Intel's powerhouse 144-core Xeon CPU preliminary benchmarks show performance trails behind...

Huesane

Posts: 13   +0
Bottom line: In the high-stakes competition for the best server CPUs, Intel's latest entry, the "Sierra Forest" Xeon with a whopping 144 cores is quite the show. However, it's not just about the number of cores, but how you use them, and here AMD's 128-core Epyc Bergamo is still holding all the aces.

Intel's Xeon Sierra Forest, an upcoming server chip with up to 144 efficiency cores, has recently surfaced on Geekbench 6, revealing its multi-core prowess. Set to launch in a matter of weeks, it's ready to rumble with AMD's Zen 4c Bergamo server CPUs and Arm-based server chips.

The test shows off a dual-socket setup, with two 144-core CPUs, effectively creating a 288-core system. However, it's important to note that Geekbench recorded this as two separate 144-core CPUs. Intel's ambition doesn't stop there, as they have also announced a staggering 288-core version of Sierra Forest, promising even more power and efficiency

The latest Xeon's performance with a multi-core score of 7,770 is quite impressive, especially when you compare it to Intel's own Xeon Platinum 8480+, which usually scores between 6,500 and 7,500. However, its single-core performance of 855 falls short when compared to AMD's Epyc 9754. This is because the Sierra Forest's 144 Glen E-cores are purposefully designed for efficiency rather than speed, unlike the Xeon 8480+'s 56 Golden Cove performance cores.

Feature Intel Xeon
(144-core Sierra Forest)
AMD Epyc 9754
(128-core Bergamo)
Core Count 144 128
Thread Count 144 (Efficiency Cores) 256 (Zen 4c Cores)
Single-Core Score 855 1,597
Multi-Core Score 7,770 16,455
Architecture Glen E-Core Zen 4c
Base Clock Speed 2.20 GHz Varies (Depends on specific SKU)
L3 Cache 108 MB 256 MB
L2 Cache 64 MB 128 MB
Efficiency Focus High Moderate
Expected Release First half of 2024 Already available

For its part, AMD's 128-core Bergamo CPU has already set a high standard in the server CPU market with a multi-core score of 16,455. The Zen 4c cores are architecturally identical to Zen 4 but are more area and power-efficient.

Bergamo also shows a multi-core score that substantially outperforms both Sierra Forest and Sapphire Rapids. AMD isn't just pushing boundaries with Bergamo, but setting new milestones in the AI space with their Instinct MI300X, which combines CPU and GPU.

It's relevant to consider that the Sierra Forest result comes from a pre-release chip, which might not reflect its full potential though. Additionally, Sierra Forest might even have some special features that Geekbench didn't catch, similar to how Intel's Sapphire Rapids was chosen for Microsoft Azure over AMD's MI300X GPUs due to its AI-accelerated instructions.

The benchmark does detail Sierra Forest's cache configuration, reporting 108MB of L3 cache. While the L2 cache count is unclear, Sierra Forest likely has a total of 144MB of L2 cache per CPU, combined with the L3 cache, making up for a total of 252MB. This is still smaller than Bergamo's L3 cache, which is 256MB.

In practical terms, choosing between Intel's Sierra Forest and AMD's Bergamo CPUs will ultimately depend on specific server needs. Sierra Forest, with its focus on efficiency and multi-core performance, is like a fleet of electric vehicles. It might not be the fastest on the road, but potentially efficient for the long haul. On the other hand, AMD's Bergamo, with its higher core count and larger cache, is a power-packed SUV, capable of handling all terrains.

Permalink to story.

 
Intel is really struggling with their GAA design. They're making progress but they've been having trouble making this "leap forward" they've been talking about since as early as, well, early into their 14nm process.

I don't see Intel overtaking AMD until 2028 and that's only if AMD gets lazy and stops innovating.

I'm not a financial advisor but I think I'll share my recent trades. I dropped nVidia stock and split the profits between AMD and Intel 50/50. I expect AMD to peak around the same time I see Intel stock soaring. Worst case scenario, Intel and AMD are still very profitable companies and their stock price keeps in line or exceeds inflation
 
"Sierra Forest, with its focus on efficiency and multi-core performance, is like a fleet of electric vehicles. It might not be the fastest on the road, but potentially efficient for the long haul."

Caught someone who never drove electric. 😂

Also: isn't AMD more power efficient as well as being faster?... Where are the numbers?
 
If 144 Ecores results in a mcore score of 7,770 - does that mean that 288 Ecores would score 15,540? Or does the story mean to say that with both 144 Ecore CPUs running they only scored 7,770 - it's just that Geekbench only lists it as a single 144 Ecore CPU instead of the 288?

Either way I look at it, it looks awful just based on the information provided alone that Intel is so far behind AMD for server CPUs. I don't have to build or maintain servers of any kind, so maybe I'm not seeing a good application use for Intel's offering....?
 
Epyc 9754 has a base clock of 2.25ghz and will boost up to 3.1 across all cores if thermals allow for it.
Then we have this new Intel CPU that has a 2.2ghz base with an unknown max clock speed, if it boosts at all. This alone could be why the Intel garbage delivered such a pitiful report against AMD.

Then we have the cache. less then 50% of the cache of AMD across the board. Nearly every application is going to see moderate gains when there is more cache on the CPU. Intel's choice to drop the L3 cache by 60% and the L2 Cache by 50% makes this new lineup an absolute joke compared to whats out in GA right now.

144cores should be matching the 128cores from AMD at the very least and we are not even seeing that in this early release data. I had hope for these more dense sockets from Intel, but not anymore.
 
If I understabd this right...not surprising since it's single threaded vs dual threaded
No, it's because E-cores are slow. They don't support SMT/Hyper Threading (as Intel calls it) but even with it, E-cores would be very slow against Zen4. Even without SMT disabled Zen4 would be much faster.
 
No, it's because E-cores are slow. They don't support SMT/Hyper Threading (as Intel calls it) but even with it, E-cores would be very slow against Zen4. Even without SMT disabled Zen4 would be much faster.
but doesn't the AMD CPU have two threads vs one?
 
Back