WEBVTT 00:00:02.159 --> 00:00:06.319 whoa hey no that's not how you do it 00:00:05.200 --> 00:00:12.000 geez i don't think nzxt sponsored this video 00:00:08.880 --> 00:00:13.840 so we could chop up their hardware 00:00:12.000 --> 00:00:19.520 they sponsored it so we can show you guys how to take a single gaming rig 00:00:16.640 --> 00:00:24.720 like this build pc and split it into two or even more virtual gaming machines so 00:00:22.240 --> 00:00:28.000 you and a couple buds can pool together and share a single 00:00:26.320 --> 00:00:33.120 working GPU even if not everyone can get their 00:00:30.720 --> 00:00:38.440 hands on their own graphics card super cool right 00:00:34.719 --> 00:00:38.440 let's show you how it works 00:00:45.280 --> 00:00:52.160 splitting a GPU used to be a huge challenge for a number of reasons first 00:00:49.680 --> 00:00:57.440 and foremost being that NVIDIA didn't want you to you see NVIDIA locked down 00:00:55.039 --> 00:01:02.239 the ability to virtualize their consumer gpus taking one physical card and 00:01:00.480 --> 00:01:05.680 dividing it up into multiple virtual ones and for years they have only 00:01:04.320 --> 00:01:10.560 allowed this functionality on their workstation and data center cards like 00:01:07.760 --> 00:01:14.479 the quadro tesla and grid series these artificial software limitations 00:01:12.479 --> 00:01:18.880 have allowed NVIDIA to price those other cards vastly higher than their GeForce 00:01:16.960 --> 00:01:23.680 options and remember how i said there were a number of reasons there weren't 00:01:21.360 --> 00:01:28.240 NVIDIA kept you from sharing your GPU resources because they wanted to charge 00:01:25.920 --> 00:01:33.200 more for the capabilities shocker i know but we've got a trick up our 00:01:30.560 --> 00:01:39.360 sleeves today everyone's favorite Windows specifically Windows 11 and some 00:01:36.560 --> 00:01:43.600 fancy tech called para virtualization para virtualization differs from 00:01:41.200 --> 00:01:48.079 standard virtualization by communicating directly with the hypervisor using 00:01:45.600 --> 00:01:53.360 drivers rather than through hardware level calls so while spinning up a 00:01:50.880 --> 00:01:57.920 virtual machine or a vm setting up the hardware and software and actually 00:01:55.680 --> 00:02:02.320 getting to a workable vm used to take a lot of skill and time 00:01:59.520 --> 00:02:07.520 the key word there is used to thanks to a cute little script aptly 00:02:04.240 --> 00:02:09.280 named easy gpup p is for partitioning 00:02:07.520 --> 00:02:14.400 which was written by one of parsec's developers almost all of the work of 00:02:11.840 --> 00:02:18.640 setting up a Windows vm is done for you it divvies up your hardware installs 00:02:16.480 --> 00:02:22.319 your os and drivers and of course installs parsec but that's not just 00:02:21.120 --> 00:02:27.440 because they want to shield their software it's kind of the secret sauce 00:02:24.879 --> 00:02:30.720 that makes this whole thing work but before we get to that we need to talk 00:02:28.959 --> 00:02:33.920 about the limitations of the setup but the most important one being hardware 00:02:32.720 --> 00:02:38.480 outputs while the virtual machines we create are 00:02:36.239 --> 00:02:43.920 able to borrow our GPU's processing power and its onboard memory they cannot 00:02:41.440 --> 00:02:48.319 borrow the physical display outputs think of it kind of like this we can 00:02:45.760 --> 00:02:52.480 slice our GPU up this way where one of the virtual machines ends up with all of 00:02:49.920 --> 00:02:56.239 the outputs but we can't slice it up this way where each one gets its own 00:02:55.200 --> 00:03:02.560 output that is where parsec's cloud gaming software comes into play it's able to 00:03:00.000 --> 00:03:07.920 snag an encoded video feed from each of our vms and then stream it to a 00:03:04.800 --> 00:03:10.239 different device like a low-powered pc a 00:03:07.920 --> 00:03:14.560 macbook or even a phone before we can even think about splitting 00:03:12.480 --> 00:03:20.400 a GPU up though we're gonna need to get one fortunately our friends over at nzxt 