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

[Chief Blur Buster]

Sources are all linked into the pictures.

Excellent contribution to the discussion. Thanks!

I like this latency heatmap. The latency patterns are quite revealing!

[dervu]

Ryzen 5600X having ~63ns in AIDA64 with 3200 with Corsair CMW16GX4M2C3200C16 (16-18-18-36).
Measuring in mlc (even with newer 3.9) version I get worse result than on old 2600 Ryzen. Maybe it is not good tool to measure it on Ryzen.

can you post 5600X MLC screenshot for research info and aida screenshot aswell to compare. ty

https://i.imgur.com/FBzchMs.png
https://i.imgur.com/eRDBYb5.png

[Bloods]

Can anyone put it in laymen terms how much difference this makes? Going from 40 ns memory latency to 60, how much extra MS will that add in total input lag? If both CPU's have identical FPS ingame etc.

[schizobeyondpills]

Can anyone put it in laymen terms how much difference this makes? Going from 40 ns memory latency to 60, how much extra MS will that add in total input lag? If both CPU's have identical FPS ingame etc.

please stop thinking about input lag only place where latency is a problem.

its a continuous stream of frames at all times the game is running that is affected by latency. not a single key or click to photon. latency is what delays the frames to screen.

if you do an input or dont the game is still rendering frames and so both non input and input are affected.

a billion transistors working at sub nanoseconds cant be put in laymen terms unless you are paying someone 5 figures to do the thinking for you.

*** EDIT from Moderator to add Moderation Note: This reply is not constructive. At Blur Busters we encourage people to ask questions, and get friendly answers. Please reply more constructively.

[Simon95]

My Program disabled DWM in WIndows 10 up until v2009, as the time of writing. Its still not released yet but it will be soon. Search for GPU Booster or n1kobg.

Do you release a demo version too?

[amezibra]

Can anyone put it in laymen terms how much difference this makes? Going from 40 ns memory latency to 60, how much extra MS will that add in total input lag? If both CPU's have identical FPS ingame etc.

while those 60ns to 40ns seems like nothing , a 5Ghz cpu each cycle last 0.2ns

this means when RAM latency is 60ns your CPU is wasting 100 additional cycles (300 cycles) than at 40ns (200 cycles) waiting for the ram to reply (or 33% of its time doing nothing)

that's why you get huge fps and frametime boost by lowering ram latency especially in CPU bound scenario (1080p gaming)

https://developers.redhat.com/blog/2016 ... th-caches/

[Bloods]

while those 60ns to 40ns seems like nothing , a 5Ghz cpu each cycle last 0.2ns

this means when RAM latency is 60ns your CPU is wasting 10 more cycles than at 40ns waiting for the ram to reply (or 33% of its time doing nothing)

that's why you get huge fps and frametime boost by lowering ram latency especially in CPU bound scenario (1080p gaming)

I understand that it's a big bottleneck looking at it directly but once again how does this translate in total time(ms?) added to input lag ?

For example I typically spend 90% of my gaming sessions playing competitive games like CS:GO / Valorant(Currently) which are heavily CPU bound. Even with my i5-3570k at stock speeds and GTX 680 I average around 130 fps on 1080p / LOW in Valorant, slightly better in CS:GO, this is with a comp from 2013.

I want to buy a 360hz monitor and will need a whole new setup to get enough frames to benefit from it, the new 5600x offers 30% more FPS in games like Valorant/CS:GO over the 10900K, yet it has higher RAM latency?? This is pretty nice considering you never want to dip below 360fps and having a bigger buffer available that can keep it above 360fps in demanding situations like multiple abilities, smokes being used at the same time is great.

Which is why I am trying to figure out how much it would actually affect me in game.

[Chief Blur Buster]

Casual “simple barely surface-scratch” answer:

It is correct that input lag is a misnomer in some ways here — latency is a horridly complicated topic and it is easy to make mistaken assumptions especially when you are not an engineer. Sometimes those nanoseconds are super important (i.e. LIDAR) and other times less so.



But it gets WAY more complex than this.

Here is a very brief surface scratching:

— There are multiple different kinds of latency, such as latency jitter (variable latency), absolute latency (tapedelay-style latency), sampling latency (higher Hz at same absolute lag feels less laggy), and they can hit different chains.

