The VR / PC Upgrade Reality Check

Subscribed; +1 OP if I could

Really looking forward to seeing the specs of the budget machine capable to run ED in VR on a CV1 for example. I don't have a PC and am going to have to build one that'll last as long as my car... or my iMac :D

Overclocking sounds quite scary:eek:
Here's the current World record for CPU overclock. http://hwbot.org/benchmark/cpu_frequency/halloffame

The only scary part is knowing your CPU will be hosed by the end of the day. Expensive weekend that.
 
I actually laughed out loud when I saw those figures (-: - hard to imagine the cost of that car crash.

http://www.octopusrift.com/building-the-rift-pc/ seems to describe a sensible build for under £700 (less the old copy of Win7Pro on disk i have) but looking at your advice; a PCI run SSD may be a better option? Also as quiet as poss is a preference;)

An SSD won't help your frame rate in VR, though it will help load times. Put that money toward getting a better gpu instead (Frontier recommends 980 or better). Only buy the SSD if you have money left in your budget after getying the best GPU possible.
 
Proper gear that supports VR is not cheap. However, it is possible to achieve decent performance on a reasonable budget as long as you take the above into consideration. Name brand components typically perform better than no-name Chinese knock-offs. The sum of all parts does matter in the world of VR. Good luck.
Sorry, but this is plain and simply wrong, even PCIe V2 is not saturated yet, because generally you aren't transferring the graphic data raw result via those lanes.
In short.
Read data information, textures from hdd/ram, cpu sends this to be processed by the gpu, this data is especially in Elite only around 4gb of data, so no issue at all.
The gpu itself is what handles the creation of the actual picture, and for VR GPU is going to be significantly more important, though yes the CPU is going to be a bit burdened as well, but not so much, all it really is doing is positioning and such data.

- - - - - Additional Content Posted / Auto Merge - - - - -

The rest I'll let people read and make up their own mind. You've taken my post far too literally. You also pick apart individual points that are designed to go together, so any context is out the window. If non-CPU/GPU components have no discernible impact on performance then please tell me how my stock 980Ti scored higher than a Titan X with the same CPU?
Because the 980Ti is more powerful then Titan X? Though not by much.
The Titan X only really has an advantage when Memory limit gets hit.

one of many reviews: http://www.extremetech.com/extreme/...itan-x-performance-at-a-fraction-of-the-price
 
Last edited:
An SSD won't help your frame rate in VR, though it will help load times. Put that money toward getting a better gpu instead (Frontier recommends 980 or better). Only buy the SSD if you have money left in your budget after getying the best GPU possible.

ok Up the GPU to a 980 - point taken and could be cheaper 2nd hand once everyone upgrades their super cooled small village light dimming GPU's at CV1 launch (-;
 
This is a really bad way to think about system performance.. (I say this as a developer and ex-performance specialist) however it's the way that sales people for PC manufacturers have traditionally done it. You see performance is all about bottlenecks, not about the performance of each component. It's been common for example, for laptop manufacturers, to talk about all kinds of different metrics which their laptops support when for the last 10 years only one thing has mattered on a laptop, the speed of the disk. Because that has been the bottleneck for almost everything.

I'm no expert on VR but I'm certain this premise holds true, to understand the performance requirements you need to understand what actually holds performance back for this one thing and knowing the MB/s of your internal bus is probably not relevant. Only bottlenecks are relevant and for the last 10 years or so that's almost exclusively been about GPU performance and GPU memory. CPUs are broadly not relevant these days and motherboards probably even less so..

While I agree that looking at the bottleneck is the right method, I'll disagree that CPU is completely irrelevant. Generally true for anyone upgrading all components on a steady basis to similar quality, but it breaks down when someone upgrades just their GPU. Doubly so when you're looking at Dx9, where CPU utilization is higher than Dx11.

