News Intel Details 10nm+ Xeon Ice Lake-SP with Sunny Cove Cores at Hot Chips 2020

[Admin]

Intel divulged the architectural details of its 28-core Ice Lake-SP Xeon variants at Hot Chips 2020.

Intel Details 10nm+ Xeon Ice Lake-SP with Sunny Cove Cores at Hot Chips 2020 : Read more

 

[JayNor]

"We also know that AMD will bring its own massive enhancements paired with (at least) an industry-leading 64 cores to market with its AMD EPYC Milan processors"

really? are they adding avx512? dlboost? bfloat16? What do you know?

 

[Sleepy_Hollowed]

"We also know that AMD will bring its own massive enhancements paired with (at least) an industry-leading 64 cores to market with its AMD EPYC Milan processors"

really? are they adding avx512? dlboost? bfloat16? What do you know?

There's like 3 applications written for paralellization that use AVX512, so that's like literally useless outside of those. Linus Torvalds had a lot to say about that:

"I hope AVX512 dies a painful death, and that Intel starts fixing real problems instead of trying to create magic instructions to then create benchmarks that they can look good on…
I absolutely destest FP benchmarks, and I realize other people care deeply. I just think AVX512 is exactly the wrong thing to do. It’s a pet peeve of mine. It’s a prime example of something Intel has done wrong, partly by just increasing the fragmentation of the market."

 

[Conahl]

avx 512 ? * yawn * keep it in the Xeon lines where i might be used more, rip it out of the consumer chips.

 

[PCWarrior]

There's like 3 applications written for paralellization that use AVX512, so that's like literally useless outside of those. Linus Torvalds had a lot to say about that:

Intel is including such functionality because clearly there has been demand for it. Here there are companies that even go ahead to develop custom cpus just to do high-volume math for AI etc. And Linus was speaking about consumer-grade cpus, not server-grade ones. But even for consumer-grade cpus his comments about AVX-512 were just wrong. He seems to be either stack in the past or he simply only cares about what he is personally involved with. Let’s not forget that he is primarily a kernel developer (and not even that engaged anymore) and kernel stuff has virtually no use for high-volume math. Linus is not using GPU acceleration either and he is generally not invested in any HPC-related stuff. Which is fine but projecting your own needs to everyone’s is not just narrow-minded but preposterous.