00:03:17.599 --> 00:03:26.640 build sent over a sweet pre-built system that is absolutely perfect for this kind 00:03:22.720 --> 00:03:28.800 of use case it's got a ryzen 9 5950x 16 00:03:26.640 --> 00:03:32.720 core processor and a near top of the line rtx 3080 ti 00:03:31.280 --> 00:03:40.000 of course you can check out this rig at the link down below but you don't need specs like this to be 00:03:37.360 --> 00:03:45.360 able to do this we tested a handful of modern gpus with really amazing results 00:03:43.280 --> 00:03:50.080 though your mileage may vary and one thing to note is that you must split 00:03:47.599 --> 00:03:55.440 your GPU's video memory evenly between each vm so if your GPU is light on vram 00:03:53.519 --> 00:04:00.799 you may have issues gaming on more than a couple of vms since we're starting 00:03:57.599 --> 00:04:03.760 with two vms and one of these 3080 ti's 00:04:00.799 --> 00:04:07.439 that's gonna give us six gigs each now regular system memory can be divided 00:04:05.760 --> 00:04:12.400 up any way you like just like with regular vms but since we've got a total 00:04:09.920 --> 00:04:16.720 of 32 gigs of RAM and most modern games are asking for eight gigs or more we're 00:04:14.560 --> 00:04:20.320 gonna divide them up with eight gigs per via now we'll get into more details 00:04:18.400 --> 00:04:27.759 later but for now we've got hyper-v manager open on our monitor but 00:04:23.280 --> 00:04:29.440 very funny by deskpad ltdstore.com so 00:04:27.759 --> 00:04:34.479 hyper-v manager open on our monitor we're going to go ahead and fire up a 00:04:31.120 --> 00:04:38.880 couple of our vms here we've got GPU p1 00:04:34.479 --> 00:04:40.800 and GPU p2 for player 1 and player 2. 00:04:38.880 --> 00:04:45.360 each of these was configured with four of our 16 ryzen cores and as i mentioned 00:04:43.199 --> 00:04:49.120 before eight gigs of RAM fortunately the parsec script that we 00:04:47.360 --> 00:04:54.479 linked in the description did all the heavy lifting leaving us with just 00:04:51.360 --> 00:04:55.919 games to install here we are yay 00:04:54.479 --> 00:05:02.080 but before we fire them up we've got a couple of housekeeping items to take care of first and foremost is the fact 00:04:59.440 --> 00:05:06.479 that these Windows are limited to 1080p resolution which doesn't sound like a 00:05:04.400 --> 00:05:12.240 huge problem until you remember that the 3080 ti in this machine is capable of 00:05:09.919 --> 00:05:15.440 way higher resolutions than that that 00:05:13.199 --> 00:05:20.160 is where cheap little displayport or HDMI dummy plugs like these ones come in 00:05:18.160 --> 00:05:24.400 even though we can't actually use the physical ports on our GPU to plug into a 00:05:22.720 --> 00:05:29.280 monitor for each vm if you plug these in they can trick the 00:05:27.120 --> 00:05:33.919 GPU into thinking that a monitor is connected which actually unlocks the 00:05:31.360 --> 00:05:39.280 resolution selector in Windows so we go ahead and set that to 4k on each of our 00:05:36.160 --> 00:05:40.720 vms and it's time to see how a 3080 ti 00:05:39.280 --> 00:05:46.479 handles two concurrent 4k gaming sessions 00:05:44.960 --> 00:05:50.639 it's pretty cool right if you want to mirror it to both one and 00:05:48.720 --> 00:05:56.680 two then we can get a display here so people can see it sure i can do that 00:05:52.720 --> 00:05:56.680 first i'm going to set it to 3840x2160 00:05:56.880 --> 00:06:03.840 duplicate is not an option no interesting well um i guess you won't be 00:06:02.320 --> 00:06:08.319 peeking into what we're doing from over there let's for the first time in my 00:06:05.440 --> 00:06:12.560 life run two games at once on the same GPU on two separate machines colin you 00:06:10.720 --> 00:06:16.080 can be my counter-strike friend i am only at 1080p right right right right 00:06:14.720 --> 00:06:20.960 we'll get to the we'll get you the real performance numbers later the developer ran into a bit of a weird 00:06:19.680 --> 00:06:26.560 issue where for whatever reason that one vm is not 00:06:23.520 --> 00:06:28.400 able to do 4k uh but he's replicated on 00:06:26.560 --> 