Intel Xeon Platinum 8592+ "Emerald Rapids" Linux Benchmarks

[phoronix]

Phoronix: Intel Xeon Platinum 8592+ "Emerald Rapids" Linux Benchmarks

Following the 5th Gen Xeon Scalable "Emerald Rapids" overview, you are likely wondering about the performance claims made by Intel and how they shake up in independent testing as well as how Emerald Rapids competes against AMD EPYC Genoa(X) and Bergamo. If so this article is for you with the Phoronix benchmarks of the new flagship Intel Xeon Platinum 8592+ 64-core processors being tested in both single and dual socket modes.

https://www.phoronix.com/review/intel-xeon-platinum-8592

[Kjell]

Hope Intel adjusts their prices accordingly to make up for the ridiculous power consumption and worse performance..

Let's be honest here. Intel could sell those CPUs at 1/3 of the price and still make a bank.

[Anux]

Let's be honest here. Intel could sell those CPUs at 1/3 of the price and still make a bank.

Only if anyone would be willing to buy them for 1/3. Not sure this is enough to mitigate the cooling costs over time. It's roughly double power requirement to AMD offerings.

[sobrus]

It is nice to see what both AMD and Intel are capable of doing with x86 architecture...

...while in consumer space we still have dual channel memory just like 20 years ago with my nForce2 chipset.
And the only performance benchmarks for AMD regarding Ryzen and Radeon hardware are FPS in games, because it is obvious that PCs are only used for gaming.

But to be honest, current situation is still better than "quad core i7" for like 10 years in a row

[Volta]

A whole page of OpenVINO so Intel doesn't look like total crap?

[avis]

These charts are very hard to read at least for me.

Would be great if you posted performance per core and performance per watt as well.

[andyprough]

It's roughly double power requirement to AMD offerings.

More like 20%, doesn't look even close to "roughly double".

[sophisticles]

I apologize in advance, intel has offered me a lot of money to shill for them and Christmas is coming up and I haven't worked since March, so i really need the money.

With that disclaimer out of the way:

EPYC 8534P 64C/128T, 2.3ghz, 3.1ghz
EPYC 8534PN 64C/128T, 2ghz, 3.75ghz
EPYC 9554 64C/128T, 3.1ghz, 3.375ghz
EPYC 9654 96C/192T, 2.4ghz, 3.55ghz
EPYC 9684X 96C/192T, 2.55ghz, 3.42ghz
EPYC 9754 128C/1256T, 2.25ghz, 3.1ghz

The first speed rating is the base clock and the second is the all core max,

Now let's compare with Intel:

Xeon Platinum 8592+ 64C/128T, 1.9ghz and a single core max boost of 3.1ghz, I could not find the all core max.

So, with a huge clock speed advantage and in some cases double the core count, AMD is able to beat Intel's offerings.

Who could have possibly expected such an outcome?

As I have said all along, AMD's secret weapon is in their superior manufacturing process, that has allowed them to cram more processing resources into a similar sized die and better power envelops.The have created these super wide cores that are able to fetch, execute and retire massive amounts of instructions per cycle through brute force.

For the Xeon Platinum 8592+ it comes with 4 accelerators that if properly used can pay huge performance dividends at very low power consumption levels and if you are willing to spend the cash, you can unlock an additional 12 accelerators for a total of 16.

As Michael mentioned, just wait until next year when Intel releases their all E core Xeons, that will have huge core counts and sip power, and I have to believe they will eventually release an all E core Xeon that also features massive amounts of onboard ram, like the Xeon Max and possibly a bunch of accelerators.

AMD has had a good run brute forcing performance with lots of fat cores, but there is only so far that strategy can take you.

Of course there is the possibility that AMD counters by releasing processors with the same fat, wide cores but with SMT-4 capability, meaning each core could handle 4 threads per cycle, for a CPU like 64C/256T.

2024 should be a very interesting year.

[brunosalezze]

amounts of instructions per cycle through brute force.

Calling it brute force is to underrate hundreds of employes that worked years to design, print, tape and release these CPUs, if it was this easy Intel would have done it first, they print CPU cores on TSMC as well now.

[Mark Rose]

Michael gets all the fun toys!

A nice incremental bump, for sure. I'm sure many of these will sell as upgrades.