Oh, one more little question. Is it possible to run higher than 60 fps?
Yup . I only have 60Hz monitors so I can’t test anything higher yet. The higher the refresh rate the lower the latency, for example at 140Hz the worst case latency will be around 7ms, at 240Hz, it will be around 4ms.
Alright, i’m pretty sure I know the answer, but where would things like Gsync stand. Being that it’s hardware implementation.
This was talked about earlier.
It is. Gsync is going to be between the monitor and GPU.
In my opinion, Gsync/freesync is a crappy technology for people who don’t know how to run games at the proper graphical settings. Run your games at the proper settings to match your GPU and Monitor’s capabilities and you don’t need any fancy synchronization.
Do you know of any online website that would help you figure out the proper settings to match your GPU and Monitor’s capabilities? If I can save 200 to 600 dollars on my next Desktop build that would be great.
It’s about benchmarking. Most games have an FPS counter, or you can run RivaTuner and inject to monitor fps. You find your monitor’s refresh rate (mine is 60hz) and aim for that framerate. I run regular Vsync, but some people don’t like that because it induces a bit of latency.
From there, just play with the settings until you hit that target framerate. Every game is going to be different.
I’m curious by what you mean by that. Like frame timing to your monitor? The dell ultrasharps are extremely good if thats what you’re worried about.
Insane! I love it. I work in Windows VM to support my work life, and switching back and forth on key/mouse/display is bothersome (doable, but bothersome).
Does this take into account VM’s HDMI audio in some shape or form? Just asking.
Audio sent over HDMI can only be received by a monitor, if you switched input on your monitor or unplugged it from the passed through GPU to see host again it can’t play it.
You can play sound fo an emulated sound card in your VM though which is how audio is normally passed to host (specifying host audio driver, also or pa, for qemu process). Otherwise, you can of-course pass physical DAC or sound card to VM and if you need both host and guest audio combined you’d need to pass their outpout in a line-in of a sound card which the host still can use.
This might be a stupid question but here goes… Won’t a higher framerates, perhaps even at higher resolutions, mean higher bandwidth on the pci lanes to the native card which is already using said bandwidth to communicate with the system. What I’m trying to say is, won’t you hit a wall at somr point? Or perhaps other things will be the bottleneck long before that?
Yes, it will, so let’s do the math.
1920 x 1200 x 4 = 9,216,000 bytes per frame * 60 Hz = 552,960,000 bytes per second / 1024 / 1024 = 527.34375 MiB per second 1x PCIe 1.0 lane runs at 250MiB/s * 8 = 2000MiB per second 1x PCIe 2.0 lane runs at 500MiB/s * 8 = 4000MiB per second 1x PCIe 3.0 lane runs at 985MiB/s * 8 = 7880MiB per second
And that’s only on 8 lanes, most people are running 16, you have a ton of headroom before it becomes an issue.
Let’s assume 16 lanes of PCIe 3 at 15,760 MiB per second and calculate the teoretical maximum at 1920 x 1200 and 4K
16,525,557,760 / (1920 * 1200 * 4) = 1793.14 Hz 16,525,557,760 / (3840 * 2160 * 4) = 498.09 Hz
Games do need much of this bandwidth, but only when they launch and are loading their textures for the first time, so there will be a performance hit but it will be marginal. In fact my TimeSpy score drops about 100 points when I am using the program, but in general gameplay it is undetectable. Also my 1080Ti is only on four lanes (AB350 Pro4 only has 4 lanes on the 2nd PCIe slot), so my score may be improved if it was in a 16x slot.
If this in the future ever doesn become a problem we can look at using the YUV420 color space which is native in hardware (so fast) but consumes only 12bits per pixel instead of 32, but at the cost of color accuracy. See other streaming services (Steam/ShadowPlay) for what that looks like
Thanks for the explanation. But yeah, yoy are right, does seem like it is far off becoming a problem. Especially considering GPU performance will probably not advance as fast as PCI speed. Heard version 5 is the next big thing. That’s my guess at least.
Thank you for your support so far, I hate to plug this in here but we are short of what’s required to continue on this.
At this point it is going to cost no less then $4,000 AUD for the components needed to build the development system, and with GoFundMe fees (AU: 7.25%, International: 8.4% + $0.30) the total that needs to be raised is around $4,336 AUD.
At the time of writing this I am at $3,105 before fees, leaving a $1,231 defect.
The $4000 AUD covers:
- PSU - needs to be 1000W or better for running the two video cards (worst case, my 1080Ti + Vega 56)
- Ryzen 1950x - This is needed for the additional PCIe lanes
- 32GB RAM - 64GB would be better but can make do with the 32GB
- Vega 56 (For both host and guest development)
- Hopefully a few cheap budget video cards of different GPU vendors for testing on the host
- Best HDD I can get with whatever is remaining
The motherboard and CPU cooler have already been pledged by Level1Techs.
PLE Computers Australia have offered to supply parts for this at cost, so final pricing at this point is an estimate.
Hardware donations to cover any of the above list would also be greatly appreciated reducing the final required amount.
Got a couple 256GB SATA SSDs sitting here you can probably have if that suits, bonus points if you are in Brisbane so shipping is free/low
Thanks! That would be very helpful! I am in Katoomba though so up to you if you want to ship them down this way . If so PM me and i’ll give you addressing details.
Also got an ASUS DirectCU 2 GTX 970 that isn’t being used. Knocks itself down to ~600MHz on the front side for no good reason, but hey at least it’s something from Maxwell. Was a dud 2nd hand purchase, it’s yours if you think you can use it.
It isn’t strictly required as I have a few NVidia cards here already but nothing of that generation. It would diversify the configurations that could be tested though. Up to you .
Some older crappy workstation cards are more what I am looking for so that the host doesn’t need to have a high end expensive card, the idea is to try to make the setup as cheap as possible for people.
This is on my desk right now. No idea what the fuck it is but I do know that it registers a 9.8 on the crapter scale.
Perfect :D. There are various Vendor specific extensions that can be leveraged on these cards to get them to perform, since we are simply passing back frame data these cards should be more then capable of rendering this ability.