[hrtabf]

Yes, the CPUs listed are in the range where customers cannot perform expensive experiments (buying both Intel and AMD to test) and rely on review sites.

Personally, when I first got access to an Epyc, I was underwhelmed by the performance. For numerical performance it was slightly worse than M2 or cheap old Intel processors. I'm now a bit skeptical of reviews.

[grues-dinner]

I listened to a review that said an i7 Thinkpad was cool and quiet with an 8-10 hour battery life. Fans scream at the slightest load, it's 45 degrees constantly and the battery life is 3 hours if you don't touch it and 1 hour if you do. And digging deeper, that's just normal for them. Serves me right for trusting a "real" reviewer.

Should have insisted on a Framework, by the sounds of it, it would actually have at least not worse battery life.

[echoangle]

Can you say which Epyc processor? And in what benchmark it was worse than which old intel processor?

[ultrnal]

I think that the current thread sentiment (downvoting and accusations of lying elsewhere), does not make it appealing to provide further details.

[bayindirh]

No, creating a throwaway account and making a superfluous claim creates the thread sentiment.

There might be valid comparisons for specific Epyc generations and their Xeon counterparts. But processors are a bit like car engines. Power and torque numbers doesn't tell the whole story.

Datacenter processors are optimized for different scenarios. Use the wrong processor for the wrong job, and you get abysmal performance. We have an AMD system which won't win any speed records, but that thing has enormous number of memory channels and PCIe lanes, so it's basically a semi with an extra long trailer.

Fittingly, that processor lives in a storage cluster and delivers tremendous amount of I/O both in IOPS and throughput. Same processor would look silly in a compute cluster, though.

[shiroiushi]

>No, creating a throwaway account and making a superfluous claim creates the thread sentiment.

Maybe this is Intel's new marketing strategy. Astroturfing has been around for decades now; Microsoft was doing it in the early 2000s.

[Ragnarork]

Very convenient.

Or you could back up your claim with actual data.

[wordofx]

^ things that never happened.

[bayindirh]

I'd rather talk in SPEC scores if we're talking about Epyc and Xeon processors, and in TOP500 lists and national supercomputing centers, but you do you.

[ywvcbk]

> and in TOP500 lists and national supercomputing centers

Why would that be particularly relevant when comparing the performance of specific individual chips?