It's also worth noting that VR does have a higher CPU demand than flat-screen gaming, because you need to maintain a tight process loop that will complete in time for the draw and update calls just-in-time to minimize latency. That means both enough CPU to run your game's physics loop and Oculus calls within the 90FPS window, and enough overhead so your CPU can handle other system events without dropping a frame.

Beyond that, RAM is the next potential component to cause bottlenecking, but there you're mostly concerned about having enough RAM to avoid page faults (and any resulting stutters, which will depend on how the game was written).
 
While I agree that looking at the bottleneck is the right method, I'll disagree that CPU is completely irrelevant. Generally true for anyone upgrading all components on a steady basis to similar quality, but it breaks down when someone upgrades just their GPU. Doubly so when you're looking at Dx9, where CPU utilization is higher than Dx11.

It's also worth noting that VR does have a higher CPU demand than flat-screen gaming, because you need to maintain a tight process loop that will complete in time for the draw and update calls just-in-time to minimize latency. That means both enough CPU to run your game's physics loop and Oculus calls within the 90FPS window, and enough overhead so your CPU can handle other system events without dropping a frame.

Beyond that, RAM is the next potential component to cause bottlenecking, but there you're mostly concerned about having enough RAM to avoid page faults (and any resulting stutters, which will depend on how the game was written).

Nicely said.

It all comes down making sure the CPU can issue as many draw calls as possible to the GPU while maintaining a steady logical & physics update frequency (the latter here running on a fixed interval). Luckily Cobra seems to have nice thread support evident when running a diagnostic tool with it. Additionally, it takes full advantage of multithreaded rendering.

It is always fun to see people saying "well my CPU usage isn't 100%, that must mean that I'll get no benefit from upgrading". It may seem logical but it isn't true. Especially on pre-DX12 interfaces where the GPU will idle while waiting for the CPU.
 
this is what i like about consoles and psvr, no nagging feeling if only a spend £££ on that "puma sniper f4tal1ty extreme XL ti" card it will all work so much better. i mean that amd pro duo "amd vr" card £1000 ... just no! :eek:

Funny how two people can see it totally different. :D

Your reason for liking the consoles is the exact reason why I like PC better. With console I m stuck with whatever performance they decide to allow me. With PC I can usually upgrade to get it where I want it.
 
Here is a u tube link to a great presentation on the subject of VR and performance bottlenecks by Valve..

Personally i think this one presentation helped me understand whats going on underneath the hood :) (car thing)
Because its not simple and some new concepts need to be considered when coding for VR, and its still not officially released yet so EARLY days.

One thing valve outlines that there pioneering.. there is ~11ms between frames @90hz and they send the draw calls BEFORE the frame sync to get max GPU utilisation..
there doing this on all there demos etc..

www.youtube.com/watch?v=ya8vKZRBXdw
 
Sorry, but this is plain and simply wrong, even PCIe V2 is not saturated yet, because generally you aren't transferring the graphic data raw result via those lanes.
In short.
Read data information, textures from hdd/ram, cpu sends this to be processed by the gpu, this data is especially in Elite only around 4gb of data, so no issue at all.
The gpu itself is what handles the creation of the actual picture, and for VR GPU is going to be significantly more important, though yes the CPU is going to be a bit burdened as well, but not so much, all it really is doing is positioning and such data.
As previously stated to another user, this is to future proof your PC for the arrival of Pascal cards which will double in bandwidth needs. I doubt you can even buy a 2.0 PCIe Mobo today. All the ones I see are 2.0/3.0. The rest of your answer is a good example of how things work. Thanks for that.
 