00:06:35.919 his side he's working on it okay this is not the most amazing performance ever 00:06:31.280 --> 00:06:38.960 but i'm at 60fps oh it's going david 00:06:35.919 --> 00:06:42.160 it'd be like whoa dawg what is this two 00:06:38.960 --> 00:06:43.919 games running at once oh someone's spawn 00:06:42.160 --> 00:06:47.840 camping us yeah they're all spawn camping us how rude come on it's 00:06:46.080 --> 00:06:52.000 counter-strike what'd you expect yeah that's fair i'm going back to 1080p okay 00:06:50.319 --> 00:06:57.199 we got a match to win here colin sure fair i mean i'm getting like 200 FPS 00:06:54.639 --> 00:07:01.680 yeah i want more like that my game is dead 00:06:59.280 --> 00:07:05.840 i'm gone steam is closed why did you do this and it's beautiful here 00:07:04.479 --> 00:07:09.039 terrible there oh yeah don't look at that and think 00:07:07.759 --> 00:07:13.759 that that is representative of the gaming experience at all hey still i got 00:07:11.759 --> 00:07:18.880 two kills okay okay but we're playing video games we're playing video games on 00:07:16.000 --> 00:07:24.319 one GPU and one CPU and we're both getting like 150 to 200 FPS yes 00:07:22.479 --> 00:07:28.880 david does this blow your mind yeah yeah he just shot me direct like 00:07:26.720 --> 00:07:33.599 straight through the door yeah but he could see like like he was 00:07:30.960 --> 00:07:38.400 tracking me i swear hax hacks question for you then in total we're 00:07:35.840 --> 00:07:44.240 getting about 400 FPS between the two of us on that GPU so running a single 00:07:41.440 --> 00:07:49.120 instance at 1080p what were we getting i only ran it at 4k balls well but at 4k 00:07:46.960 --> 00:07:52.880 we're getting 462 FPS so theoretically assuming that we are 00:07:51.759 --> 00:07:57.280 still CPU limited i have no idea i just don't 00:07:55.520 --> 00:08:03.680 even know what to expect from this like when you split the GPU across multiple workloads in a 00:08:01.199 --> 00:08:07.520 game like cs go where you're typically CPU rather than GPU bound could you 00:08:05.680 --> 00:08:11.120 potentially end up with more FPS's or less fpss 00:08:10.319 --> 00:08:16.240 it looks like a little bit less right now 00:08:13.520 --> 00:08:18.960 given that we're running 1080p but 00:08:17.120 --> 00:08:23.759 what if we were to split it more ways 00:08:20.879 --> 00:08:27.680 would it scale better or worse we can find out i guess we can as 00:08:26.240 --> 00:08:32.800 impressive as it is that this is running at all i'm sure what you really want to know is 00:08:30.319 --> 00:08:38.640 objectively speaking how well is it running right so right now we're getting 00:08:36.959 --> 00:08:45.279 anywhere from 350 to 400 FPS shared between the two of 00:08:42.959 --> 00:08:48.320 us right so how does that compare to running this same card 00:08:47.440 --> 00:08:52.880 but just in a single instance on bare metal 00:08:51.120 --> 00:08:57.440 even though we only have one of these running at 4k right now we've actually 00:08:54.399 --> 00:09:01.160 got some numbers for you so cs go on a 00:08:57.440 --> 00:09:03.920 single 3080 ti managed about 00:09:01.160 --> 00:09:11.839 460 FPS average then with our 50 vm so this is half of a 00:09:07.120 --> 00:09:15.360 3080 ti we managed 175 FPS so it looks 00:09:11.839 --> 00:09:17.040 like we actually do get less than half 00:09:15.360 --> 00:09:21.920 of the performance the scaling is not perfect which is i have to admit not 00:09:19.680 --> 00:09:27.600 quite what i expected csgo being a mostly CPU-bound game i would have 00:09:24.959 --> 00:09:32.720 thought would allow our GPU to stretch its legs more but who knows maybe we're 00:09:30.480 --> 00:09:36.160 not dividing it up enough we've got more demos right we do got 00:09:34.320 --> 00:09:39.680 more demos all right let's try let's try something new yeah i know i'm averaging 00:09:38.080 --> 00:09:43.440 like 50 FPS even in fairly busy parts of town 40 00:09:42.000 --> 00:09:48.200 dipped down to 40. my 99th percentile is like 25 00:09:46.880 --> 00:09:55.200 that is absolutely playable in a game like this so to be 00:09:52.160 --> 00:09:58.240 