According to the leaked text now removed (taken by THG before dropping), the AMD Epyc 7742 is a 64-core CPU with 128 threads, 256 L3 cache, TDP 225W, and TDP 225W, and 2.25 base clocks / clocks GHz and 3.4GHz, respectively. The released Epyc 7601 is 32C / 64T, 180W TDP CPU, with 64MB of L3, and 2Base clock boost of .2GHz / 3.4GHz which is almost identical. Xeon Platinum 8280 is 28C / 56T, 2.7GHz base, 4GHz boost, and TDP 205W, while Xeon Gold 6138 (also included as a reference) is 20C / 40T, 2GHz / 3.7GHz, and 125W TDP.
If this rumor is accurate, AMD has managed to double the number of cores and slightly increase the clock in the TDP envelope 1, 25 times higher. I'm not sure what "RDY1001C" refers to at the bottom of the results, even though this setting is always the fastest on the list. At Google, the term turned out not to produce results.
There's more THG testing than we've done here; See your article for complete results. And, as always, treat all results with a bucket. This is a filtered result. Even if they are accurate, they can reflect engineering samples that do not represent final performance.
SVT is a video encoder that is highly optimized for Intel CPUs, but optimization for Intel chips often works well for AMD CPUs as well, and we certainly see it here. Neither encoder can scale very well by adding more cores, so we won't try to understand dual numbers. 7742 single is significantly faster than Xeon Platinum 8280 and 7742 is more than twice as fast as 7601.
At HEVC, performance numbers change. Here, Intel and AMD as a whole are equivalent, but 7742 is a big improvement over the Epyc 7601.
POV-Ray 3.7 scales with increasing number of threads, but 1x CPU gain to 2x CPU is much less than 7742 compared to 7601. AMD only requires 24 percent more performance when adding 64 cores, compared at 42 percent scale for the Xeon Platinum 8280. Difference in scaling this means that a dual Xeon 8280 pair almost matched an Epyc 7742 pair, despite the fact that an Epyc 7742 is significantly faster than a Xeon Platinum 8280.
Blender, and more general rendering, are tests that are generally controlled by AMD CPUs. AMD convincingly won this test, although oddly, we also see signs of a significant increase in scale for Intel CPUs. This could only reflect the fact that Intel CPUs have far fewer cores. The Xeon Platinum 8280 is just a 28-core chip compared to the performance of the 64-core chip. That is a great advantage for AMD. Of course, there are questions about price and position, too: Intel generally gives Xeon's price well above the AMD Epyc CPU, and we tend to prioritize price comparison over other factors.
However, readers should be aware that we might see a scaling issue on AMD CPUs due to the large number of cores: 128C / 256T, while the Xeon Platinum CPU only featured 56 cores in a 2S configuration. The application itself may not scale well on this type of thread count.
If these figures are accurate, they suggest that AMD 7nm Epyc will be a major challenge for Intel in a broader range of markets, which is exactly what we expected based on previous third-generation Ryzen and AMD statements about Epyc 2. Please note It features Bob Swan's recognition of the surge in competitive markets, and we have a stage to prepare where Intel will cut Xeon's price, either directly or by launching Cooper Lake (currently expected in the first half of 2020). Historically, the price of Intel's CPUs has been much higher than AMD's, but it's hard to know exactly how much more, because the company's price list (the best indicator we have to pass) doesn't reflect the volume that it actually does. customers pay.
If Rome's AMD is as good as it sounds, we should see an increase in OEM parts adoption compared to the first-generation Epyc, as well as some reaction from Intel. A client server needs several generations of products to move to a new provider, but they end up paying attention.