AMD Ryzen discussion

Well for anyone interested I finally gave up. While benchmarking apps showed a 15%-20% performance penalty (which sucks but for me is livable) some games I tried (like Grim Dawn) went from 60fps native to 15fps. The cause of both (the 20% and the gaming degradation) is apparently a bug with the Linux KVM Hypervisor and Ryzen.

The other Linux hypervisor (Xen) was supposed to get me much better performance (due to not having that bug) but the version that comes with Ubuntu doesn’t support Ryzen, and building the newest versions wasn’t working for me for whatever reasons.

Since the KVM bug is a software bug that can be fixed, I’ll probably try again in another 6-8 months (I have the hardware so why not) but for now it’s windows native with a Linux VM for me.

Fuck me. I spent so much time trying to get the GPU passthrough setup I didn’t really do basic stuff with the processor, like Gaming.

It all started with me having hard locks when running the Unige Heaven benchmark. 100% of the time it would get to this one spot and my whole computer would hard lock with the display going black. Even in a virtual machine it would go black, and eventually my Linux host would freeze (though not right away, and if music was playing with spotify the music would oddly keep playing).

3dmark would run fine so I thought it was just a weird issue with the Heaven benchmark (dumb assumption I know) but I ignored it. Fast forward to today where I’m playing Grim Dawn and the same thing happens.

I know it’s not my video card, because I’ve had this R9 Fury for almost a year now and never had an issue. I do some gooogling and find instability with original batches of Ryzen (https://www.phoronix.com/scan.php?page=article&item=new-ryzen-fixed&num=1) where Linux users could easily reproduce a segmentation fault compiling GCC over and over again (https://github.com/suaefar/ryzen-test). I run this test and lo and behold it fails within a minute :-/.

The processor has 1726SUS on it, which apparently means it was made the 26th week of this year, so it’s one of the early batches. Sent an email to AMD and hopefully they replace it quiickly without hassle :(

Damn. Sorry to hear about the bad luck fighting the bleeding edge on this one. GL.

I’ll give props to AMD customer service at least. I emailed them Monday night with all my info and saying I think I have the early batch seg fault bug. By the time I woke up they gave me an RMA form to fill out. Wednesday morning I woke up with a fedex label and mailed it off. They received it thursday and my replacement came today (Friday). Unfortunately, I was on the way out the door to go out of town when it came so I haven’t been able to test it out.

Hey look!

Judging from our results with the i7-8700K, we have a meaningful change from what we’ve come to expect from Intel. Its flagship processor gets a jump in core/thread count and the numbers show how much of an improvement it makes in CPU-heavy tasks that are able to leverage more than four cores.

However, there isn’t much to point to when it comes to performance in gaming, since games are heavily reliant on the GPU and aren’t geared to take advantage of more than four CPU cores quite yet.

You quoted yourself from May saying something that literally no one on the thread disagreed with just to pat yourself on the back?

Who in the world was claiming that a 6th core was what they really needed for 4k gaming or whatever?

Hey guys, just because Coffee Lake supports 64GB of RAM and that can be useful if you have a lot of virtual machines running or other RAM intensive tasks, having 64GB of RAM will not matter for running Overwatch.

I will quote myself in 6 months when history proves this bold statement prescient.

Skylake has supported 64GB for quite some time and it was the first Intel consumer chip to bump the limit from 32GB.

I don’t know if anyone cares but whatever, I’m excited.

Previously when I tried to do GPU passthrough to a vm (Linux host, windows virtual machine for gaming) I had a massive frustrating time. Several things have changed which made me want to try again.

  1. 90% of my issues before turned out being a bad power supply. My power supply I had previously used in my i5-3950k was either not good enough for my R9 Fury + Ryzen 1600X, or it just degraded. Either way buying a new (and good quality) power supply made my Ryzen box work well for gaming.
  2. The mainstream virtual machine hypervisor in Linux (KVM) had some major patches that fixed a lot of performance issues people with AMD CPUs had, bringing performance theoretically close to bare metal in gaming.
  3. There’s a project called Looking Glass that does very quick framebuffer copies directly from the GPU’s vram into your host computer’s vram, which allows you to have your windows virtual machine inside of Linux without any perceptible latency and no fps loss. I haven’t gotten to try it yet.