Thanks for the qualifications, still a little bemused by the "grossly inaccurate" comment but hey that's just me. All of your points are well made and accurate apart from... the criticisms about buying new components being an invalid option, why? Yes, there are other options but this is still a valid one. The point about CPU and GPU being the core focus within a budget again is valid but so is the OP's you are just nitpicking here IMHO. Your comments about PCIe saturation, again (again) point to someone who is looking at "making do" rather than building "intelligently" (in your words) with an eye on getting the best longevity out of a system. Mocking the Ferrari in a tractor analogy is just petty, to a lay person it is an acceptable example (yes, completely wrong in a number of ways, but sufficient to make a point). Same goes for the condescending memory, half point comment. More reference to "budgets", great to know but... not what the OP is talking about. OK now for the overclocking bit, yes we know overclocking is "easy" these days and "safe" (then why do Intel sell insurance specifically aimed at overclockers?), oh wait, no they are not, not even slightly (to do correctly, or should I say "intelligently"). Overclocking is an option and a valid one but let's not start making out that it will be easy and trouble free, please.

On your denouncement of his false economy point,

Overclocking will allow you to retire richer

A well-known benefit of overclocking is that you can purchase a lower performance, cheaper component and overclock it in order to attain the performance of a more expensive one. However, people often forget that overclocking incurs new costs that may or may not offset these potential savings. For example, you might need to purchase a high-performance fan or acquire a water-cooling system in order to better handle the heat generated by your overclocked component. It must also be said that overclocking results in higher power consumption which may in turn raise your energy bills. At the end of the day, you should consider overclocking more like a hobby than an investment.
(http://www.lavalys.com/community/blog/2010/04/top-7-myths-about-overclocking/)

The point about red lining the CPU, see above. The comment saying it's not cheap, well hang on all you have done is harp on about budgeting?? And finally the last comment is just puerile, or maybe it is just my low-grade English atoms that are at fault.

You make some points that need balance and IMHO a slightly less "all knowing" approach. I am not trying to start an argument here, I am just pointing out that your response could have been more helpful and less biased on a point not even focused on by the OP who acknowledges his approach is not for those without the budget, perhaps you should have written your own piece entitled "VR on a budget"?
 
Getting a whiz bang GPU running on a five year old mobo with mismatched clock speed RAM sticks is like dropping the engine from a Ferrari into a tractor. It might work, but not well.

This statement is grossly inaccurate. X58 came out in 2008 (8 years ago), it sports 40 PCIe 2.0 lanes, triple channel DDR3 (which is now matched in raw bandwidth by the recently released Z170, dual channel DDR4) and is capable of running a 6 core Xeon, if you're keen, which can overclock to 4Ghz on air without much fuss. Running a R9 290X is, in most scenarios, no different performance wise to a bleeding edge system. In the few exceptions where it is slower, the difference is within 5-10% in raw FPS (but were talking CPU bound benchmarks, or tests where the game was crippled on purpose to illustrate the difference between old and new).

I was running Elite Dangerous VR on an X58 with an R9 290X, and it ran very well on medium with some form of SSAA (either in game or using the new fandangled DSR). I only upgraded to bleeding edge because I can and I wanted the best experience possible.

My point is, PCs from 8 years ago are quite capable of providing an excellent VR experience provided you upgrade your GPU. I have no doubt that a 980Ti will not be hindered to any significant degree if installed in an X58 system. With Vulcan and DX12 it's likely we will start to see two things. Early on it will close the gap between older and newer systems (more efficient threading) but as implementation becomes more wide spread the gap will widen again as efficiency of the newer architectures are realised.
 
Last edited:
For best VR performance your mobo should support V3.0 of this standard, minimum. The various versions have significantly different data transfer rates.

PCIe 3.0 is only available on bleeding edge platforms (X99 enthusiast only, and Skylake). Haswell (dual channel DDR3 with 8 PCIe 2.0 lanes) has generally been favored over Haswell-e (Quad-channel DDR4 with 40 PCIe 3.0 lanes + 4 PCIe 2.0 lanes) as a gaming platform with benchmarks showing no tangible difference in performance, even in crossfire/SLI scenarios where additional PCIe bandwidth would be beneficial.
 
Last edited:
.

