Core i7 8700K vs. Ryzen 7 1800X For NVIDIA/Radeon Linux Gaming

[Spooktra]

The problem with tests like these is that it ignores one simple thing, these processors, both AMD's and Intel's high core count cpu's are not designed to run a single application at a time, no matter if it's well threaded or not. They have tremendous throughput and even if it looks like a single app is loading up all the cores within the cores there are still many resources not being used.

It's like being handed the keys to a diesel Nissan Titan or Ford F250 and choosing to test 0-60 and 1/4 mile times; yes they may be relatively quick, the test may be fun and of some value for comparison sale but it completely ignores that they are made to pull and or carry very heavy loads for extended periods of time across many grades of road inclination and declination and many different weather conditions.

In short these types of gaming tests and next to useless and this goes for every web site that uses them, which means every web site out there.

[LinuxID10T]

Sigh, another case of Ryzen inter-ccx latency problem (unless Michael ran the 1800x with low speed DDR4 which could also be a problem. The review states that 3200 ram was used but that doesn't mean it was running at that speed).
This is seriously harming AMD's brand under Linux

It's a pity since all it generally takes to solve is a simple taskset -c 0-7 %command% in Steam.
I'm especially disappointed by Feral : Dawn of War 3 was released well after Ryzen.
They should have known by then that scheduling is really important with Ryzen.

The 8700k has a 4.3ghz all core frequency and a 7 to 10 % IPC advantage.
So it should be around 4.3/3.7 * 1.1 ~= 30% faster in games (because no game use more than 6 core / 12 threads)
It could reach 40% in extreme cases but not more.
That's why these benchmarks with a ludicrous 70% lead for the 8700k are not representative of the performance that Ryzen is capable of.
Unfortunately for AMD though, they are representative of the out of the box experience

I've done the tests on this. The reduction in CPU usage from other threads offsets the inter-CCX latency. Plus, Linux is MUCH better at keeping threads on the same cores than Windows.

[LinuxID10T]

Michael

Thanks for add 1080p tests

See results intel is 1080p* game king

*Without forget 1080p is more desktop used resolution, in many cases at 144hz

Hopefully amd can improve for ryzen 12nm



Yeah for disgrace amd single thread is too low compared with intel**, especially with coffelake*** higher frecuencies

**Sadly ryzen oc have similar single thread of haswell at 4.2 - 4.4ghz but haswell 4.2 and upper stay avalaible since 2014

***Various models can stay around 5.0ghz but with delid have better chances

This situation appears in dolphin benchmark and others for courtesy of anandtech

The Ryzen IPC is certainly better than Haswell. Normally it ends up sitting between Broadwell and Skylake. That being said, it highly depends on the task involved. Some tasks that run in the L1 cache can run far faster than even Sky/Kaby/Coffeelake. It is a much wider architecture and is capable of more IPC than ever Intel processor ever made. The best way to see this effect is to do IPC calculations for something like C-Ray The problem is feeding the beast and AMD has a real bottleneck when it comes to RAM memory latency which heavily affects gaming performance.

[theriddick]

Yeah looks like games in general under Linux are not really trying that hard to thread the work load around if at all. Nothing new there, not sure if updated Ryzen cpu firmware or something can help in that respect.

[partizann]

Well at MindBlank Tech youtube channel, they were basically able to reach i7-7700k@5GHz levels of performance from [email protected] or even top it when using high frequency memory with very low latency - at 3466MT/s and CL 14 (both CPUs were using same rams and settings).
What it seems to imply, is that ryzen has more of a latency problem in games than a IPC deficit.

[mlau]

The Ryzen IPC is certainly better than Haswell. Normally it ends up sitting between Broadwell and Skylake. That being said, it highly depends on the task involved. Some tasks that run in the L1 cache can run far faster than even Sky/Kaby/Coffeelake. It is a much wider architecture and is capable of more IPC than ever Intel processor ever made. The best way to see this effect is to do IPC calculations for something like C-Ray The problem is feeding the beast and AMD has a real bottleneck when it comes to RAM memory latency which heavily affects gaming performance.

That's it in a nutshell, and Intel's manufacturing advantage, allowing much higher turbo clocks, which makes the *lake cpus appear so much more powerful in select benchmarks.

[humbug]

Michael, was the RAM running at 3200Mhz on the Ryzen system for these tests? It's needed for Ryzen to perform well in games.
Unless you specifically set it in the bios it will run at something lower by default.

At launch people had problems getting RAM to run at this speed but I believe it's better now if you update to the latest BIOS for your respective motherboard.

[oooverclocker]

Firstly, I would like to know more about the RAM speed. I have 3200 MHz Dual Rank Modules and I can't see a difference in the average gaming performance between my Ryzen [email protected] and an i7-7700K.
But when I run the RAM with 2400 MHz Ryzen is a bit slower. In my opinion to measure the CPU performance you should try to eliminate RAM bottlenecks. And the most Ryzen-configurations I have seen on the net have between 2933MHz Dual Rank and 3600MHz Single Rank because people are aware of this massive impact on Ryzen performance. So I usually tend to look at benchmarks with less than 3200MHz RAM clock as being not really helpful.

With reasonable RAM clocks it is unlikely that the difference in some games is that significant.

Secondly, as Nvidia seems to bully the open source community one might think about ignoring their closed source drivers completely and only using Nouveau for Linux benchmarks in the future. I think they should get back the same ignorance they gave us.

[L_A_G]

I expected the 8700 to come out on top, but not by anywhere near this much.

Then again when the R7 series launched it performed rather poorly in a lot of games on Windows as well, but they've mostly fixed that with patches to the games (which also benefit highly multi-threaded Intel parts) and BIOS updates to motherboards. Most linux games being much more single-threaded due to being run trough heavily single-threaded DX11-to-OpenGL translation software may explain this poor performance at least partly. Something that seems to have an unusually large effect on Ryzen CPUs is the RAM timings and clocks. Apparently the inter-CCX communication runs off the same clock source as the RAM.

[Michael]

Firstly, I would like to know more about the RAM speed. I have 3200 MHz Dual Rank Modules and I can't see a difference in the average gaming performance between my Ryzen [email protected] and an i7-7700K.
But when I run the RAM with 2400 MHz Ryzen is a bit slower. In my opinion to measure the CPU performance you should try to eliminate RAM bottlenecks. And the most Ryzen-configurations I have seen on the net have between 2933MHz Dual Rank and 3600MHz Single Rank because people are aware of this massive impact on Ryzen performance. So I usually tend to look at benchmarks with less than 3200MHz RAM clock as being not really helpful.

With reasonable RAM clocks it is unlikely that the difference in some games is that significant.

Secondly, as Nvidia seems to bully the open source community one might think about ignoring their closed source drivers completely and only using Nouveau for Linux benchmarks in the future. I think they should get back the same ignorance they gave us.

It was running DDR4-3200.

And only using the NVIDIA open-source driver is outright silly.