AMD had a one rack entry of their El Capitan nodes in the top500 list … 19.65 petaFLOPs.
“one computing rack each from El Capitan’s “Early Delivery System” (EDS), LLNL’s newest unclassified supercomputer RZAdams and its unclassified “sister” system Tuolumne. All three registered 19.65 petaFLOPs (nearly 20 quadrillion floating point calculations per second) on the High Performance Linpack (HPL) benchmark used by the Top500 organization to determine the world’s fastest supercomputers. “
]]>>>And they are still buying apparently
that claim doesn’t seem to be valid either based on past four quarterly reports. AI scaling is a nasty problem. The narrative that it is a chip problem and hence we can beat Nvidia with a bit more memory or some other clever trick kind of propaganda is the root of all these problems.
]]>There is only one reason AMD isn’t participating in industry benchmarks, and no taunting will help: the results don’t show well. Some startups have complained about the costs to participate in MLPerf. Cerebras concluded it wasn’t worth the resources which one can understand for a cash strapped startup. AMD at this point can’t make a similar claim, they’ve got plenty of resources, and there is no way they are not running these benchmarks internally. It’s also odd because their customers (and potential customers) also run that data on AMD chips as part of their evaluations. And they are still buying apparently. So effectively the only perception AMD are managing by holding back on MLPerf is the public’s. But the public by and large, aren’t customers for these type of chips.
]]>I don’t think Halo will have the CPU and GPU talking to the same RAM using the very same pointers.
]]>Isn’t that what Strix Halo is going to be? An APU with 16 Zen5 cores with 40 CU of RDNA 3.5 giving an iGPU that’s about a 4070 equivalent with say 128GB fast DDR5 unified shared memory?
]]>I agree as well! And Ricardo, it is so good to see you here in the comments. We appreciate you, and I can tell you that now.
]]>I agree. I want a PC made of this.
]]>I will taunt them a bit….
]]>Isn’t that question being a thorn all along on AMD’s side? Where are the benchmarks? Why not run MLPerf and publish what you see. I mean they seem to be proud of what they are doing. So, why hide the results?
]]>