7 Gamers, 1 CPU - Ultimate Virtualized Gaming Build Log

[Str_Mape]

So are we just going to pretend that the Nano card was the perfect choice for this build because of its physical characteristics or are we going to hear about the limitations of this card as well? 

Power wise it was either the 980 or the nano.

Roughly 10-20% faster at the resolutions they'll be running.

Easier to get single slot I/O.

Short enough to not be ridiculously difficult to remove.

TLDR: No real reason to go green team on this one.

[EricX2]

I completely disagree with pretty well everything you just typed (as far as unRAID being "easier" to use than ESXi).

Yes, unRAID has the ability to use nVidia cards, via a workaround (that was admitted to be 'spotty' in the 2 gamers 1 tower video), but "easier" to configure otherwise, I completely disagree.

 

It's easier for Linus because he's got contacts with LimeTech who will help him configure it, (AFAIK) he doesn't have contacts with VMware to help set it up.

[middleclasspoor]

This would be an awesome addition to the LTT Folding team!!!......just sayin'.

[aw_]

Power wise it was either the 980 or the nano.

Roughly 10-20% faster at the resolutions they'll be running.

Easier to get single slot I/O.

Short enough to not be ridiculously difficult to remove.

TLDR: No real reason to go green team on this one.

 

You're missing the point and going back to the physical, power, and performance characteristics of the specific card, while assuming functionality is equivalent.  What I'm saying is that newer AMD cards often have problems with reset, which become a significant functionality limitation in a GPU assignment configuration.  Working well with GPU assignment should have been the first consideration, followed by power, size, performance, etc.

[bstevenson]

Umm Linus you could have gotten the Super Flower 2000W PSU. EVGA PSUs are rebranded Super Flower ones.

 

http://www.super-flower.com.tw/products_detail.php?class=2&sn=16&ID=119〈=en

[LabRat]

Why Linus did not use Nvidia graphic card?

you should have watched the video........... anyways, too much nvidia stuff in everything. time for something different.

 

is this what Linus was doing while his sidekick was doing the WAN show Friday night?........... would have been better doing the videos separately ( or not in the "same room ") so we could have heard the "play by play" as the machine was being put together. This assembly was more important and more informative to me than any WAN show and more infromation ( and video ) put before the viewer would have been great. This was a once in a lifetime oportunity and you blew it.

 

you spent more time dragging out that last 3way build off.

[OGFatal]

r9 Nano's with G-Sync monitors ohhh linus..

[scanferr]

What was used to do the virtualization part? HyperV? Citrix?

[Stadin6]

Is that an driver limit or an hardware limit?

i think the problem is directx (11). so an API limit. not completely certain though 

[Howhurl]

Holy hell, in the video you mentioned that there are a myriad of issues when you built this system, do you by chance kept a diary/log of all the issues and such? I would be interested in the ultimate behind the scenes of building this monster.

[michrech]

I doubt it's the hypervisor doing that over a service (dhcpd, something else?) doing so...

 

 

You dont even need the router, the hyperviser can supply a dhcp

 

[Centurius]

To everyone mentioning the 7 gamers 1 cpu thing, it probably was a joke on the fact that so many people call pc's cpu's, the very start of the video already shows it as 7 gamers 1 tower.

 

As for the project itself, this thing is insane and awesome, well done.

[michrech]

According to a link I found on their web page, they're using either Xen or KVM.

http://lime-technology.com/unraid-6-virtualization-update/

 

 

What was used to do the virtualization part? HyperV? Citrix?

 

[aw_]

unRAID uses KVM and VFIO for GPU assignment

[Yoshigriff]

"it's impractical...but if nothing else, it proves we have the technology."  This thing is sweet!

 

Definitely would be cool to see another video with more details on the issues that were had in setting this up.   Though like Linus pointed out...it's incredibly impractical and costs as much (and way more in many cases) as a brand new car.  So none of us really could do anything like this realistically.  But it would still be fun to watch.

[AnonymousGuy]

3 comments:

 

-I could have told him that 1 D5 pump wasn't going to be able to drive that many blocks.

 

-Also, it's not impossible to modify a dual slot GPU to be single slot.  I did it with a cheapie card desoldering the DVI connector (with a vacuum desoldering pump).  That' would have been a lot moar hardcoar, but it's timeconsuming as fuck.  Ideally you would only have to desolder HDMI or DP or something with less pins than DVI.

 

-Why wouldn't he just go dual PSU if needed?  When I built my system, I planed ahead to make sure the case could support dual PSUs should I go quad SLI.  I already use 1200W under full load with 2 cards.

[xgn]

Only problem with lanparty on this system you can see what you're friends are doing  

  also 3 of the gamers will only get 8x pci slot i dont know if everyone got their own gpu or if all share the full workload but its not a huge issue.

[limetechjon]

The hypervisor takes care of that task.  The TL:DR version is -- the CPU cores aren't "dedicated" to each VM, and unusable anywhere else...

That is incorrect, they are dedicated to each VM.  That is how we achieve such solid performance.  We do not emulate a CPU, we pass through the host's CPU cores directly to each instance.  This means 0 emulation overhead for CPU work.  We also do not let each VM traverse across each others dedicated cores.  This is to prevent costly CPU context-switching which would deteriorate performance to the point of it being unplayable.

[aw_]

We do not emulate a CPU, we pass through the host's CPU cores directly to each instance.  This means 0 emulation overhead for CPU work.

 

That's not how it works.  Using passthrough mode on the CPU only means that the VM sees the full feature set of the CPU.  The vCPU is sill a process thread running on the host CPU.  There's still emulation, the overhead is not zero.  The physical CPUs are still usable by the host, unless isolcpus was used, which isn't specified in the video.  Pinning vCPU certainly improves a number of aspects, but let's not go overboard.

[Sam Z Man]

Uses 2 CPU's, calls video 7 gamers 1 CPU.

*Dafuck*

[fueg]

Amazing! BTW, what's that keyboard you're using near the end of the video? It looks fu**ing nice!

[Marinatall_Ironside]

Maan!! Whoow!

 

@LinusTech I expected a really long video of this one. Especially considering that it is a $30k PC.

 

You could have asked Super Flower for their 2000w PSU and throw in 7 x 980 Ti in there.

7 980 Ti's wouldn't even fit in there since most of them are dual-slot cards. Even the K|NGP|N version of those cards would be extremely expensive, both in terms of price and power consumption, let alone he would have needed to go with a CaseLabs MAGNUM case and dual power supplies to even install and power up the system, plus extra radiators to properly cool the 300-watt TDP K|NGP|N cards.

[Cbeeman]

You should try this with 21 monitors. I want to see SEVEN surrond setups on one tower!

[ASDPinoyGeekBoy]

HOLY BALLS!

But, I'm not able to get to the CES 2016, so can you give me (or us) a sample on how it performs?

[Raggou]

This is simply awesome, I am always amazed at the creativity and resourcefulness of the Linus Media Team. I like how they have begun to expand their videos beyond just simple product reviews into really cool ideas like the Two Gamer's on CPU on to even crazier ideas like this. Keep up the fantastic work and I look forward to ever crazier and better videos.