So I got my RX 550 in the mail and gave it a go. I was able to get up and running in about an hour or so and so far I’m really happy with the setup.

Initial benchmarks are weird but encouraging. Unigine Heaven benchmark gave me 72.2fps on the virtual machine where as I got 62.2 on bare metal windows prior to nuking it. That’s bizarrely backwards so I’m just assuming that something was running in the background without my knowledge (the vm is a brand new install).

3dmark timespy gave me a graphic score of 4888 (33.19 fps and 27.05fps) on the virtual machine vs 4902 (33.54 fps and 26.99 fps). The CPU score was notably lower (2914 vs 5928) but that’s to be expected since on bare metal it could use all 6 cores/12 threads but I’m only giving the vm 3 cores / 6 threads.

3dmark firestrike ultra gave me a graphics score of 3950 (20.77 fps and 14.64 fps) on the vm vs 3904 (20.57 fps and 14.45 fps) on bare metal.

All in all, so far very encouraging. Too bad I need to go to bed now instead of actually testing it out with actual games.

At least I care; in the opening post of this thread my big question mark on Ryzen was whether they could get the virtualization story for passthrough GPUs right. Took a year, but sounds like it’s getting there. Thanks for the report :)

Wow, Looking Glass works amazingly well. I have my windows VM with 3dmark running at full speed in a window in the corner of my screen. I can just move my mouse to the window and it automatically gets captured by the vm, and moving the window to the boundaries puts it right back into the linux host. I haven’t noticed any lag, though all i’ve tried is the windows GUI, Shadow Tactics and 3dmark.

I’m blown away.

You’re making me want to install Linux on my desktop for the 1st time in 10 years…

Zen+ is out of NDA today. As expected, basically unchanged IPC but on a slightly smaller process so it runs at higher clockspeeds. Roughly a 10% performance improvement overall. The 2700X will clock 4.2Ghz all-cores or boost up to 4.3Ghz on select cores, compared to Zen non-plus which topped out at 4.0Ghz at most.

More interesting is the new clock boost tech in Zen+ which works dynamically on a per-core basis rather than pre-set steps. That basically means the CPU runs more cores at higher clocks than Zen non-plus, and will use more power and generate more heat until it reaches its thermal design threshold where it backs off.

Looking at the same tradeoff versus Intel as the Zen non-plus; you get more cores but they run slower. Difference is now that tradeoff is markedly smaller, so it’s a much more attractive choice.

Also all of the Zen+ CPUs come with heatsinks and fans now, and the highest-end one in the 2700X compares well against $30 aftermarket coolers like the CM Hyper 212+. So if you aren’t looking to overclock heavily, knock thirty bucks off the cost.

And finally, the new 400-series motherboard chipsets are finally mature, and no longer super picky about RAM.

If I were purchasing today, I would go with Zen+ over Intel Coffee Lake. Screw that Meltdown noise.

I’m generally happy with my Ryzen 5 1600 I bought to be my main Plex server.

The gaming benchmarks in the Anandtech review are crazy, showing Ryzen 2700x winning every game against i7-8700k. Usually by pretty significant margins.

https://www.anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600/15

Other reviewers aren’t showing that (even when testing the same game on the same preset), so it’s pretty mysterious. Since it’s just AT showing these results, odds should be that something went wrong with their testing. On the other hand, AT were running with the latest Spectre and Meltdown patches while some other review sites apparently weren’t. So probably worth waiting a few days just to see where the dust settles.

Given that even AMD only projects a 10% overall performance gain over OG Ryzen, those numbers do seem unlikely.

I wish these hardware review sites would toss Planet Coaster into the benchmark mix. That’s the one game that can bring my i7 4770k to its knees. Wondering if more cores would help.

I think this is because Anandtech seems to be pairing each CPU with it’s highest officially supported RAM frequency. So the 7700 gets 2400 MHz ram, the 8700 gets 2666 MHz, and the 2000 series Ryzen chips get 2933 MHz. Most other reviews tend to test everything with equally fast memory.

I watched a side-by-side video of the RyzenX vs. an 8-gen i7 in ArmA3. The i7 still had a significant, but not obscene, lead. Then again A3 is its own thing.

Is AMD chipset weirdness a thing of the past?