Hands On & Initial Benchmarks With An Ampere eMAG 32-Core ARM Server

[ms178]

Are you sure about that? The last notice I have is that they will use ARM for the central processor and RISC-V for the accelerators.

You made me re-check my claim. In 18 Min 55 Sec - 19 Min 31 Sec of that presentation it was mentioned that they are developing that RISC-V accelerator first but over time they want to add more IP to it to move it into a full RISC-V general purpose processor in the end with a high degree of European IP. To bridge the gap until they will have achieved this they use ARM for the general purpose CPU.

[ms178]

I was speaking about ARM CPU cores vs. x86 CPU cores in general, not about a specific chip. Adding to that, I did not "assert" anything. This is all based on what I have seen. OTOH, I have not experienced the performance of any ARMv8 chip, so I might be somewhat outdated.
If you do have data, feel free to share it.

I recommend the recent Anandtech review of the new Apple smartphone where they go deep into the architecture of the A12 and have some SPEC performance numbers in the embedded link to another article of the Cavium Thunder X2: https://www.anandtech.com/show/13392...icon-secrets/4

"What is quite astonishing, is just how close Apple’s A11 and A12 are to current desktop CPUs. I haven’t had the opportunity to run things in a more comparable manner, but taking our server editor, Johan De Gelas’ recent figures from earlier this summer, we see that the A12 outperforms a moderately-clocked Skylake CPU in single-threaded performance. Of course there’s compiler considerations and various frequency concerns to take into account, but still we’re now talking about very small margins until Apple’s mobile SoCs outperform the fastest desktop CPUs in terms of ST performance. It will be interesting to get more accurate figures on this topic later on in the coming months."

[juanrga]

I recommend the recent Anandtech review of the new Apple smartphone where they go deep into the architecture of the A12 and have some SPEC performance numbers in the embedded link to another article of the Cavium Thunder X2: https://www.anandtech.com/show/13392...icon-secrets/4

"What is quite astonishing, is just how close Apple’s A11 and A12 are to current desktop CPUs. I haven’t had the opportunity to run things in a more comparable manner, but taking our server editor, Johan De Gelas’ recent figures from earlier this summer, we see that the A12 outperforms a moderately-clocked Skylake CPU in single-threaded performance. Of course there’s compiler considerations and various frequency concerns to take into account, but still we’re now talking about very small margins until Apple’s mobile SoCs outperform the fastest desktop CPUs in terms of ST performance. It will be interesting to get more accurate figures on this topic later on in the coming months."

The problem is that anadtech testing is wrong. Their Xeon vs EPYC review crippled the performance of Xeons by huge amounts. Broadwell Xeons got SPEC scores 40% lower than expected, whereas the performance of Skylake Xeons was crippled by 60% even more

http://www.realworldtech.com/forum/?threadid=169894&curpostid=169970

http://www.realworldtech.com/forum/?threadid=169894&curpostid=170012

Apple must have matched Intel on IPC, or even can be slightly above. But Anandtech numbers aren't valid.

[ms178]

The problem is that anadtech testing is wrong. Their Xeon vs EPYC review crippled the performance of Xeons by huge amounts. Broadwell Xeons got SPEC scores 40% lower than expected, whereas the performance of Skylake Xeons was crippled by 60% even more

http://www.realworldtech.com/forum/?threadid=169894&curpostid=169970

http://www.realworldtech.com/forum/?threadid=169894&curpostid=170012

Apple must have matched Intel on IPC, or even can be slightly above. But Anandtech numbers aren't valid.

Thanks for hinting me that the Anandtech SPEC results are in dispute. As I don't have access to any of these machines I cannot judge it myself of how trustworthy their numbers are.

By the way, I have found a similarly configured Dell server with EPYC from Feb. 2018 which is a little bit faster than the one Linus Torvalds linked over in that second post, but just about 120 Points. That would still be ca. 10 % slower than the Platinum 8176 which you linked there at the end of that thread.

And we are right on topic again, I very much would like to see such a fair comparison in different workloads across different ISAs on the most recent software.

[chkaib]

Thanks phoronix for this review, it's super useful. Was wondering if you had been able to measure cache access times by any chance? Using an approach like https://stackoverflow.com/questions/...ache-latencies

Thanks!