Is it true Ryzen has higher input lag than Intel? Is there conclusive data to prove this?

[orangetuner]

How does AMD's infinity fabric play into this? Can it close the gap? Because I want to build a new pc and this isnt the only forum where ive seen people say they feel less responsive on Ryzen.

If you care about having the highest 0.1% lows and rawest mouse input possible in high fps games then you should only buy Intel.

How much of a difference in milliseconds are there between Ryzen and Intel?

[deama]

I'd be interested in an actual input latency test between similiar rigs but one using intel and the other amd.
They'd also have to be the same Ghz as I've noticed from my own tests that higher Ghz affects input lag, at least on my ryzen rig.

EDIT: Ah I forgot to mention. I did a test where I disabled SMT on my CPU and saw a 1 frame latency reduction (4.16ms), using the intel latency measuring tool, it gave me 180-ish nanosecond latency, but with SMT enabled it gave me 340ns. Normally an intel chip, with hyperthreading disabled, would be around 80 nanosecond latency (got this from another user who posted his benchmark here). If we assume it's about 80, that's about 100ns reduction from the ryzen one (mine at least), then that would mean that it should be about 2ms difference or so, because I saw a difference of 4.16ms from 340ns down to 180ns, that's a reduction of 160ns, so 160ns reduction will give us 4.16ms reduction in input lag. The difference between 180ns and 80ns is 100ns, so it should give us a reduction of 2-3ms in input lag.

This is all just speculative as I haven't noticed a reduction in some other latency figures shown further down below in the tool when SMT was disabled, but at the very least a user with an intel system instead of ryzen should see a 2-3ms difference in input lag.

[schizobeyondpills]

I'd be interested in an actual input latency test between similiar rigs but one using intel and the other amd.
They'd also have to be the same Ghz as I've noticed from my own tests that higher Ghz affects input lag, at least on my ryzen rig.

EDIT: Ah I forgot to mention. I did a test where I disabled SMT on my CPU and saw a 1 frame latency reduction (4.16ms), using the intel latency measuring tool, it gave me 180-ish nanosecond latency, but with SMT enabled it gave me 340ns. Normally an intel chip, with hyperthreading disabled, would be around 80 nanosecond latency (got this from another user who posted his benchmark here). If we assume it's about 80, that's about 100ns reduction from the ryzen one (mine at least), then that would mean that it should be about 2ms difference or so, because I saw a difference of 4.16ms from 340ns down to 180ns, that's a reduction of 160ns, so 160ns reduction will give us 4.16ms reduction in input lag. The difference between 180ns and 80ns is 100ns, so it should give us a reduction of 2-3ms in input lag.

This is all just speculative as I haven't noticed a reduction in some other latency figures shown further down below in the tool when SMT was disabled, but at the very least a user with an intel system instead of ryzen should see a 2-3ms difference in input lag.

>They'd also have to be the same Ghz as I've noticed from my own tests that higher Ghz affects input lag, at least on my ryzen rig.

why would I willingly give up 1Ghz of clock (5GHz) on my Intel? Its not Intel's problem AMD cant go higher than 4GHz. If you want to go fast, you have to run fast, not cry "please wait" "downclock yourself for me to keep up".

benching CPUs of same GHz while one can do 1GHz more than another is like limiting the speed limit to 10km/h and then comparing a battery scooter vs F1 race car and saying they're the same. yeah, they are because you nerfed one 50 times lower than what it can do.

>Normally an intel chip, with hyperthreading disabled, would be around 80 nanosecond latency.

wrong. Games dont saturate 100% of your RAM sequential read/write bandwidth or load, nowhere close, games are most dependant on RANDOM ACCESS, not sequential. meaning you need to optimize for ~ 1gb/s-3gb/s saturation(its prob even lower than that) of RAM load specifically for single read/write operation, because thats what games do most, random pointer chasing, game world -> player collection -> player object -> player information -> player position -> player information -> gun information .