clear anno 1800 not the world's most 00:09:55.200 --> 00:10:00.480 demanding game but it's also not the 00:09:58.240 --> 00:10:04.240 world's easiest to run game like this all this is all very reasonable 80 00:10:02.240 --> 00:10:09.680 degree hotspot like fine yeah that thing's not working that hard actually 00:10:06.000 --> 00:10:13.360 so dual cs goes is for whatever reason 00:10:09.680 --> 00:10:16.399 heavier than dual anodes 00:10:13.360 --> 00:10:16.399 is that because 00:10:16.839 --> 00:10:22.240 the CPU is scaling really well in cs go 00:10:20.640 --> 00:10:27.680 and making it so hold on i got to win lunch so GPU usage is right up around 70 00:10:25.839 --> 00:10:31.600 75 i'm surprised it's not just pinning it i 00:10:29.600 --> 00:10:35.920 wonder if the host vm is using some of it could be 00:10:33.600 --> 00:10:40.959 clearly it works and all things considered it was actually pretty easy 00:10:38.160 --> 00:10:45.519 to set up colin here who set up and wrote this video has literally never 00:10:44.000 --> 00:10:49.920 touched hyper-v yet he was able to get this up and 00:10:47.440 --> 00:10:53.440 running in a couple of hours so what does it take to do it well apart 00:10:52.079 --> 00:10:57.920 from what we've already mentioned you're going to need a pro or enterprise 00:10:55.360 --> 00:11:00.640 version of Windows 10 or Windows 11. we would suggest the latter as it is a bit 00:10:59.760 --> 00:11:07.040 easier you start by downloading the files from the easy gpup github and extracting them 00:11:05.120 --> 00:11:10.480 to your desktop then you grab the latest drivers for your GPU from NVIDIA's 00:11:08.720 --> 00:11:14.880 website directly now comes the most important step you 00:11:12.320 --> 00:11:18.079 need to enable virtualization both in the BIOS 00:11:16.079 --> 00:11:21.760 and in Windows now on most systems you'll find the BIOS 00:11:19.839 --> 00:11:27.760 settings under CPU settings and they'll either be called svm mode on AMD or 00:11:24.640 --> 00:11:29.360 Intel vtx on well Intel 00:11:27.760 --> 00:11:34.000 save and reset and once you're back into Windows hit the Windows key and type 00:11:31.440 --> 00:11:38.240 turn Windows features on or off scroll down to hyper-v and enable 00:11:35.920 --> 00:11:41.279 everything inside and reboot your pc while it's booting by the way why not 00:11:39.519 --> 00:11:45.600 hit the subscribe button we've got a lot more cool Windows 11 content planned 00:11:42.959 --> 00:11:48.959 like running it on a phone now that our operating system is set up 00:11:47.040 --> 00:11:52.640 properly we can run the easy gpup scripts because it's possible that these 00:11:51.040 --> 00:11:55.959 steps will change i'm going to refer you to the github page to find the latest 00:11:54.240 --> 00:11:59.920 information and if you get stuck the linustechtips.com forum is a great 00:11:57.680 --> 00:12:04.880 resource once you configure your script you can rerun it over and over to keep 00:12:02.720 --> 00:12:09.200 making vms just make sure to change the vm name each time before creating 00:12:06.800 --> 00:12:12.720 another and remember that NVIDIA limits the number of concurrent end bank 00:12:10.720 --> 00:12:16.560 encoding streams to three on their consumer graphics cards meaning the 00:12:14.240 --> 00:12:22.000 practical player limit is four one local and three remote ones 00:12:20.000 --> 00:12:26.320 if you did everything right in hyper-v manager you should see your new vms just 00:12:24.320 --> 00:12:30.399 like we saw earlier you double click the vm to boot it up and then you can 00:12:27.760 --> 00:12:35.600 install your games and configure parsec now all that's left is for us to go 00:12:32.399 --> 00:12:39.120 ahead and make three vms each with four 00:12:35.600 --> 00:12:42.079 CPU cores 25 of our GPU and eight gigs 00:12:39.120 --> 00:12:47.600 of RAM to see if we can run four gamers off this single CPU and GPU 00:12:45.440 --> 00:12:51.120 which which one is alt is it command is that command is win key oh it's got to 00:12:50.399 --> 00:12:55.440 be option okay um 00:12:54.160 --> 00:13:00.079 i think we might just have to lose the macbook