AMD Ryzen 7 1800X Linux Benchmarks

[garwynn]

Update: Contacted a member of Red Team Plus who also had the MSI motherboard, verified this booted Ubuntu Live CD without issue. Another friend bought ASUS Crosshair VI and is setting up, he'll be testing and reporting back today.

As for the GIGABYTE AORUS... I have a hard time recommending this one for Linux at the moment. Journaling won't start and "unexpected irq trap at vector 07" spams dmesg. 17.04 server will install but failed to boot without acpi=off nomodeset (going to verify if second one is really needed). I'm trying to install ubuntu-desktop this way and see if it will finally let me in. Have tried BIOS defaults, adjusted, enabled/disabled IOMMU... most people aren't going to want to go through this for an install.

[Nille_kungen]

I saw some tests on windows that showed some interesting result when SMT was turned off in games.
I wonder if there are interesting results also in linux games?

[mlau]

From the benchmarks I gather:
- the core is at least as good as intels
- the avx unit sucks (it's half as fast as intels)
- lot of improvements to be made in turbo (intel's turbo is still much better/aggressive), and memory controller (p7zip is comparatively slow on zen)
- it compiles stuff super fast, which, as a Gentoo user, is always nice.

[garwynn]

GIGABYTE progress, FINALLY!

acpi=off is a must at the moment. Also the out of range error is specific to HDMI, changed to DVI and I have a working display. Now to see how it runs and if we can improve that at all.

[jrch2k8]

From the benchmarks I gather:
- the core is at least as good as intels
- the avx unit sucks (it's half as fast as intels)
- lot of improvements to be made in turbo (intel's turbo is still much better/aggressive), and memory controller (p7zip is comparatively slow on zen)
- it compiles stuff super fast, which, as a Gentoo user, is always nice.

1.) yea in some reviews well done they actually tested IPC at the same frequency and it is between skylake and kabylake pretty much, of course a quad core can reach a higher frequency
2.) i'm not so sure is that slow but i agree is slower
3.) As far as i understood the proper reviews, there is no drivers for the CPU on any OS at the moment, so is very likely once win10 and linux get the fixes the performance will improve
3a.) As it happened with intel before i suspect there is something wrong with the Memory controller at the BIOS/EFI level, that latency make no sense but well some manufacturers are aware and it seems BIOS updates are cooking to fix it and the memory speeds which are botched at the moment, this should drastically help p7zip and the likes.
4.) fuck yeah

[schmidtbag]

From the benchmarks I gather:
- the core is at least as good as intels
- the avx unit sucks (it's half as fast as intels)
- lot of improvements to be made in turbo (intel's turbo is still much better/aggressive), and memory controller (p7zip is comparatively slow on zen)
- it compiles stuff super fast, which, as a Gentoo user, is always nice.

Seems like a reasonable (or should I say, ryzenable! LOL *crickets*) synopsis.

I don't know much about AVX, but I'm guessing it's not so great because it isn't so relevant anymore, considering OpenCL and CUDA. I think the turboing is just fine, but the memory controller leaves much to be desired. As stated before, there are supposed to be fixes for that so higher frequencies can be achieved. As pointed out by Nille_kungen, there are SMT issues that need to be resolved too, though, it seems Linux is less affected by it than Windows.

I suspect the Ryzen 5 and lower models will scale better. By the time they're released, most of the BIOS, microcode, and OS updates should be settled, and these models will have at least 4 fewer threads, which means the memory controller won't be so overloaded.

[droidhacker]

Haha if you think 1/10th the budget is small uh... AMD's total assets (so, the company's worth as a whole) is almost 1/4 of Intel's net revenue. In other words, Intel could buy all of AMD 3x over and still have billions of dollars left over. Even the FX series CPUs would be impressive when you consider how little AMD has in comparison.

Net revenue doesn't mean what you apparently think that it means. It doesn't account for the cost of doing business.

So you have sales that equals a billion dollars this year... that's nice. What did it COST you to make those sales? Did it cost you $2B to make sales of $1B? Your company could be WORTHLESS, even with a net revenue of a billion dollars!

Here, read more about it;

https://www.entrepreneur.com/article/66442

Note: Not saying that intel isn't way bigger than amd, just that you're pulling your justifications for that claim out of your a$$.

[jrch2k8]

Seems like a reasonable (or should I say, ryzenable! LOL *crickets*) synopsis.

I don't know much about AVX, but I'm guessing it's not so great because it isn't so relevant anymore, considering OpenCL and CUDA. I think the turboing is just fine, but the memory controller leaves much to be desired. As stated before, there are supposed to be fixes for that so higher frequencies can be achieved. As pointed out by Nille_kungen, there are SMT issues that need to be resolved too, though, it seems Linux is less affected by it than Windows.

I suspect the Ryzen 5 and lower models will scale better. By the time they're released, most of the BIOS, microcode, and OS updates should be settled, and these models will have at least 4 fewer threads, which means the memory controller won't be so overloaded.

well AVX is very relevant for HPC but is still too new for most developers, so everybody target SSE2 generic x64 to avoid specialization but once you use AVX vectors the win is massive in many scenarios(specially in parallel SIMD vectored code)

[mlau]

1.) yea in some reviews well done they actually tested IPC at the same frequency and it is between skylake and kabylake pretty much, of course a quad core can reach a higher frequency
2.) i'm not so sure is that slow but i agree is slower
3.) As far as i understood the proper reviews, there is no drivers for the CPU on any OS at the moment, so is very likely once win10 and linux get the fixes the performance will improve
3a.) As it happened with intel before i suspect there is something wrong with the Memory controller at the BIOS/EFI level, that latency make no sense but well some manufacturers are aware and it seems BIOS updates are cooking to fix it and the memory speeds which are botched at the moment, this should drastically help p7zip and the likes.
4.) fuck yeah

if i remember correctly, some tech slides published by amd in the past indicate zen's avx unit is half as wide as haswell's, which may explain the performance difference. the avx2 unit in haswell+ is really impressive, espcially when fed with properly tuned code (e.g. llvm 4 code of scimark 2 sparse matmult, I hit more than 12000pts at 3.9GHz).
amd probably though that initially avx performance was enough for current needs, and I'm sure they'll have competitive performance with zen2/3.

Did Michael publish scimark2 results for the 1800X?

other reviews I've read indicated that the bios code of the test boards is really early beta quality. I hope they can increase performance there. I was a bit disappointed that
at full 64gb the memory clock is apparently dropped to ddr4-1800 or lower.

Overall it's a great first iteration. Looking forward to letting gentoo loose on one.

[schmidtbag]

Note: Not saying that intel isn't way bigger than amd, just that you're pulling your justifications for that claim out of your a$$.

Maybe if you actually spent the time researching Intel's net income instead of wasting your time lecturing me, you'd realize that my statement still holds true.

Look whose "a$$" is spewing unjustifiable claims now!

Maybe "revenue" was the wrong word, but it still doesn't change my point. Intel's net income averages around $10 billion. That's not operation costs, that's their profit.