cancel
Showing results for 
Search instead for 
Did you mean: 

Low latency video cards

JonnyCache
Honored Guest
Hey, folks! We know latency is as critical as frame rate for VR, but I'm having a really hard time finding latency benchmarks for consumer video cards. So...

Doesn't anyone have any suggestions for cards or card features that we should keep an eye out for?
Do any cards handle split frame rendering (SFR) exceptionally well?
Does SFR really reduce overall latency or increase it like alternate frame rendering (AFR) does through buffering?
Do single core GPU cards have a lower average latency then multi-core GPUs in the same price range?
How does sli/crossfire effect latency in general?
What are the best solutions for sli/crossfire micro stutter?

Also does any one have any pro tips on reducing latency through tricks like reducing the number of frames your card buffers?

Thanks all!

PS Here's some interesting reading I've been able to dig up so far.
http://en.wikipedia.org/wiki/Scalable_Link_Interface
http://www.tomshardware.com/reviews/nvidias-response-early-diy-quad-sli,1299-3.html
Intel(R) Core(TM) i7-3930K; ASUS Sabertooth X79; NVIDIA GeForce GTX 780 (*2); 32768MB RAM; Window 7 x64 SP1
40 REPLIES 40

KBK
Protege
My vague recollection of it is that we are dealing with a single frame output, with two warp patterns superimposed.

That the given engine is putting out 'side by side', as a single frame.

Thus, any SLI set ups are considered to be more problematic than a single good card.

it all depends on what you are going to be doing with the card. What software is to be rendered.

The preferable is to super sample down to 1280x800, in side by side, and V synced at 60hz. I've heard thoughts on gtx cards being a good choice, in the 650Ti and up range. This will do service for the Tuscany demo, etc. Possibly not so well with high rez stuff brought down to the 1280x800 rez of the panel.

The latency is more in the control of the hardware than the hardware itself. The correct software settings are key to achieving the lowest latency you can with the given hardware in use.

I have not touched PC's in about 7 years (used my brain pan elsewhere), so I'm just getting up to date on all this info myself.
Intelligence... is not inherent - it is a point in understanding. Q: When does a fire become self sustaining?

JonnyCache
Honored Guest
@KBK Thanks! Regarding the GTX thing, I heard something very similar.

It was a podcast with PlamerL where he mentioned that the GTX690 was a better choice for VR then the new Titan card, he also mentioned some stuff on sli's effect on VR/latency (which I don't clearly recall). That's actually what got me curious about what we should be looking for in VR oriented hardware and settings.

But besides getting more knowledgeable on video cards in general, I haven't made much progress in nailing down many facts on achieving minimal latency.

I'm building a new primary development PC this year with a VR focus (my current machine will become a target spec test bed) so latency is a prime factor. I'm excited so I thought I'd share. 🙂
Intel(R) Core(TM) i7-3930K; ASUS Sabertooth X79; NVIDIA GeForce GTX 780 (*2); 32768MB RAM; Window 7 x64 SP1

lordvtp
Protege
There has been allot of good work of late on multi-gpu frame latency issues. 😄

http://techreport.com/review/24553/insi ... ture-tools
http://www.pcper.com/reviews/Graphics-C ... nce-Testin

For the record I am not affiliated with either site.

An interesting possibility for the rift would be a new split frame mode built to divide each perspective to one card and combining the two to create the final side by side output... Also Sli AA modes(up to 128x!) are a heck of thing for dealing with low rez displays with very little latency involved.

lo4d
Explorer
I don't have anything to contribute to the hardware side of things, but in Joe Ludwig's presentation "Lessons Learned Porting Team Fortress 2 to Virtual Reality" (http://media.steampowered.com/apps/valv ... VR_GDC.pdf) he mentioned something I hadn't heard before. It fits the theme in this thread of achieving minimal latency:

The first is that you really want to turn off Aero. The compositing that Windows does in Aero adds at least a frame of latency when you are running in a window.

cybereality
Grand Champion
I've heard that SLI can add latency. For example with AFR, where you can add at least 1 frame of delay. Personally I've been running SLI on my home rig (dual GTX 470) and I haven't perceived much of a problem. I haven't got to test the Rift too much on that rig, so I can't really comment on the VR aspect of latency.

