AMD Ryzen 7 1800X vs. Intel Core i7 7700K Linux Gaming Performance

[nuetzel]

The BIOS version not being listed is due to a shortcoming of Linux. I was using for this board its version 117 [what arrived on the board to me was 115], which was supplied to me by AMD yesterday as the latest; at least as of yesterday, no BIOS versions were listed on the MSI website for this board.

But beyond that, the BIOS version isn't exposed (maybe via dmidecode when running as root, haven't checked in this particular instance) and thus isn't something PTS can parse and report to the benchmarking results. If there happens to be some interface available to reliably read a board's BIOS version, happy to add support to PTS for it.

What? ;-)

Parse dmesg.

Example:
[ 0.000000] SMBIOS 2.6 present.
[ 0.000000] DMI: FUJITSU PRIMERGY TX150 S7 /D2759, BIOS 6.00 Rev. 1.19.2759.A1 09/26/2012

Greetings,
Dieter

[torsionbar28]

Funny... Although given the manpower available at AMD it might be depressingly accurate.

I think it is accurate, although it's nothing to do with AMD's manpower. Think about it, the heavy corporate Linux contributors are all server vendors. They don't give a crap about Linux on the desktop (aside from professional workstations which also use server chips), and they certainly don't give a crap about Linux desktop gaming. I'm talking about IBM, HP, Red Hat, SuSe, etc. Unfortunately for us desktop users, we may have to wait for the Zen server chips to get closer to market, before we see the big players spin up and contribute code.

[dungeon]

After looking at many reviews, Windows doesn't have this problem so this issue is Linux specific.

Previously i make you spot Metro 1080p on Windows.

And now i present CS:GO on Windows from LinusTechTips

Put this as wallpaper if you think that Windows does not have this problem




Here is "our Linux-specific" 60% alcohol shows up Which might be also some nVintel compiler crap actually

[GruenSein]

Lets's wait and see how things look one or two bios revisions down the road with the memory timings figured out. That's a huge question mark at the moment.

Also, since Ryzen is close to Intel in IPC for multi-threaded applications, I suspect something going on with the boost frequencies. In massively parallel code, there is not much boosting and shifting frequencies. The CPU settles at its highest sustainable clock and that's it. In games, loads change all the time (see how much the freq governor already influences the result?). Intel has made lots of improvements to its boost technology in terms of how quickly it shifts from estate to state. Maybe, AMDs boost technology just can't respond to high CPU demand spikes as quickly. It's not always just about the bare frequency numbers.

anyways, until the memory issue is figured out, this is all just speculation

[torsionbar28]

The Ryzen CPU's also appear to be poor overclockers with most topping out BELOW 4.1ghz which is quite low, most intel cpu's can handle 4.4ghz without much sweat. Ryzen performs best doing workstation intensive work such as rendering etc... where its 16threads can be put to use. I suspected this might end up being the case.

Of course an 8 core 16 thread chip is going to excel at multi-threaded tasks like rendering, and be not quite as great at single threaded. It's not about the Ghz. After all, an FX9590 can do 4.7 Ghz base 5.0 Ghz turbo and that's stock. But Ryzen kicks the FX9590's butt, even though Ryzen has lower clocks and much lower TDP. The impressive thing here, is that in a number of benchmarks, Ryzen manages to beat intel in both single-threaded and multi-threaded. AMD has a winner on its hands here.

[finalzone]

Not quite as good as the 7700k. But not as bad as we see here.

Then i7 6900K, the competitor for Ryzen 1800X, is not as good as i7 7700K.

[dungeon]

After looking at many reviews, Windows doesn't have this problem so this issue is Linux specific.

Boring, maybe we are reading entirely different reviews....i watched now this third one from Gamers Nexus and same story, enough to me... could you provide anything that back up your words on reading many reviwes and to get that conclustion? As everything what i watched speak different story

[sarfarazahmad]

tomshardware is always thorough !

[L_A_G]

Seems like it's much the same case as with game benchmarks under Windows...

Still, I wonder how much of this is up to memory latency? Because I remember one site doing some rather fine grained synthetic benchmarking that included per level memory latency benchmarks from L1 cache to RAM and finding that Ryzen suffered from pretty terrible memory latency once you missed the L3 cache (where latency wasn't all that stellar either) and had to retrieve stuff from RAM. Throughput was just fine (on par with the 7700K), but latency (which is what really matters to a CPU doing real time work like running a game) seemed to have been something of an Achilles heel.

An interesting factor is also that a number of other sites have discovered that if you disable SMT you actually get a decent boost in performance across the board in games. I haven't seen anyone test memory latency with SMT disabled, but it wouldn't surprise me if this reduction in stress on the MMU leads to a considerable improvement in memory latency. This could however be worth testing, so if you have the time Michael , I'd recommend doing it.

If it really is RAM latency I wonder if it can be fixed or at least mitigated with BIOS or microcode updates (apparently microcode fixes are coming and the microcode was apparently holding up the release) or if it's a hardware problem that can't fixed with software. When it comes to potential hardware problems two things spring to mind, the first is the MMU not being able to handle the workload when the data being processed is spread all over RAM, causing excessive cache misses, and the second is that it's a design compromise made because of AMD planning to use the same AM4 socket for APUs (which have GPUs that are much more sensitive to bad throughput than bad latency) as well.

Still, I'm not canceling my order as playing games is of secondary importance to me. As long as I get decent performance at 1080 or 1440p with high settings I'm satisfied.

Edit: Found the benchmark and it turns out that Ryzen actually does worse in RAM latency than their previous chips:
http://hexus.net/tech/reviews/cpu/10...nm-zen/?page=7

[mitch074]

It's not the first time a brand new chip needs tinkering with its firmware - personally I'll wait for later revisions of the core (my Haswell CPU is still working well), but my next rig will probably be a 6- or 8-core AMD chip, except if Intel decide to drop the price on their 8-core i7's - something I doubt will happen. Intel are much more likely to enable HT on i5s and drop the price of that generation - it only requires a firmware update and they can spin it as a new chip for that reason alone. After all, they just did that with the Pentium G4600 which is basically an i3 with both AVX and 1Mb of cache disabled.