Comments on: AMD Now Has More Compute On The Top500 Than Nvidia https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Mon, 25 Nov 2024 16:22:06 +0000 hourly 1 https://wordpress.org/?v=6.7.1 By: emerth https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240316 Thu, 21 Nov 2024 13:24:55 +0000 https://www.nextplatform.com/?p=145000#comment-240316 In reply to Paul Berry.

The FP64 rate on things like A40, L40, L40s, RTX gaming cards – is abysmal already. Off the top of my head nV only does full rate FP64 on A30, A100, H?00. Not sure if there is an L30.

]]>
By: emerth https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240315 Thu, 21 Nov 2024 13:20:43 +0000 https://www.nextplatform.com/?p=145000#comment-240315 In reply to Timothy Prickett Morgan.

Get one for me, please. My VII is so lonely.

]]>
By: Calamity Jim https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240225 Wed, 20 Nov 2024 02:17:37 +0000 https://www.nextplatform.com/?p=145000#comment-240225 I reckon this here Top500 is interesting, and I especially like the #20 El Dorado (for its inspiring name; better than IronMan …) but, hold your horses a bit, what about that most arduous of red-headed stepchild competitions, the lower bookend of the librarian County Fair, the red-hot picante sauce of the exhibition, the HPCG, that one doesn’t seem to have received that much love this time around, as all the MI300s just skipped it outright.

Two retirees (Summit and Trinity) made room for 3 new entries in the top 25 (CEA-HE, Gefion, Miyabi-G) but not much else seems to have moved at first sight. It’s still Fugaku, Frontier, Aurora, LUMI, Alps, Leonardo … as it was in June (no El Cap, no Eagle, no HPC6, no Tuolumne, …). Here’s to hoping that Mr.DIMM shows up soon enough to shake things up some in this here rodeo show, especially for CPU only bullriders of the Crossroads, Shaheen III, and MareNostrum 5 GPP variety, where it should shine most brightly, and even for accelerated systems if it helps them at all. Imagine if Fugaku had met Mr.DIMM, it could have 32 PF/s its way through the HPCG barrell race and be virtually untouchable (I think)!

]]>
By: Slim Albert https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240220 Wed, 20 Nov 2024 00:40:21 +0000 https://www.nextplatform.com/?p=145000#comment-240220 In reply to HuMo.

Speaking of growth, let’s not forget the conjoined MI300A triplets that sure saw some improvements since this past June. “Grizzly Adams” saw a 24% HPL improvement without (it seems) additional hardware, jumping from 19.6 to 24.4 PF/s (from 61% to 76% of peak perf, now #49). The “American Cro-Magnon of Tuolumne” grew even more to 9x of its original core structure and 10x its previous performance, now at 208 PF/s and #10. And of course, the formerly #46 “early deliveroo” has blossomed into the swashbucklingly charming #1 we all know and love!

They grow up so fast don’t they (and each with their own eventual personnality)! (eh-eh-eh)

]]>
By: Akhanaton https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240213 Tue, 19 Nov 2024 20:40:06 +0000 https://www.nextplatform.com/?p=145000#comment-240213 At the end of the RFP process, there was a price below which nVidia was unwilling to go. Its the Keystone Problem, i.e. Quad-Keystone pricing is mandatory – cost is not the basis for the price.

]]>
By: HuMo https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240203 Tue, 19 Nov 2024 17:36:16 +0000 https://www.nextplatform.com/?p=145000#comment-240203 Great analysis! It’s good to see AMD’s EPYC quiet Zen Instinct meditating to the top of the rankings (and share) with the boisterous vigor and swashbuckling stronger swagger of the buccaneering El Capitan, no longer tip-toeing through the dancehall, but rambunctiously tap dancing over all manners of HPL, like a youthful corsair of HPC, at a blistering 1.7 EF/s!

Interesting also to observe the choreographic preparation of its likely next adversary, Jupiter, with a jazz ballet target of better than 1 EF/s, at 18 MW of effort (or 56 GF/W) … already met by the new champ. But the JETI, and the JEDI, suggest an even better telekinetic ability to effortlessly perform computational dance moves, as if floating through thin HPC air, with 68 to 73 of these GF/W! The future will be fascinating if such levitation can meet the challenges of scale, in a fullsize exaflopping ballroom blitz! Already, little ROMEO, and little Adastra 2, seem anxious to step into the dance shoes of their bigger chanllengers in this contest …

And its great seeing the Alps growing to nearly twice its former size, reaching new peaks of gut-warming performance in the Top500 waltz fondue! Nowhere else in the world could such hearty melding of HPC choreography and gastronomy be achieved, while surrounded by glaciers! 8^b

]]>
By: Paul Berry https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240152 Mon, 18 Nov 2024 23:41:42 +0000 https://www.nextplatform.com/?p=145000#comment-240152 If you look at the big inflection points in the list:
mid-90s the unix workstation market brought commodity(ish) 64-bit processors at much lower cost than vector mainframes. The early 00s brought the same, but they were even cheaper desktop-derived chips. The late 00s brought multi-core processors, which again provided more push for the same price. Then the mid 10s GPUs provided more ALUs per dollar if you can port back to vectors. At the moment, no inflection in site.

HPC isn’t a big enough market to drive massive innovation alone. It rides the coat tails of the broader IT market. At the moment all the IT investment is into low-precision matrix math engines. One wonders when Nvidia will put out a GPU that stops supporting 64bit math?

]]>
By: Timothy Prickett Morgan https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240151 Mon, 18 Nov 2024 23:33:03 +0000 https://www.nextplatform.com/?p=145000#comment-240151 In reply to Eric Olson.

OK. I will ask for a baker’s dozen.

]]>
By: Jure Pečar https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240146 Mon, 18 Nov 2024 21:35:35 +0000 https://www.nextplatform.com/?p=145000#comment-240146 In reply to Timothy Prickett Morgan.

Three 😀

]]>
By: Eric Olson https://www.nextplatform.com/2024/11/18/amd-now-has-more-compute-on-the-top500-than-nvidia/#comment-240145 Mon, 18 Nov 2024 21:19:02 +0000 https://www.nextplatform.com/?p=145000#comment-240145 Please ask for three.

]]>