it was a funny meme but uh we're 00:12:58.399 --> 00:13:07.040 just we're having this game we're having some issues hey 00:13:03.279 --> 00:13:10.639 macbook works oh heck yeah i got my LTT 00:13:07.040 --> 00:13:12.800 crew it's Linus and the Linus tech tips 00:13:10.639 --> 00:13:16.000 okay that's a cue Mac Address you know and it's like you're the little brother 00:13:13.920 --> 00:13:20.480 and he's just holding the controller but not actually doing anything yeah this is 00:13:18.000 --> 00:13:23.760 what i feel like right now i think we're doing it guys oh i'm playing oh i got 00:13:22.880 --> 00:13:28.800 one wait i killed ltd benchmark tertiary 00:13:26.560 --> 00:13:33.040 we're not all on the same team so we're actually getting a grand total 00:13:30.800 --> 00:13:36.800 of around 300 frames per second does that sound about right to y'all jeff's 00:13:34.880 --> 00:13:41.279 experience being on like native is clearly better uh yeah that looks a 00:13:39.600 --> 00:13:45.360 little more responsive because i'm i'm getting a lot of chop on 00:13:42.959 --> 00:13:49.680 my own look like that wow oh let's go 00:13:46.959 --> 00:13:54.800 yeah with only a quarter of the GPU he's still getting like 150 180 FPS 00:13:54.880 --> 00:14:00.240 GPU's definitely working harder than it was with anno right now though 00:13:58.880 --> 00:14:05.279 you know what though i actually don't think it's working as hard as it was with two instances of cs go can you all 00:14:03.600 --> 00:14:11.120 tab for a sec whoa GPU core load is 98 00:14:09.040 --> 00:14:15.279 i lied i think the fan's just running heavier okay 00:14:13.440 --> 00:14:20.079 remember guys not only is it running four instances of the game but it's also 00:14:18.079 --> 00:14:23.680 running four instances of NVIDIA's hardware and coding oh there's a guy 00:14:21.839 --> 00:14:28.079 there and theoretically that hardware encoder 00:14:25.839 --> 00:14:31.760 is sort of separate hardware but in practice it does put additional strain 00:14:30.160 --> 00:14:35.920 on the GPU it does produce additional heat when it's operating and so it can 00:14:33.839 --> 00:14:41.120 affect your overall performance so how do i change guns so press b to 00:14:37.920 --> 00:14:42.720 buy guns oh yeah what's to buy yeah you 00:14:41.120 --> 00:14:47.760 have to buy guns i mean logically imagine what this would be capable of if 00:14:44.639 --> 00:14:49.440 NVIDIA allowed us to access them locally 00:14:47.760 --> 00:14:53.360 rather than remotely you could have a single gaming machine in the middle of a 00:14:51.040 --> 00:14:57.279 big table and just have four separate sets of peripherals hooked up to it and 00:14:55.680 --> 00:15:03.519 all be playing a game off the same hardware that's freaking incredible 00:14:59.920 --> 00:15:04.720 but the 3080ti is pretty expensive what 00:15:03.519 --> 00:15:10.720 if we tried it with something more reasonable like a 2070 super we did 00:15:07.920 --> 00:15:14.320 and as long as you're only gonna play fairly light games things like rocket 00:15:13.440 --> 00:15:20.399 league it's actually surprisingly usable as well 00:15:18.639 --> 00:15:23.519 though there are some things to note about this setup if you're going to try 00:15:22.240 --> 00:15:29.839 it for real hardware that crazy like top of the line is not 00:15:27.440 --> 00:15:34.160 strictly necessary but it definitely helps and perhaps even more important is 00:15:32.399 --> 00:15:38.000 your network speed parsec suggests having an upload speed of over 50 00:15:36.320 --> 00:15:42.480 megabit per second if you plan on hosting two or more vms over the 00:15:40.399 --> 00:15:47.680 internet and even if you're running locally you will need a directly wired 00:15:45.120 --> 00:15:51.839 connection wi-fi or mesh networking might have the throughput for modern 00:15:49.199 --> 00:15:56.079 applications like 4k netflix or what have you but their latency is not nearly 00:15:54.639 --> 00:16:00.240 good enough for this you may also encounter some black 00:15:57.920 --> 00:16:05.040 screens with Windows security prompts and stuff like that so having parsec 00:16:02.160 --> 00:16:09.440 