AMD Ryzen 7 1800X Linux Benchmarks

[C_Monkey]

I hope when you do the linux game testing you try with both SMT on and off. On windows systems Ryzen shows considerable gaming performance with SMT off and wondering how it compares.

[C_Monkey]

Hope we can get some Linux gaming benchmarks with and without SMT enabled. Windows benchmarks with SMT off have shown improved performance and it'll be interesting to see on the linux side.

[daemon32]

Really interested to know as well, Ryzen would be one killer VM host with a couple of GPUs.
I have seen some kernel commits regarding the IOMMU stuff, but I´m not sure what the status is.

I want to add myself to those interested in ACS on CPU ports and IOMMU support by the chipset.
Also I was planning to buy MSI X370 GAMING PRO CARBON so it would be great if the MSI EFI implementation supported it.

Well, unfortunately, it appears that the AMD-Vi is broken on at least one board (ASUS PRIME X370-Pro)

[Firmware Bug]: AMD-Vi: IOAPIC[17] not in IVRS table
[Firmware Bug]: AMD-Vi: IOAPIC[18] not in IVRS table
[Firmware Bug]: AMD-Vi: No southbridge IOAPIC found
AMD-Vi: Disabling interrupt remapping

EDIT: I found out what Michael was talking about with the 'I/O page faults'

AMD-Vi: Event logged [IO_PAGE_FAULT device=24:00.0 domain=0x0003 address=0x000000f400228680 flags=0x0010]

So it appears that at least on the MSI X370 XPOWER GAMING TITANIUM board, the IOMMU is trying to do its job.
It also appears in lspci!

Slot:    00:00.2
Class:    IOMMU [0806]
Vendor:    Advanced Micro Devices, Inc. [AMD] [1022]
Device:    Device [1451]
SVendor:    Advanced Micro Devices, Inc. [AMD] [1022]
SDevice:    Device [1451]

And I bet you can guess what device 24:00.0 is...

Slot:    24:00.0
Class:    VGA compatible controller [0300]
Vendor:    Advanced Micro Devices, Inc. [AMD/ATI] [1002]
Device:    Fiji [Radeon R9 FURY / NANO Series] [7300]

EDIT2: It appears that, at least in some configurations, IOMMU groups aren't setup correctly.
However, they do state in the video description:

Also note: There maybe ACS patches for KVM virtual i/o passthrough that DO work, but we need to do more testing.

[Klassic Six]

It came in fucking wooden box? that's very cool! anyway that's nice to see finally AMD can competition with Intel but unfortunately it looks like in gaming (based on Michael's gaming benchmark) AMD still behind the Intel. Michael can you post the photo of it's original cooler? if any.

[pixo]

EDIT: I found out what Michael was talking about with the 'I/O page faults'

AMD-Vi: Event logged [IO_PAGE_FAULT device=24:00.0 domain=0x0003 address=0x000000f400228680 flags=0x0010]

So it appears that at least on the MSI X370 XPOWER GAMING TITANIUM board, the IOMMU is trying to do its job.
It also appears in lspci!

Slot: 00:00.2
Class: IOMMU [0806]
Vendor: Advanced Micro Devices, Inc. [AMD] [1022]
Device: Device [1451]
SVendor: Advanced Micro Devices, Inc. [AMD] [1022]
SDevice: Device [1451]

And I bet you can guess what device 24:00.0 is...

Slot: 24:00.0
Class: VGA compatible controller [0300]
Vendor: Advanced Micro Devices, Inc. [AMD/ATI] [1002]
Device: Fiji [Radeon R9 FURY / NANO Series] [7300]

What is interesting is once the driver kicks in and initialize the GPU the errors stop.

Michael could you populate 2 of the CPU PCIe slots, 1 from the X370 and run find /sys/kernel/iommu_groups/ -type l

[mitcoes]

Now you need to add some ARM SoCs running Linux, as the new System 76 server, or those with Chrome OS than can run at least Crouton
better if it has one synthetic test at the end and some GPU tests as notebook check has.

Last but not least if you want to make extra money with benchmarking a linked list with all the models for every main SoC and links to buy each model (with returns for you) as the page KIMOVIL has, like this https://www.kimovil.com/es/donde-com...uk-z2-4gb-64gb in spanish where I found my info to buy my phone on their cheaper link and they made their revenue.

[Adarion]

Does anyone know if the Ryzen still support ECC? Could be a great chip for home servers if they do.

Yesterday's ask me anything on reddit and some other infos say yes. Iirc. Lisa herself said yes to ECC, but it's not as thoroughly tested as the upcoming Naples will be. So you can build a "consumer" machine with ECC (Register-RAM not supported but you only need that on servers with lots of mem, for that will be Naples). It might depend on the mainboard manufacturerer how well they'll support ECC, though. (My knowledge lacks info if you need actual HW on the _mainboard_ for ECC (I think not really, buffer/reg. needs) - but you might want to have it supported in FW for machine bringup though I've read somewhere that the kernel can also activate and use the ECC if present later).
The memory controller of Ryzen can do ECC, so if you have UDIMMs with ECC your chances are already good.

[k1e0x]

So

In this case is not reliable argument. I said that 7700k wins in games and I meant if a customers do not need MT (read as "play games"), Intel is preferable.

Wins in games today optimized for intel and 4 cores. Games maybe will use more cores in the future with PS4 and Xbox one each having 8 making Ryzen potentially better choice now for the coming 2-5 years. (tho we've heard that before)

It is probably an optimization type thing tho because the compute benches show it's a good performer. not sure.

[pszilard]

Michael Where can the configure options you pass to each software be checked. Why is that not listed anywhere? E.g. I really hope you are not compiling FFTW without SIMD optimization (i.e. --enable-sse --enable-avx --enable-avx2 --enable-fma should probably all be passed here)!

[Manuel Jose]

I think that the low result for Himeno is because AVX2.

I copy from phoronix site:

pts/himeno-1.2.0 [20 Jan 2016 19:22:28 UTC]
- Use AVX2 by default if available.

pts/himeno-1.1.0 [07 Dec 2011 09:49:26 UTC]
- Use -O3 cc flag by default and add CFLAGS var.

pts/himeno-1.0.0 [06 Dec 2010 14:52:26 UTC]
- Initial import into OpenBenchmarking.org