AMD EPYC 9755 / 9575F / 9965 Benchmarks Show Dominating Performance

[lejeczek]

But feels bit weird, as if AMD dropped the ball on RAM speeds/performance - overconfident, underestimated its arch enemy...?

[mSparks]

It was an exciting 2 weeks for intel, but all that now comes to an end.

Even on desktop, they are advertising their latest and greatest as only 5% SLOWER than AMD's yesteryear...

An on enterprise, amd casually walks all over intel's greatest improvement in more than a decade.

Quite.
6980P Processor Sets Record $17,800 Flagship Price

Higher power consumption and less performance than
AMD EPYC Genoa 9684X, €4112.14 ex VAT

I actually feel sorry for Intel, these results are devastating.

[reavertm]

I don't think there's any question that Intel has the superior technology overall...


Sorry, wrong decade.

[schmidtbag]

Here I was thinking AMD would have some real competition. I guess Intel still has a lot of catching up to do.

[r1348]

That OpenSSL performance is otherworldly.

[sophisticles]

The main strength of the Xeon 6 Granite Rapids besides memory intensive workloads where MRDIMMs can help is for workloads able to leverage Advanced Matrix Extensions (AMX). With Intel's OpenVINO AI toolkit being a leading example for demonstrating AVX usage, the Xeon 6980P processors were leading with some AI models. At the extremes a single Xeon 6980P could outperform two EPYC Turin processors.

This right here is why I say that AMD makes scam processors, they appeal to the lowest common denominator, just brute force performance with raw core performance with no attempt at any finesse.

Clearly it's working, people are talking about them and they have gained market share, but the reality is that all CPU makers better get their act together and leave the American muscle car mentality behind and start following IBM's lead to make processors like that found in the z16 or what Qualcomm's NPU offers:

https://www.tomshardware.com/pc-components/cpus/arm-powered-snapdragon-x-elite-laptop-shown-outperforming-intel-core-ultra-by-up-to-10x-in-ai-tests-qualcomm-fires-early-npu-shots-at-intel

[HEL88]

Strongly AMD jumped.

[coder]

Tonight, sophisticles cried himself to sleep. He'd poured his life's savings into Intel stock, after his crack statistical analysis of its share price convinced him it was a sure bet.

It wasn't so much the sense of existential dread he experienced upon seeing this drubbing (AMD's 18.4% better performance at just 72.9% of Intel's price) and the entirely underwhelming Arrow Lake announcement, but rather the unshakable pangs of doubt that they just might be right and perhaps he is indeed a pretentious clown; not the genius day trader he fancied himself as.

🤡

[coder]

Well, intel still does better in some rather edge use cases,

Mostly just those few places they can deploy their low-precision AMX or derive an outsized benefit from MRDIMMs.

the most valid reason to keep using intel is running contracts and production capacity starvation of AMD.

But given the actual CPU design merit, perhaps intel should be prohibited from bidding for TSMC capacity for its "runt of the litter" lineup...

Huh? Granite Rapids and Sierra Forest are manufactured entirely by Intel fabs. You must have this confused with their latest client processors, which are indeed using mostly TSMC.

[coder]

But feels bit weird, as if AMD dropped the ball on RAM speeds/performance - overconfident, underestimated its arch enemy...?

They indeed have been working on MR-DIMM support, as well.

https://www.techpowerup.com/306762/amd-and-jedec-create-ddr5-mrdimms-with-17-600-mt-s-speeds

They just got leapfrogged by Intel. On the plus side, Turin drops into the same socket as Genoa. So, partners and customers have a tried and true platform. I'm sure their next EPYC will support MR-DIMMs.

BTW, I think it was rumored that Intel was working on a nonstandard MCR-DIMM-based solution. I wonder if that was ever true or maybe just an intentionally leaked red herring.