access to the host as well as the vms is a good idea so you'll be able to squash 00:16:07.600 --> 00:16:13.199 those little interruptions another thing is you should definitely 00:16:11.199 --> 00:16:18.480 temper your performance expectations as we found 400 FPS on bare metal 00:16:15.680 --> 00:16:25.199 absolutely does not result in four times 100 FPS on four vms you'll also need a 00:16:22.320 --> 00:16:28.560 Windows license for each of your vms definitely don't just spin up a new one 00:16:26.720 --> 00:16:33.600 whenever your license expires another thing to discuss is the cost 00:16:30.800 --> 00:16:37.360 you're absolutely saving some money using a single machine to run four 00:16:35.600 --> 00:16:42.320 instances instead of four separate machines you only need one case you only 00:16:40.000 --> 00:16:46.000 need one power supply you only even really need one SSD 00:16:44.320 --> 00:16:50.959 but it should be noted that you're going to end up shelling out for higher 00:16:48.000 --> 00:16:56.639 quality hardware versus if you only had to run a single client workload say take 00:16:53.519 --> 00:17:00.320 the SSD for example if you were to use a 00:16:56.639 --> 00:17:02.079 basic basic qlc SSD with no dram cache 00:17:00.320 --> 00:17:06.240 you could get away with that for a single gaming machine you try to have 00:17:04.319 --> 00:17:10.000 four clients working off of that thing at the same time you're gonna have an 00:17:08.079 --> 00:17:14.160 extremely bad time and the same goes for a power supply never mind just the 00:17:12.160 --> 00:17:17.679 wattage that it's rated to output but let's talk about the duty cycle 00:17:15.919 --> 00:17:21.360 manufacturers when they warranty these things when they decide about how long 00:17:19.679 --> 00:17:25.520 they expect them to last they have an expectation for how many hours a year 00:17:23.839 --> 00:17:29.600 you're probably going to have it running and how hard it's going to be hit as we 00:17:27.760 --> 00:17:35.520 discovered with all four of our gamers hitting that 3080 ti it is pinned at 100 00:17:34.559 --> 00:17:40.960 usage these and other components like your motherboard or a mini core CPU are 00:17:39.600 --> 00:17:47.440 things that you would want to consider when you're configuring your system with someone like of course nzxt who is the 00:17:45.600 --> 00:17:51.120 sponsor of this video so in conclusion this is a really cool and novel use of 00:17:49.360 --> 00:17:54.640 parsec it's normally used to have a friend remote in to play a multiplayer 00:17:52.640 --> 00:17:58.160 game or to stream games from your home pc to another of your own devices kind 00:17:56.640 --> 00:18:03.120 of like having your own private cloud gaming service but here you're turning 00:18:00.799 --> 00:18:07.760 it into all of you and your friends private 00:18:05.440 --> 00:18:11.679 cloud gaming service which is a seriously legitimate possible use case 00:18:09.840 --> 00:18:14.720 for it especially in these troubled times now in addition to performance 00:18:13.520 --> 00:18:18.640 drops though you'll experience additional connection latency about a 00:18:16.720 --> 00:18:22.960 frame on a local network and a couple of frames on a remote one and you'll be 00:18:20.720 --> 00:18:27.200 giving up some image quality even a high bitrate hevc stream is going to have 00:18:25.120 --> 00:18:30.799 more blocking and artifacts than a hardwired displayport connection 00:18:29.280 --> 00:18:34.559 especially if you crank your frame rate past 60. but i'm super excited for the 00:18:33.039 --> 00:18:39.600 future of this kind of technology and i think it's an awesome little project that they created over there so go give 00:18:37.840 --> 00:18:44.640 them some love all right if you enjoyed this video go check out six workstations 00:18:41.679 --> 00:18:49.760 one CPU or seven gamers one CPU for an example of doing this with actually very 00:18:47.760 --> 00:18:53.360 little performance drawback because we loaded up the system with gpus back when 00:18:51.600 --> 00:18:56.640 you could actually find them okay who just killed me 00:18:55.360 --> 00:18:59.640 rude rude