Everything else you said below is wrong because of how RAM latency scales with frequency

1. higher CPU freq is more impacted by ram latency(5ghz is 0.2ns per clock - 1 clock is enough to issue a mem load - 50ns is like 250 clock cycles, so 1:250 ratio) - meaning, 4GHz is 0.25ns per clock - 1 clock load issued - 50ns ends up being 1:200 ratio)
What this means is that lower frequency CPU clock has less cycles wasted since they dont tick in that time-slice bottlenecked by fetch delay. in this napkin tier example those 50 ticks (250-200) never even tick on 4GHz CPU.
2. since request:load clock ratio is so high the scaling is *** NOT *** LINEAR. its chrono-exponentional. meaning, 50ns-40ns performance gains are around 10% perf increase. but so is 40ns-35ns. 35-30ns is ~ 20% increase.

the only way to solve input lag on AMD setup is to sell it and buy Intel with focus on higher frequencies and good low timings RAM
and yes, I know someone is already ready to send me some FPS benchmarks, which fall apart the moment u would paste the link, why? because those tests are done on 3200MHz CL16 RAM, with 3 prerendered frames as default setting

[Happyalive]

Yes, intel's 11th gen being chiplet is set in stone, and no you can't make chiplet as low latency as monodie. Intels 11th gen also probably wont support w7 either so yeah, just get a high tier 10th gen rig and use it for the next 5 years before going linux or quitting pc.

Why would you test intel vs amd at the same frequencyt? That's so fucking useless, because in reality intel can easily get 600 mhz higher. It's like deciding you have to test intel vs amd @ 3200 cl16 xmp. Also intel is more than 2-3ms lower latency than amd, and way smoother as well. Not even comparable.

[deama]

I'd be interested in an actual input latency test between similiar rigs but one using intel and the other amd.
They'd also have to be the same Ghz as I've noticed from my own tests that higher Ghz affects input lag, at least on my ryzen rig.

EDIT: Ah I forgot to mention. I did a test where I disabled SMT on my CPU and saw a 1 frame latency reduction (4.16ms), using the intel latency measuring tool, it gave me 180-ish nanosecond latency, but with SMT enabled it gave me 340ns. Normally an intel chip, with hyperthreading disabled, would be around 80 nanosecond latency (got this from another user who posted his benchmark here). If we assume it's about 80, that's about 100ns reduction from the ryzen one (mine at least), then that would mean that it should be about 2ms difference or so, because I saw a difference of 4.16ms from 340ns down to 180ns, that's a reduction of 160ns, so 160ns reduction will give us 4.16ms reduction in input lag. The difference between 180ns and 80ns is 100ns, so it should give us a reduction of 2-3ms in input lag.

This is all just speculative as I haven't noticed a reduction in some other latency figures shown further down below in the tool when SMT was disabled, but at the very least a user with an intel system instead of ryzen should see a 2-3ms difference in input lag.

>They'd also have to be the same Ghz as I've noticed from my own tests that higher Ghz affects input lag, at least on my ryzen rig.

why would I willingly give up 1Ghz of clock (5GHz) on my Intel? Its not Intel's problem AMD cant go higher than 4GHz. If you want to go fast, you have to run fast, not cry "please wait" "downclock yourself for me to keep up".

benching CPUs of same GHz while one can do 1GHz more than another is like limiting the speed limit to 10km/h and then comparing a battery scooter vs F1 race car and saying they're the same. yeah, they are because you nerfed one 50 times lower than what it can do.

>Normally an intel chip, with hyperthreading disabled, would be around 80 nanosecond latency.

wrong. Games dont saturate 100% of your RAM sequential read/write bandwidth or load, nowhere close, games are most dependant on RANDOM ACCESS, not sequential. meaning you need to optimize for ~ 1gb/s-3gb/s saturation(its prob even lower than that) of RAM load specifically for single read/write operation, because thats what games do most, random pointer chasing, game world -> player collection -> player object -> player information -> player position -> player information -> gun information .

Everything else you said below is wrong because of how RAM latency scales with frequency

1. higher CPU freq is more impacted by ram latency(5ghz is 0.2ns per clock - 1 clock is enough to issue a mem load - 50ns is like 250 clock cycles, so 1:250 ratio) - meaning, 4GHz is 0.25ns per clock - 1 clock load issued - 50ns ends up being 1:200 ratio)
What this means is that lower frequency CPU clock has less cycles wasted since they dont tick in that time-slice bottlenecked by fetch delay. in this napkin tier example those 50 ticks (250-200) never even tick on 4GHz CPU.
2. since request:load clock ratio is so high the scaling is *** NOT *** LINEAR. its chrono-exponentional. meaning, 50ns-40ns performance gains are around 10% perf increase. but so is 40ns-35ns. 35-30ns is ~ 20% increase.

the only way to solve input lag on AMD setup is to sell it and buy Intel with focus on higher frequencies and good low timings RAM
and yes, I know someone is already ready to send me some FPS benchmarks, which fall apart the moment u would paste the link, why? because those tests are done on 3200MHz CL16 RAM, with 3 prerendered frames as default setting

Cause I donno if ghz = input lag phenomenon is related to just amd, or to all CPUs.

why would I willingly give up 1Ghz of clock (5GHz) on my Intel? Its not Intel's problem AMD cant go higher than 4GHz. If you want to go fast, you have to run fast, not cry "please wait" "downclock yourself for me to keep up".

