The first performance tests of a processor are coming Intel Sapphire Rapids-SPso we are talking about an engineering sample with 56 cores and 112 processing threads under the Golden Cove architecture with Intel 7 lithography which ran at a base/turbo frequency of 1.90/3.70GHz (3.30 GHz for all cores).
This processor was tested in a Dual Socket configuration, so we know the performance of the 2x Xeon combination platinum 8476 or Xeon platinum 8480 with 64 GB of DDR5 @ 4800 MHz memory and has been benchmarked against the performance of two AMD EPYC 7773X processors (MILAN-X with 3D cache) with 64 cores and 128 threads each CPU at 2.20/3.52 MHz Base/Turbo frequency. In this way they were compared 112 hearts and 224 threads from Intel vs. 128 hearts and 256 AMD wires.
The quick summary is that in single-core performance the two CPUs are very close, and that’s that in CPU-Z V17, V19, and V19 AVX512 we’re talking 507.6/157.6/725.6 points in front of the 482.7/167.9/561.8 points AMD EPYC processor. In Cinebench R15, R20 and R23, we are talking about 109/439/1105 dots in front of 189/446/1143 stitches.
If we talk about multicore performance, here the differences are already quite remarkable, and it is that just taking Cinebench R15, R20 and R23 we are already talking about performance of 8830 against 12308 points (R15); 27571 against 36076 points (R20) and 68200 against 96383 points (R23).
It is obvious that we are facing an engineering sample and that this performance must increase, but that it increases enough to be able to withstand Milan-X is a miracle, and even more so if we take into account that by launch dates its real rival will be the AMD EPYC GENOA, which make the leap to Zen4 with TSMC’s 5nm lithography. So it seems that everything will have to come down to this Intel offers a pretty good price in order to balance the scales in the absence of being able to offer a higher number of cores which solves the decrease in multi-core performance.
Going through: @yuuki_ans