Blur Busters Forums

I've tested CS:GO on both Intel and AMD and there is a difference, indeed, but rather in mouse movement itself than in input-lag. It was more like comparing mice with different sensors.

On my Ryzen system I had to mod BIOS file to change some settings which are unavailable (not shown) by default. For example, when I have "XHCI Gen1" and "XHCI Gen2" both on "Auto" (which usually means "Enabled") my mouse movement feels like shit, but when I force one of them to disabled -- mouse movement immediately becomes a lot better. Also there are some hidden power-saving options turned on by default (on overclocker's motheboard, wtf?) like "Chipset Power Saving Features", "C1E Support", "PCIe Speed Power Policy", "C6 Mode" and so on.

ffs_ wrote: ↑

24 Jul 2020, 01:43

I've tested CS:GO on both Intel and AMD and there is a difference, indeed, but rather in mouse movement itself than in input-lag. It was more like comparing mice with different sensors.

On my Ryzen system I had to mod BIOS file to change some settings which are unavailable (not shown) by default. For example, when I have "XHCI Gen1" and "XHCI Gen2" both on "Auto" (which usually means "Enabled") my mouse movement feels like shit, but when I force one of them to disabled -- mouse movement immediately becomes a lot better. Also there are some hidden power-saving options turned on by default (on overclocker's motheboard, wtf?) like "Chipset Power Saving Features", "C1E Support", "PCIe Speed Power Policy", "C6 Mode" and so on.

Any idea where I could learn how to mod my bios? I have the MSI B450 Gaming Plus which doesn't really feature many settings.

senny22 wrote: ↑

24 Jul 2020, 12:38

Any idea where I could learn how to mod my bios? I have the MSI B450 Gaming Plus which doesn't really feature many settings.

https://www.overclock.net/forum/11-amd- ... ectly.html

Friard wrote: ↑

23 Jul 2020, 15:43

https://youtu.be/3UhaZ8VT4fU

It's certain that AMD has higher input delay than Intel. In the video amd has 18% higher delay in fortnite (18.3/15.5) and 34% higher in csgo (16.3/12.2). However im not sure how reliable those tests are.

Firstly, as alot of people already pointed out, mobos usually have lots of power saving options enabled and hidden. You need a custom bios to disable them. So we don't know which mobo had more power saving which had less in those tests.

Secondly, he caps the fps. Which makes sense since he doesn't want to include input delay reduction from higher fps. However higher fps (in lower/competitive settings) is a big selling point for intel. So i feel its unfair that he didn't include uncapped results.



Thirdly, pairing a 10900k with 3600mhz cl16 is kinda dumb. On ryzen it makes sense since if you go above 3766 mhz (i believe) you suffer from a massive memory latency penalty. A more realistic situation would be 10900k+4400mhz cl19. Not to mention that if u manually tune the timings u can get 35ns memory latency quite easily on an intel rig and I've seen a fair amount of 31-32ns as well. However I've never seen amd system go below 60ns (usually its above 70ns).

It should be worth mentioning that amd 4000 series looks better than 3000 series in terms of memory latency, but its still nowhere near what you can get from an Intel rig.



The source are linked in the images themselves, click on them to visit.

Has someone done a test of memory speed / memory latency and input lag?

MaxTendency wrote: ↑

24 Jul 2020, 18:07

On ryzen it makes sense since if you go above 3766 mhz (i believe) you suffer from a massive memory latency penalty.

Would you elaborate? I use 3866 CL16 with Ryzen 3600x, perhaps I should switch to 3733, hmm...

MaxTendency wrote: ↑

24 Jul 2020, 18:07

However I've never seen amd system go below 60ns (usually its above 70ns).

Yes, though ~65 ns would be not too hard to achieve with good memory.

MaxTendency wrote: ↑

24 Jul 2020, 18:07

It should be worth mentioning that amd 4000 series looks better than 3000 series in terms of memory latency

I think it's mostly because 4333 CL14 have been used, although L1 cache speed was increased by ~25% as well.

MaxTendency wrote: ↑

24 Jul 2020, 18:07

It should be worth mentioning that amd 4000 series looks better than 3000 series in terms of memory latency

I've just noticed that 4000 series CPU from the screenshot you posted is Renoir, i.e. APU based on Zen 2, not Zen 3.

ffs_ wrote: ↑

25 Jul 2020, 01:58

Would you elaborate? I use 3866 CL16 with Ryzen 3600x, perhaps I should switch to 3733, hmm...

Above 3733mhz the MCLK and FCLK de-synchronizes into a 2:1 ratio causing memory latency to go up.

ffs_ wrote: ↑

25 Jul 2020, 03:11

I've just noticed that 4000 series CPU from the screenshot you posted is Renoir, i.e. APU based on Zen 2, not Zen 3.

That's true, however unlike the 3000 series, Renoir is a monolithic cpu not a chiplet. I don't think I've seen any zen 3 benchmarks yet.

ffs_ wrote: ↑

25 Jul 2020, 01:58

Yes, though ~65 ns would be not too hard to achieve with good memory.

65ns is doable. Still far from Intel.

ffs_ wrote: ↑

25 Jul 2020, 01:58

I think it's mostly because 4333 CL14 have been used.

I don't think its the memory frequency but rather the fact that MCLK and FCLK are synchronized in a 1:1 ratio that's causing the lower memory latency.

MaxTendency wrote: ↑

25 Jul 2020, 20:18

Above 3733mhz the MCLK and FCLK de-synchronizes into a 2:1 ratio causing memory latency to go up.

This is silicon lottery. You can use 1:1 with any IF frequency, but most CPUs will run at 1866, some at 1900 and only few at 1933.

MaxTendency wrote: ↑

25 Jul 2020, 20:18

That's true, however unlike the 3000 series, Renoir is a monolithic cpu not a chiplet.

3300X is also monolithic, but latency difference is probably 4-5 ns (lower). I think it's about architecture itself since Zen 2 has high latency (higher than Zen+, for example). Perhaps we will see memory improvements/optimizations in Zen 3 so it could benefit from using monolithic format.

MaxTendency wrote: ↑

25 Jul 2020, 20:18

I don't think its the memory frequency but rather the fact that MCLK and FCLK are synchronized in a 1:1 ratio that's causing the lower memory latency.

I still doubt you will get such result with 4333 CL19 even at 1:1.