Intel Xeon 6980P "Granite Rapids" Linux Benchmarks

[ddriver]

Good to see intel appearing somewhat competitive finally, let's see if that lasts tho.

The usual cadence is to test the mainstream parts on plebs before releasing enterprise products.

Before everyone rushes to buy stonks, wait to see if this rushed product doesn't blow up in their face. Not just metaphorically, but also literally, given the TDP figures...

[blackshard]

Uhhh, is it really true? Intel releases a server CPU that isn't worse than AMDs last generation and it might even be on par in power efficiency. I can't wait for the power numbers ...

From what we know of Zen 5 this might be the closest in competition we have seen in years. And in some cases the Intel CPU is twice as fast as the Zen 4 one, I don't believe that Zen 5 will give such massive boosts but it might still be slightly faster overall.

256T vs 256T and roughly similar performance and efficiency. Hopefully I don't wake up in a few seconds ...
Makes me a little more confident in the leaks about desktop 200 series.

Also can't wait for the Zen 5 comparison, thanks Michael.

Mmmh, about power efficiency, I don't know... the 2P 6980P processors have a 1kw TDP summed up
Zen 5 provides quite an interesting bump in terms of raw performance. I guess a lot of the juice here comes from the MRDIMMs and their bandwidth, but let's wait and see...

Interestingly enough, after disappearing from the upcoming desktop core 200 series, hyperthreading is back here Something that should be asked to all those Intel-loving people that was continously repeating like a mantra SMT was bad

[Weasel]

This is a tech forum, not investing. Please stop pumping (or trashing) stocks on here.

His comment was actually useful especially to newbies at investing. As in, he's mostly right in general way.

In general you buy stocks when they're low and hit hard by people panicking to sell, not stocks that are sky high. Buy Low, Sell High.

Or Buy Low and hold. But you still want to buy low so you get more shares (and thus dividends). Simple logic.

[coder]

AMX is the new AVX, any software running on the dedicated extension will run circles around the cpu that hasn't got it yet...

It's also far less general than any of the AVX-series extensions. Right now, it supports only int8 and bf16 data formats, which basically restricts it for use only on AI. There are better options for doing AI, however, like Intel's own upcoming Gaudi 3.

I think AMX was mostly done as a hedge against Intel's other AI bets not paying off, but it consumes a real chunk of die space. Basically, you can take Linus' original comments about AVX-512 and multiply them by 10, for AMX, because it wastes even more die area on a feature that's even more specialized to specific workloads and creates even more burden for the kernel.

[coder]

The usual cadence is to test the mainstream parts on plebs before releasing enterprise products.

They did, more or less. This uses the same P-cores that shipped in Meteor Lake, at the end of 2023. The process node is slightly different, however.

Interestingly enough, after disappearing from the upcoming desktop core 200 series, hyperthreading is back here

Your chronology is backwards. These use an older generation of P-cores than what's in Lunar Lake & Arrow Lake.

Something that should be asked to all those Intel-loving people that was continously repeating like a mantra SMT was bad

Intel actually said they're retaining hyper-threading in their server CPUs. So, there's no contradiction or backpedaling vs. their messaging on the subject.

[sophisticles]

It's being compared against their previous generation. Intel on a 3nm-class node vs. AMD on a 5nm-class node. It's no wonder that it beats EPYCs with fewer or smaller cores.

Good job on Intel for getting this out the door before AMD could launch Turin. Once that happens, it'll be a return to the usual picture where AMD is back on top.

The issue is that there is a practical limit as to how small you can go.

This means eventually both AMD and Intel will be on the same node, meaning that advantage will not exist any longer.

You are also assuming that AMD will be able to execute 3nm without any issues, and the reality is any company can stumble, even one that has been firing on all cylinders that past few years.

[coder]

His comment was actually useful especially to newbies at investing. As in, he's mostly right in general way.

That's not the point. If you want to talk investing, there are many other (and better) sites for that.

[coder]

The issue is that there is a practical limit as to how small you can go.

This means eventually both AMD and Intel will be on the same node, meaning that advantage will not exist any longer.

Enough silliness. The technology roadmaps from the fabs (Intel included) and IMEC show a pathway for continued improvement well into the next decade. We've seen Intel and Samsung falter numerous times, over the previous decade, so it's far from a foregone conclusion that everyone will end up on basically identical nodes any time soon.

You are also assuming that AMD will be able to execute 3nm without any issues, and the reality is any company can stumble, even one that has been firing on all cylinders that past few years.

AMD already shipped the very same Zen 5 CCDs in their Ryzen 9000 desktop CPUs that they're using in Turin. So, those are locked and loaded.

The 3 nm Zen 5C CCDs are the ones we have yet to see, but those will go in the 196-core EPYCs, not the 128-core models.

[sophisticles]

This is a tech forum, not investing. Please stop pumping (or trashing) stocks on here.

People on tech forums do invest and more importantly a companies financial health dictates how well it is able to bring new tech to the market.

This is not pumping Intel stock, this is some advice based on a detailed market and stock price analysis and it is also a prediction of the performance of future Intel products,

Intel's execution in the market with new products will dictate what it's sharers trade at, so when I predict that it will hit 27 shortly, implicit in that prediction is the claim that the products due to be released by end of 2024 will be very competitive.

This in turn leads to the 35 dollar prediction; the 55 and 100 dollar predictions are forward looking statements assessing how Intel's future products are expected to perform relative to the competitions.

Likewise, when I say AMD is overvalued at 158 and change, the implication is that the performance of AMD's product line relative to the competitions does not warrant such a valuation.

I am sorry that you lack the sophistication to make the connections on your own,, but there are some people that may find this type of analysis beneficial, especially now that i have explained it.

[sophisticles]

Before everyone rushes to buy stonks, wait to see if this rushed product doesn't blow up in their face. Not just metaphorically, but also literally, given the TDP figures...

I'm not looking at just the performance of Intel's current products, I am also looking at the contracts and cash infusion it has recently acquired. 3 billion dollars as part of the CHIPS act, a 5 billion dollar investment from Apollo, a 3.5 billion dollar chop deal with the Pentagon, there's a lot of interest in Intel right now:

https://fortune.com/2024/09/22/intel-apollo-offer-investment-5-billion-qualcomm-takeover/

https://www.investopedia.com/intel-reportedly-gets-usd3-5-billion-contract-to-make-chips-for-the-pentagon-8713225

https://www.cnbc.com/2024/09/16/intel-awarded-up-to-3-billion-under-chips-act.html

For a company that was supposedly facing an existential crisis there is a lot of positive news for perspective investors to be excited about.