Running The Ryzen 7 1700 At 4.0GHz On Linux

[DanL]

Michael, thank you for all of your Ryzen coverage. I wanted to ask about the sensors on the MSI X370 mobo you have. In your initial Ryzen review, you said:

When running LM_Sensors' sensors-detect with the X370 motherboard, no onboard sensors were detected and thus no support for monitoring the board's fan speeds, voltages, temperatures, etc.

I can't see exactly what Super I/O chip your board has, but it is a Nuvoton of some sort, and maybe it is an NCT6793D. If that's the case, try manually modprobing the nct6775 module:

modprobe nct6775
sensors

The 3.4.0 version of lm-sensors (and sensors-detect) found in Debian/Ubuntu does not yet support automatic detection of the NCT6793D. You would have to use the latest git/svn for that to work: https://github.com/groeck/lm-sensors...master/CHANGES

[VikingGe]

With ease I was able to push the Ryzen 7 1700 to 4.00GHz by feeding the CPU 1.3875V rather than the default 1.1~1.2V.

Looks like you have quite a decent chip there, some reviewers seem to have trouble reaching 3.8 GHz. And that voltage should be accetable vor 24/7.

mrpime doesn't currently detect the CPU properly and will pick the K10 kernel, which doesn't use AVX/AVX2

This should make almost zero difference on Ryzen, given that its 128-bit pipelines can be kept busy using pure old SSE code. In fact, the SSE version should stress the frontend more, given that it's basically twice the code doing the same work compared to AVX.

And while mprime is great as an initial stability test and to check whether the cooling is good enough, it doesn't necessarily mean that the system is stable. While trying to find the highest possible overclock on my Phenom, I could run mprime for an hour but some PTS tests (John the Ripper in particular) would almost reliably crash at some point, plus some other random application crashes. Needed to bump the voltage even further.

[Mark Rose]

This should make almost zero difference on Ryzen, given that its 128-bit pipelines can be kept busy using pure old SSE code. In fact, the SSE version should stress the frontend more, given that it's basically twice the code doing the same work compared to AVX.

And while mprime is great as an initial stability test and to check whether the cooling is good enough, it doesn't necessarily mean that the system is stable. While trying to find the highest possible overclock on my Phenom, I could run mprime for an hour but some PTS tests (John the Ripper in particular) would almost reliably crash at some point, plus some other random application crashes. Needed to bump the voltage even further.

I guess that depends on what in particular you're trying to stress, so fair point on using applications that are less memory-bandwidth-bound. I focus on getting the most throughput on LL tests. On Intel, the AVX2 code will stress the memory system more (I actually disable turboboost on my i5-6600 cluster because dual-channel dual-rank DDR4-2133 can't keep up with any more than 4 core Skylake at 3.3 GHz).

[schmidtbag]

Until somebody figures out why some of the gaming performance is *much* worse than on windows with some titles and vulkan (that confuses me), I will be watching and waiting. Still leaning towards buying a Ryzen system as it hits the ball out of the park on everything other than gaming right now.

It isn't "much" worse, it's just sometimes noticeably worse if you're playing a game at 1080p at more than 120Hz. For the vast majority of people (including those who have higher-res screens) the CPU can play games just fine.

But, it's not a gaming CPU, nor was it really marketed toward gamers.

[dungeon]

And while mprime is great as an initial stability test and to check whether the cooling is good enough, it doesn't necessarily mean that the system is stable. While trying to find the highest possible overclock on my Phenom, I could run mprime for an hour but some PTS tests (John the Ripper in particular) would almost reliably crash at some point, plus some other random application crashes. Needed to bump the voltage even further.

That is why people run mprime together with some unigine benchmark in the same time... that should burn both CPU and GPU and hopefully something will not burn really in the process

If that survive at least 2 hours, OC is stable at least for gaming

And if someone wants 100% stable machine ultimately under any condition, then do not overclock it at all of course

[LeJimster]

It isn't "much" worse, it's just sometimes noticeably worse if you're playing a game at 1080p at more than 120Hz. For the vast majority of people (including those who have higher-res screens) the CPU can play games just fine.

But, it's not a gaming CPU, nor was it really marketed toward gamers.

It seemed like on windows I was seeing around 20% down on kabylake benchmarks @ 1080p. The first benchmarks Michael posted on Linux were showing 50% down in some games.

I know it isn't exactly a gaming CPU but my worry is the r5s won't be able to go much faster either until they've spent more time understanding and maturing the 14nm process.

[leipero]

There's amusing typo in article:

As a reminder, the Ryzen 7 1800 eight-core processor has a 3.0GHz base clock frequency and a 3.7GHz turbo frequency.

.

[andrei_me]

Michael power consumption please, performance per watt is a strong point of 1700

[funfunctor]

This CPU just made Gentoo a viable distro for me now!

[qsmcomp]

Michael Do not use that cheap MSI B350M to overclock Ryzen 7 1700. Your mobo might explode like your MSI X99. Try using ASROCK X370 Taichi for overclocking.