Usually what people say is to get the fastest single-GPU card you can afford, to avoid any of the issues associated with multi-GPU rigs.
AMD Ryzen 7 1800X | MSI X370 Titanium | G.Skill 16GB DDR4 3200 | EVGA SuperNOVA 1000 | Corsair Hydro H110i Gigabyte RX Vega 64 x2 | Samsung 960 Evo M.2 500GB | Seagate FireCuda SSHD 2TB | Phanteks ENTHOO EVOLV

RedWorm
Honored Guest
I think you guys worry too much about latency, EVERYONE says that the oculus latency is good as it is and with time it will become perfect.

Low latency in game comes with low fps
I use a 120HZ monitor and I can say if the game is running above 80 fps, it's always instant. I mean even with 2 MS on my screen, I can feel it. But maybe it's all different when you are in there right infront of your eyes..
I dont know but for me it's all about how much you are willing to spend for a graphic card and not "finding" the right low latency card.

The best thing to wish for is better screen for the oculus.

lordvtp
Protege
The absolute best solution is to somehow get a 120hrz panel for the rift...Not too likely I know.

About the 'fasted single card you can afford'... In practice this works well, except when the current flagship card doesn't meant your performance needs.

Sli does inherently add a degree of latency and it is quite noticeable in 4 way v-synched @ 60 hrz...as your usually rendering 3 frames ahead... So I like to say 3 to 4 way Sli and v-synch don't mix for twitch games. At 120hrz however (or v-synch disabled) this is largely mitigated ... and with v-synch disabled removed (very nearly)entirely. So long as you can push 120fps or greater to the display your golden.

Honestly I wish I could foresee a rift enabled game that would require more then two titans, but unless we get 2560x panel that's just not an issue...yet.

jwilkins
Explorer
"RedWorm" wrote:
I think you guys worry too much about latency, EVERYONE says that the oculus latency is good as it is and with time it will become perfect.

Low latency in game comes with low fps
I use a 120HZ monitor and I can say if the game is running above 80 fps, it's always instant. I mean even with 2 MS on my screen, I can feel it. But maybe it's all different when you are in there right infront of your eyes..
I dont know but for me it's all about how much you are willing to spend for a graphic card and not "finding" the right low latency card.

The best thing to wish for is better screen for the oculus.


This is just wrong. FPS is not latency or else we would be talking about FPS. You can have 1000 FPS and five days worth of latency if it took 5 days for the input to affect the screen. 60FPS means that latency can be between 0ms and 16ms at a minimum.

The other important thing is that even if you have a 120FPS panel and a graphics card that can render your scene at 120FPS you still do not get down to 0ms-8ms of latency for free. It really means you only cut out 8ms and if your latency was 64ms, you only cut it to 56ms. If you could do a million FPS on your screen then your latency would still be 48ms!

Look up Amdahl's law.
(╯°□°)╯︵┻━┻

JonnyCache
Honored Guest
@lordvtp Thanks a ton for those articles! I just finished the techreport.com piece and am about to dive into the one from pcper.com (Christ these things are long!)

@RedWorm I've traditionally been an "FPS is king" gamer, but I've found fps alone doesn't tell the whole story. If your card micro stutters (like my old crossfired 6850s) 80 fps translates to only 40 visible frames, and worse, the 40 visible frames gets cut down to 30fps with v-sync enabled which we've been told isn't good enough for VR.

Also, regarding latency, 20ms is the magic number we keep hearing. If you've got a GTX Titan your definitely going to get a silky smooth frame rate thanks to that 6gig of ram buffering frames, but how many frames get buffered? Say your card is running at 60fps, that's 16.67 milliseconds of latency, which sounds awesome! But if your card is buffering only 3 frames your going to have 50ms of lag between moving your head and seeing the image update before it even gets to your monitor.

Add in the Rift's screen delay you mentioned and your probably going to have a pretty poor VR experience. Most of the sli/cross fire setups I've seen buffer a min of three frames (one being drawn, one on the first GPU and one one the second GPU) , but many buffer more by default.

I totally understand where your coming from though, and I agree that getting a fast panel and faster pixel switching times is the top of the hit list. Personally, I've only recently started learning about this stuff, but its really changed the way I'm looking at my next hardware purchase. My gut reaction to horse power was to go 3 Titans (for super sampling), but the latency that would introduce would be terrifying. 😞

Sorry for being so long winded, I finally got off work and have been thinking about this all day
Intel(R) Core(TM) i7-3930K; ASUS Sabertooth X79; NVIDIA GeForce GTX 780 (*2); 32768MB RAM; Window 7 x64 SP1