Intel Xeon 6766E/6780E Sierra Forest vs. Ampere Altra Performance & Power Efficiency

Written by Michael Larabel in Processors on 5 June 2024 at 11:22 AM EDT. Page 6 of 6. 10 Comments.
Xmrig benchmark with settings of Variant: GhostRider, Hash Count: 1M. EPYC 9684X was the fastest.
Xmrig benchmark with settings of Variant: GhostRider, Hash Count: 1M. EPYC 9684X was the fastest.
Xmrig benchmark with settings of Variant: GhostRider, Hash Count: 1M. EPYC 9684X was the fastest.

With the Xmrig mining software, the Xeon 6780E was delivering 2.2x the performance of the Ampere Altra Max M128-30. On a performance-per-Watt basis the Ampere Altra Max was delivering comparable efficiency to prior generation Intel Xeon Scalable CPUs but at around 61% the performance-per-Watt of the Xeon 6766E.

Stockfish benchmark with settings of Chess Benchmark. EPYC 9684X was the fastest.
Stockfish benchmark with settings of Chess Benchmark. EPYC 9684X was the fastest.
Stockfish benchmark with settings of Chess Benchmark. EPYC 9684X was the fastest.

With the very scalable Stockfish chess benchmark as another nice cross-platform reference, the Intel Xeon 6766E was yielding nearly twice the performance per Watt of the Ampere Altra Max M128-30. With the Xeon 6700E series Intel has made great strides in power efficiency compared to prior generation processors and making them more competitive to AMD EPYC. It will be interesting to see how AmpereOne compares to Intel Xeon 6700E if/when I find hardware access.

Geometric Mean Of All Test Results benchmark with settings of Result Composite, Intel Xeon 6700E Sierra Forest vs. Ampere Altra vs. AMD EPYC. EPYC 9684X was the fastest.

When taking the geometric mean of the wide range of benchmarks conducted across these single socket processors and the workloads that worked both on x86_64 and AArch64, here's the result. The Ampere Altra Max M128-30 came out similar to the Intel Xeon Platinum 8380 (Ice Lake) overall, which is actually not too bad considering Ampere Altra Max and Intel Xeon Ice Lake were both shipping in 2021... Mind you over the past three years there has been a lot of AArch64 improvements to the Linux kernel, compilers, and other open-source software to help enhance the 64-bit ARM performance so the Ampere Altra Max performance has matured rather well during its lifespan compared to when those processors first shipped. With this geo mean, the Xeon 6780E was about 68% faster than the Ampere Altra Max M128-30 and the Xeon 6766E around 54% faster than that ARM server competition.

CPU Power Consumption Monitor benchmark with settings of Phoronix Test Suite System Monitoring.

When looking at the CPU power consumption over the entire span of benchmarks, the Ampere Altra Max M128-30 on average was consuming around 126 Watts with a peak of 253 Watts. The M128-30 has a 183 Watt rated TDP. Meanwhile the Intel Xeon 6766E is rated for 250 Watts and the Xeon 6780E at 330 Watts. But as shown in these benchmarks, on average the Xeon 6700E processors with these workloads were averaging in the 140~169 Watt range. So while the Intel Xeon 6766E/6780E were consuming 18% and 34% more power on average, these Sierra Forest processors on average were 54~68% faster than the currently-shipping Ampere Computing top-end cloud native processor.

The Intel Sierra Forest processors are delivering better power efficiency overall than the ARM-based Ampere Altra Max processors. It will be interesting to see how AmpereOne compares when hardware actually becomes available. One area where Ampere Altra still maintains an advantage is for those wanting a high core count server processor but really trying to maximize value. The Intel Xeon 6780E now post-launch is on Intel Ark and has a recommended customer price of $11,350 USD and the Xeon 6766E for $10,257. Granted, hyperscalers and large customers typically are able to obtain the CPUs for much less. Meanwhile with the mature Ampere Altra it's possible to buy a 128-core processor and ASRock Rack motherboard together for under $3k USD. Plus in using DDR4 server memory is a less initial investment as well compared to DDR5 prices. So for those wanting a dense server but really after maximizing value, there still are opportunities there for Ampere Altra if not after the best performance and power efficiency.

Following AMD Zen 4C last year and now Intel Xeon 6700E series, Ampere Computing really needs to move on with making AmpereOne availability if they want to continue promoting their cloud native power efficiency. There is also the matter of Ampere Computing having only announced AmpereOne models at 144, 160, and 192 cores (and the higher core counts next year) while leaving 128 cores and less to Ampere Altra... Unless there's been any quiet/unannounced change in plans for lower core count AmpereOne processors, that's leaving more opportunities ahead for AMD Siena (64 cores and less) and Bergamo (112 to 128 cores) and Intel Sierra Forest at 64 to 144 cores for those wanting very power efficient servers and leading performance-per-Watt for medium sized workloads but not necessarily building out for a very dense deployment. Ampere Computing's focus on the high core count parts only made a lot of sense with their public cloud provider wins, but now that Amazon, Microsoft / Azure, Google / GCE, and others are designing their own ARM server processors there is more pressure in that space too.

In any event thanks again to Intel for providing the Xeon 6 reference server and Xeon 6766E/6780E processors for benchmarking at Phoronix. In my continued testing of these Xeon 6700E processors it's certainly been a refreshing surprise to see how far they have pushed their power efficiency in one generation while waiting to see where Intel Xeon 6900P Granite Rapids will land at the top end for raw performance.

If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.


Related Articles
About The Author
Michael Larabel

Michael Larabel is the principal author of Phoronix.com and founded the site in 2004 with a focus on enriching the Linux hardware experience. Michael has written more than 20,000 articles covering the state of Linux hardware support, Linux performance, graphics drivers, and other topics. Michael is also the lead developer of the Phoronix Test Suite, Phoromatic, and OpenBenchmarking.org automated benchmarking software. He can be followed via Twitter, LinkedIn, or contacted via MichaelLarabel.com.