— One lag can appear while another lag disappears, for lag that is parallelized, internally in CPU, GPU or the interaction between CPU and GPU. For example SLI is higher frame rates but laggier frames. Likewise, a higher frame rate might also occur simultaneously with more lagged USB processing or lagged main RAM processing or lagged asset streaming processing. What might happen is, for example, you get a higher frame rate, but a slightly laggier mouse, or such — many possible scenarios.

— Etc. There are literally infinite number of causes of latencies — whether a line of computer code, a transistor on a CPU by themseles, or the new latencies generated by multiples of them cooperating with each other. And there are millions of lines of codes — and there are billions of transistors — and literally infinite combinations for latency to go wrong. I know I’m absurdly oversimplifying things, but it helps gain an appreciation.

Going into further detail about nuances of latency ends up going into engineer-speak, but this reply is to allow one to gain an appreciation of the complexity of the latency subject. For those who are so fascinated about this, should consider looking into the appropriate education and work fields.

Needless to say, it is not a simple yes/no answer about Ryzen vs Intel; each is superior at different things depending on the software and what is being done in the hardware/software combinations — there is more lags and less lags fighting against each other throughout the whole machination of a system.

I know some people went to university because of Blur Busters reading, to learn more about electronics, computing, engineering, because they were so fascinated. This topic is one of these; it is definitely very hard to distill in laymen’s terms but questions are always welcome even if not easily answered in forum posts.

Even a game may be superior on Intel, and a different game may be superior on Ryzen, simply because of how they optimize to the respective virtues of the different architectures. Some gamers prioritize on higher smooth sampling latencies rather than absolute latencies, while others want the lowest absolute latency.

We all know how deliciously massively multicore a Threadripper is, and they are really great for those workloads. When used with certain heavily multithreaded games or other software, those intercore latencies can actually perform worse/better (lag-wise or performance-wise, etc) depending on how you force thread affinity (e.g. if a game uses 2 or 3 or 4 threads, they should be thread-affinity allocated in adjacent cores on the surface of a Threadripper to keep lag lower than if they were spread throughout the Threadripper. Some optimizations like these make big differences for some multithreaded processing that utilize frequent inter-thread communications; but not all software need frequent inter-thread communications. But I’m cherrypicking “one nuance”.

As CPUs get more massively abstracted (pipelined, multicore, virtualized, hyperthreaded, etc) we’re getting lots and lots more untraceable latency noise. Ever tinier buildups of imprecision (whether latency, jitter or otherwise) in the refresh rate race to retina refresh rates when sub-millisecond latencies slowly start to show human-visible jittering as the Vicious Cycle Effect (higher rez, higher Hz, clearer motion) unveil the curtain and lower noisefloors, Rube Goldberging its way to the The Amazing Human Visible Milliseconds. Cross-core latencies are way too diffuse to be worried about but in the picture of trying to simultaneously process an 8000Hz device, run dozens of threads, and a game engine, on various random cores, we get those intercore latency in the noise. How it bornes out into human degradations requires a lot more research that is far more specialized than what Blur Busters does.

I know this does not answer your question, and I know that the Ryzen vs Intel debate will continue. But it brings an appreciation to the delicious complexity of the latency subject.

[howiec]

https://developers.redhat.com/blog/2016 ... th-caches/

But it gets WAY more complex than this.

Yes, people need to realize that our systems (and how humans perceive things) are so much more complex than what the layman is aware of.

People really need to have a background in science/engineering or have self-taught themselves well about these topics beyond just the surface before making 1-dimensional "opinions & claims w/o data" - on the other hand, questions are always welcome of course.

[howiec]

Inter-ccd communication is still very slow. But honestly, the core to core latency within the ccd looks quite nice, so one ccd processors like the 5800x might not be too bad. Memory latency is still pretty bad compared to intel, lowest I saw was 48ns (aida).

Yeah, it's great that AMD acknowledges the need to improve in this area and are doing so.

I'm tempted to get a 5950X as an interim system for a RTX 3090 until Intel can do something better but I'm wary of the 2 CCXs, although technically you could carefully assign core affinities.