NVIDIA GH200 CPU Performance Benchmarks Against EPYC Zen 4 & Xeon Emerald Rapids

[sophisticles]

I'm pretty confident that the above statements are completely and ridiculously false.

You base this on what exactly?

[qarium]

What benchmarks are you looking at?

it is not important what benchmark exactly.

but most people completely do it wrong they do benchmarks on field where ASIC is available and faster...
many blamed apple's M1/M2/M3 for beeing slow CPU in tests who these SOCs have ASIC... thats nonsense.
apple does not optimise their cpus for stuff they have ASIC for.
to make a valid CPU test you first have to check: does it run faster on GPU ? does it run faster or with much less power consumtion on ASIC ?
only if this is all not a option then you have a valid CPU test.

there are ZERO people who buy a epyc or threadripper and then do not install a Intel or nvidia or amd gpu ...

[coder]

And as i mentioned i used to do consulting work for a music video production company out of Florida, so am I an expert?

Depends on what sort of consulting. Did you spend years exploring the ins and outs of different codecs and compression software, mastering hundreds or thousands of videos for commercial distribution? If so, I'd say you might be an expert. If you have an advanced degree in video compression and made meaningful contributions to popular implementations or published novel research, I'd say you might be an expert.

If you just maintained their PCs and ran through the steps of encoding a few videos, I think most people wouldn't consider that an expert.

Even so, I've know people who've spent the fabled 10,000 hours performing a skill and I still wouldn't consider them experts. A lot of it comes down to mindset, and you've exhibited such overconfidence I'm not certain you could ever truly be an expert at anything.

You will note that the tests were not done with live action footage, which is industry standard way of testing encoder quality, they were done with video game captures.

Because you don't understand either the point of those GPU-accelerated codecs or the point of the article. Those GPU codec implementations are intended for fast & realtime single-pass encodes, not mastering commercial content. That's why he was using them for game streaming, which is the use case the article was aimed at.

You will also note that the speed comparison was of a top end Raptor Lake vs an older NVIDIA card and the author completely ignored that consumer NVIDIA cars are capable of encoding two streams at full speed simultaneously.

That only helps if you've got two streams you want to encode, simultaneously. If you're just encoding a single stream, then it doesn't matter that you could do a second one.

[Times Two]

You base this on what exactly?

The poster is implying that the gh200 grinds to dust 3-5 mi300a/mi300x. The AMD offerings
have comparable compute capabilities to Nvidia's, so he's lying through his teeth. The lie doesn't
get much better by anchoring the statements to the gh200 memory.

[GPTshop.ai]

I'm pretty confident that the above statements are completely and ridiculously false.

Prove it!!! If you can not prove it, you are just trolling.

[Times Two]

Prove it!!! If you can not prove it, you are just trolling.

It's you that was trolling. You put before us what are most certainly
farfetched claims as if they were facts, with zero backing.

The onus of proof is on you, not me.

[GPTshop.ai]

It's you that was trolling. You put before us what are most certainly
farfetched claims as if they were facts, with zero backing.

The onus of proof is on you, not me.

The claims are based on publicly available information from AMD https://www.amd.com/en/products/acce...00/mi300x.html https://www.amd.com/en/products/acce...00/mi300a.html https://www.amd.com/en/products/prof...deon-pro-w7900 and our benchmarks. Which will be released here on Phoronix soon.

[sophisticles]

The claims are based on publicly available information from AMD https://www.amd.com/en/products/acce...00/mi300x.html https://www.amd.com/en/products/acce...00/mi300a.html https://www.amd.com/en/products/prof...deon-pro-w7900 and our benchmarks. Which will be released here on Phoronix soon.

Welcome to Phoronix.

By now you must have realized that there is at least one poster, who i suspect posts under various names, that is an absolute AMD fanatic while hating NVIDIA and Intel; any benchmarks that do not show AMD beating the two by significant margins will be dismissed as flawed, biased or inconsequential because there are better methods of performing the task.

This will be their stance even when clear proof is provided that the above is not the case.

In the interest of fairness, from the links you provided, using the easiest metric for comparison:

AMD Instinct MI300A - Peak Double Precision FP64 performance 61.3 TFLOPs

NVIDIA H100 GPU - FP64 34 TFLOPS

Five AMD Instinct MI300A would be 306.5 TFLOPS

The Grace CPU is capable of 7.1 TFLOPS according to NVIDIA, so that's a total of 41.1 TFLOPS for the NVIDIA superchip.

Now this is only for FLOPS and it could be that in actual workloads the massive bandwidth the chip has allows it to match five MI300A.

[sophisticles]

If you just maintained their PCs and ran through the steps of encoding a few videos, I think most people wouldn't consider that an expert.

That only helps if you've got two streams you want to encode, simultaneously. If you're just encoding a single stream, then it doesn't matter that you could do a second one.

I did a lot more than that, but truth be told there is so much to know about video that I do not consider myself an expert.

I do consider myself way more knowledgeable than most people and knowledgeable enough that i was hired when the COO of the company was Googling for a solution and ran across my posts on videohelp and after 10 pages of me arguing with people and posting test encodes, he concluded that I was the man to help them and reached out to me via PM to offer me a job.

I thought it was a scam by someone trying to unmask me but the money offer was good, so I set up an LLC and opened an account with a bank i was not affiliated with and decided to give it a shot.

Turned out they were on the level.

The ability to encode multiple stream simultaneously is important because few users only have one encode to do, they usually have multiple videos to encode.

[coder]

i was hired when the COO of the company was Googling for a solution and ran across my posts on videohelp and after 10 pages of me arguing with people and posting test encodes, he concluded that I was the man to help them and reached out to me via PM to offer me a job.

Ah, okay. So that helps explain why you're trolling on here. You think lightning might strike twice and you get a second job out of it? If so, I think there are much more reliable and time-efficient ways to find work.

BTW, convincing a non-technical person of your technical prowess is about as hard as impressing a 5-year-old. Some people just buy into the Sheldon stereotype. Or, at least they did in the past... these days, knowledge of video production and most other technical matters is a lot more commonplace.

P.S. I happen to know who sponsored your Phoronix Premium membership and they quickly concluded it was a mistake.

The ability to encode multiple stream simultaneously is important because few users only have one encode to do, they usually have multiple videos to encode.

All things being equal, sure. However, if you had reason to prefer CPU encodes, you can easy buy a machine more powerful than a Raptor Lake. Plus, CPU encodes don't use a fixed amount of CPU power, so you also have flexibility over whether to run fewer encodes faster or more encodes concurrently.