My point is, PCs from 8 years ago are quite capable of providing an excellent VR experience provided you upgrade your GPU. I have no doubt that a 980Ti will not be hindered to any significant degree if installed in an X58 system. With Vulcan and DX12 it's likely we will start to see two things. Early on it will close the gap between older and newer systems (more efficient threading) but as implementation becomes more wide spread the gap will widen again as efficiency of the newer architectures are realised.

The main drawback i found of an older system for VR (my 2011 Sandy I5 2500k on asus P8P67 was simply a lack of USB 3.0 ports. mine has 2 so probably doable but then i would have none spare. Nothing that a pci card cant fix tho.
 
The main drawback i found of an older system for VR (my 2011 Sandy I5 2500k on asus P8P67 was simply a lack of USB 3.0 ports. mine has 2 so probably doable but then i would have none spare. Nothing that a pci card cant fix tho.

My P6X58D Premium (2010) has USB 3.0 via an onboard realtek chip, My Rampage III Extreme has one as well. The only problem is it runs on a single PCIe lane so bandwidth is limited to 500MB/s (only enough to saturate one port).
 
Last edited:
This statement is grossly inaccurate. X58 came out in 2008 (8 years ago), it sports 40 PCIe 2.0 lanes, triple channel DDR3 (which is now matched in raw bandwidth by the recently released Z170, dual channel DDR4) and is capable of running a 6 core Xeon, if you're keen, which can overclock to 4Ghz on air without much fuss. Running a R9 290X is, in most scenarios, no different performance wise to a bleeding edge system. In the few exceptions where it is slower, the difference is within 5-10% in raw FPS (but were talking CPU bound benchmarks, or tests where the game was crippled on purpose to illustrate the difference between old and new).

I was running Elite Dangerous VR on an X58 with an R9 290X, and it ran very well on medium with some form of SSAA (either in game or using the new fandangled DSR). I only upgraded to bleeding edge because I can and I wanted the best experience possible.

My point is, PCs from 8 years ago are quite capable of providing an excellent VR experience provided you upgrade your GPU. I have no doubt that a 980Ti will not be hindered to any significant degree if installed in an X58 system. With Vulcan and DX12 it's likely we will start to see two things. Early on it will close the gap between older and newer systems (more efficient threading) but as implementation becomes more wide spread the gap will widen again as efficiency of the newer architectures are realised.
Entirely accurate, the ASUS Sabretooth X58 is a premium mobo (I have an ASUS Crossfire V Formula-Z) and should have no issues which is what I was saying. There are always hardware exceptions and people with the knowledge to get around obstacles. However, my point was that overall most people do not have the gear to run VR. I pulled the 9% number from the Steam VR test tool results, but there are even more pessimistic numbers out there.

"Only 13 million PCs worldwide will have that kind of power in 2016, according to the Santa Clara, California-based graphics chip maker. That is less than 1 percent of the 1.43 billion computers expected to be in use this year, according to research firm Gartner data cited by Bloomberg."

Article: http://www.cnet.com/news/reality-check-your-computer-is-most-likely-too-weak-to-run-vr-headsets/

To be fair the 9% number was generated (mostly) on gaming PC's where the 1% is taken from all computers, regardless of their main function, but it helps highlight the problem. Most computers (by design) cannot support VR.

All I wanted to do was let those with little or no technical knowledge know that they needed to look at their system overall before running VR. I've seen too many posts on various threads that say "just upgrade your GPU" and you can run VR. That may work for a lot of folks, but not everyone. A lot of people are going to be complaining when their VR gear arrives and their systems performance comes up short. The people who know how to overclock, who know how to invest in the right hardware combination, who know the difference between an i3 and i7 are the one in ten who are ready for VR. The rest aren't.

I predict many howling threads of "My Elite VR isn't working properly on ULTRA settings, it's all Frontiers fault" that will turn out to be caused by poor hardware, for the most part.
 
Back
Top Bottom