs solely a tad quicker. Nonetheless, each already obtainable CPUs are presently untouchable by Intel’s choices which have but to return to market.
One factor that YuuKi_AnS identified is that new BIOS variations have a tendency so as to add efficiency to Intel’s Sapphire Rapids CPUs. So, assuming that there might be additional microcode optimizations of those processors, manufacturing servers and workstations based mostly on these CPUs will provide greater efficiency than present samples. However till these machines arrive, all now we have left is to investigate outcomes obtained on pre-production CPUs.