AMD EPYC Launching 20 June, Are You Interested?

[torsionbar28]

Michael, yes yes yes please please please for some EPYC benchmarks! I have 500+ Opteron 6300 cores at work, and 24 Opteron 4300 cores at home. All running Linux. Would love to see how the new stuff compares.

[Veerappan]

I'm curious to see how well the zen architecture scales, but I'm definitely not in the target market for this chip. My interest caps out around 8-16 core, after which the frequency trade-off doesn't make as much sense for what I'm doing.

If you have to spend money on the chip, I'd save your money and wait for Vega, or possibly the ThreadRipper launch instead.

[polarathene]

Yes, kernel compile speed on EPYC will be EPIC!

That's been done here: https://www.forbes.com/sites/tiriasr.../#62224c372584

Dual sockets boards: x2 Epyc(15.7 sec) vs x2 Intel Xeon E5-2699A v4(22.5 sec)

https://blogs-images.forbes.com/tiriasresearch/files/2017/05/AMD-Epyc-7-1200x682.jpg?width=960

x1 Epyc(33.7 sec) vs x2 Intel Xeon E5-2650 v4(37.2 sec)

https://blogs-images.forbes.com/tiriasresearch/files/2017/05/AMD-Epyc-11-1200x671.jpg?width=960

[polarathene]

I'm also interested in how Epyc or Threadripper performance scales due to them using NUMA in single socket, specifically in QEMU/KVM vm setup. Does the new Intel chips also provide many cores in the same approach causing them to use NUMA topology too?

I don't know too much about NUMA other than it usually wasn't an issue for single socket systems and related to their resources(PCI-e, RAM), that performance decreased if you crossed had to cross the interconnect to access the other sockets NUMA node to use it's resources. AFAIK that now is an issue on single socket with these CPUs?

[nevion]

You bet I'm interested. HIghly anticipating them and especially want to measure the overhead of things like openmp (potentially with the AOCC's libomp). If I read the marketing and design goals right - this will be the lowest overhead for threading to date. It's such bag of hair on general purpose processors in comparison to things in the embedded space or GPU space where things operate with much lower overhead. Might be my next hardware build in the home. Of course I'm interested in comparing benchmarks with intels latest as well - and on that note Phoronix is probably going to be the only case of reason vs all the other noise that comes from every pretender who thinks they can benchmark these accurately with games. I wonder if these differences will show up suitably with 32 cores on the existing benchmark software though. We're bumping up into core numbers the PTS has less benching with and I don't know if any individual test in the multicore benchmark can illustrate the lower overhread of thread communication or utilize (or feed) that number of threads.

[boxie]

Server and workstation benchmarks would be super interesting - both Threadripper (how awesome is that name) and EPYC platforms.

[Jeff99]

I'm pretty interested in Epyc, but honestly I'm more interested in a Threadripper HEDT setup, though I'm worried that ECC support and on-die 10GbE is going to be omitted there.

[moilami]

Michael, please provide some method of reporting typos within article. I am all the time so disapointed, when I come to read some rants of different fanboys and all I get are some typos reports ;-)

I disagree, I find it amusing to see has Mr. Spell Checker found anything this time But I agree with you regarding rants, it just seems there are less and less of them, could this indicate more mature software, I wonder, and also call for more heated arguing in the forums, maybe Michael could put some effort in trolling and be not too much afraid of those blaming himself of that Now these forums are like people would really be on meds instead of Good Old Times when people suggested to take meds all the time in the heat of Argument

[Luke_Wolf]

I disagree, I find it amusing to see has Mr. Spell Checker found anything this time But I agree with you regarding rants, it just seems there are less and less of them, could this indicate more mature software, I wonder, and also call for more heated arguing in the forums, maybe Michael could put some effort in trolling and be not too much afraid of those blaming himself of that Now these forums are like people would really be on meds instead of Good Old Times when people suggested to take meds all the time in the heat of Argument

There hasn't been any new controversial software for a while, and with Canonical having finally surrendered the only current longstanding conflict has been resolved. Don't worry though, more are coming and the next battle may be sndio vs Pulse. Just wait a while and someone will pop up with a long bitter crusade ahead of them to fix something that's crappy in Linux with Luddites fighting that change.

[moilami]

There hasn't been any new controversial software for a while, and with Canonical having finally surrendered the only current longstanding conflict has been resolved. Don't worry though, more are coming and the next battle may be sndio vs Pulse. Just wait a while and someone will pop up with a long bitter crusade ahead of them to fix something that's crappy in Linux with Luddites fighting that change.

Laughed so hard reading this, have to say love this Free Software community.