So the tests are valid?

wrong. Games dont saturate 100% of your RAM sequential read/write bandwidth or load, nowhere close, games are most dependant on RANDOM ACCESS, not sequential. meaning you need to optimize for ~ 1gb/s-3gb/s saturation(its prob even lower than that) of RAM load specifically for single read/write operation, because thats what games do most, random pointer chasing, game world -> player collection -> player object -> player information -> player position -> player information -> gun information .

Well, when I disabled SMT it reduced input lag, so what are you trying to say? I shouldn't believe my own eyes?

[deama]

Yes, intel's 11th gen being chiplet is set in stone, and no you can't make chiplet as low latency as monodie. Intels 11th gen also probably wont support w7 either so yeah, just get a high tier 10th gen rig and use it for the next 5 years before going linux or quitting pc.

Actually I don't think intel has supported windows 7 since 6th gen or so, unless you made specific changes to windows 7 to get it to work.

Why would you test intel vs amd at the same frequencyt? That's so fucking useless, because in reality intel can easily get 600 mhz higher. It's like deciding you have to test intel vs amd @ 3200 cl16 xmp. Also intel is more than 2-3ms lower latency than amd, and way smoother as well. Not even comparable.

I'm not really interested in "It felt smoother and more responsive", I'm interested in hard numbers, how much ms lower than amd?

[schizobeyondpills]

its rude and dishonest to simply respond with automated answer based on how you feel.
Higher clock impacts input lag because ryzen has ram latency floor at 80ns while intel has at 50ns.
So yes, pushing 4ghz into 4.5ghz on AMD makes it worse because of 80ns ram latency.

the tests would be just as valid, you are comparing two different producs inside the same set. if you adjust their properties to be same then its not a valid test. if one can do 5Ghz while other struggles with 4.5Ghz its a valid test because those are their individual performance properties.

I am saying that u should know what you are talking about before making claims and atleast know which number u should look at in intel memory latency checker tool.

[Happyalive]

>Well, when I disabled SMT it reduced input lag, so what are you trying to say? I shouldn't believe my own eyes?

>I'm not really interested in "It felt smoother and more responsive", I'm interested in hard numbers

Pick one. The MOMENT you start saying "lol prove it" about things are are inductively true you no longer have the right to ever use your subjective experience as a relevant point of proof ever again, period.

You misunderstand some things about intel vs amd. Higher frequency = less input lag on EVERY SINGLE CPU PLATFORM EVER CREATED, and it will continue to be that way for the rest of human existence. If anything, difference is bigger on intel beacuse they don't have dogshit memory latency, and thus less cycles are wasted waiting around.

I don't care about intels "official" support, I care about real world compatability. In reality, pre z490 compat was fine, yeah you needed usb drivers or w/e but who cares. Some z490 mobos just straight up won't work with w7 afaik, and 11th gen will probably be even worse in that regard. Intel and microsoft are friendly companies, to the benefit of both, there is no reason for intel to not phase out w7 support over time, since microsoft wants userbase all on datafarm os.

[deama]

>Well, when I disabled SMT it reduced input lag, so what are you trying to say? I shouldn't believe my own eyes?

>I'm not really interested in "It felt smoother and more responsive", I'm interested in hard numbers

Pick one. The MOMENT you start saying "lol prove it" about things are are inductively true you no longer have the right to ever use your subjective experience as a relevant point of proof ever again, period.

I measured mine using a high speed camera.

You misunderstand some things about intel vs amd. Higher frequency = less input lag on EVERY SINGLE CPU PLATFORM EVER CREATED, and it will continue to be that way for the rest of human existence. If anything, difference is bigger on intel beacuse they don't have dogshit memory latency, and thus less cycles are wasted waiting around.

Fair enough, I thought it would have capped out at 2 or 3Ghz or something.

[Happyalive]

https://youtu.be/dsbVSknUK7I?t=250 Not the gretest test since I really doubt he has optimized hw / windows but it proves ryzen has more than intel pretty conclusively. Ryzen will only be viable once they cut at least 15ns off their memory latency and go to 8c per ccx, and at that point they would still need a very solid frametime advantage over intel @ w/e price point.