New gaming system build for rFactor2 - Z170 or X99

Discussion in 'General Discussion' started by SpeedLab, Nov 12, 2015.

  1. DocJones

    DocJones Registered

    Joined:
    Jan 26, 2011
    Messages:
    403
    Likes Received:
    5
    Hardware websites and gfx manufacturers can't expect everyone to invest in a quality psu. Lots of people indeed spend several hundered bucks for a badass graphics card without batting an eye but then are too cheap to pay for a good quality psu and end up with a shabby 1000W junk psu that don't always deliver (to put it mildly) what the manufacturer say they are capable of. Add a bit of marketing talk from other manufacturers and hearsay from people on hardware forums... Your psu is absolutely capable of supplying 2 superclocked 980ti cards, not even breaking a sweat.

    If you're interested, here's some more numbers: Your EVGA Superclocked 980ti is factory overclocked. To run overclocks stable generally requires higher power supply compared to running stock speeds. Here's a 980ti comparison on a german site (just scroll to the graphs) I trust when it comes to power consumption measurements. They measure their test system (complete PC) with a standart Nvidia 980ti (default layout and clockspeeds) at 353W and with a EVGA superclocked at 381W under full system load, so about 30W more for the factory overclocked EVGA.
    Here's another test on a different german site. In contrast to the first site they have measurement equipment to check only for the power consumption of the graphics card itself and they ended up with 250W under full load for a standart 980ti. Add the 30W your EVGA superclocked needs over the default card to that and we end up with 280w for a single or 560W for dual EVGA superclocked cards. You see where this is going.
    Here's a test of a i7 6700k. The first set of graphs has power consumption readings (full system again). Since we know what the gfx card(s) require(s), we are interested in the "max. CPU-Last (Prime 95, AVX-Test)" numbers, that's a benchmark that puts the CPU under heavy load on all cores without putting load on a graphics card. 141W for the 6700k system.
    So we're at approx. 420W with a single GPU and 700W if you decide you want the 2nd 980ti. Note these are rather theoretical numbers that you'll probably never see in real gaming situations, it's only benchmark numbers. Power consumption differs a bit between mainboards, number of HDDs/SSDs, number and kind of RAM sticks etc, but you see there is still lots of room for a proper cpu overclock.
    Given that you are unsure about possibly putting a 2nd 980ti in the rig, I'd say you made a very good choice with the Corsair RM850i.
     
    Last edited by a moderator: Nov 14, 2015
  2. DurgeDriven

    DurgeDriven Banned

    Joined:
    Mar 20, 2012
    Messages:
    6,320
    Likes Received:
    43
    Then if he is possibly getting another GTX980Ti ..........................

    He never told me that :)

    Why are we talking about 6700K ?

    I spend $2000 on video cards to run at 8x 8x when for a few hundred more I can run 16x 16x :confused:
     
  3. Spinelli

    Spinelli Banned

    Joined:
    Jan 28, 2012
    Messages:
    5,290
    Likes Received:
    32
    If somebody has the money and/or desire to build a "top" system and run SLI then screw the "mainstream" line; either get X99 Haswell-E (i7-5930K or, if you have money burning a hole in your pocket, i7-5960X) or GTFO!!! :)

    End of discussion (regarding chipset/CPU)!!! :)
     
  4. SpeedLab

    SpeedLab Registered

    Joined:
    Nov 25, 2012
    Messages:
    86
    Likes Received:
    0
    @rogue22 Really not planning on the multi GPU as I really only play rFactor2 and I know it works better with one GPU.


    @DocJones Sounds like it might be tight even with an 850w PSU with two GPUs. I'm really not thinking I ever will put two cards in unless something changes in the gaming landscape in the near future. Most likely I would sell off my 980 ti and then just upgrade to whatever is the best card then. My main concern was what you had said earlier just having enough good quality power so I could overclock my CPU and GPU and run the rest of my system without having to worry about some kind of "poof"!!!

    @DD I didn't really mention about putting in a second 980 ti because I'm not really planning on it. This was one of the reasons I didn't go for the x99...really not thinking about multiple GPU's and seems the general consensus is if you're running just a single card the new Skylake chips are a bit faster.

    Also, and of course I could be totally wrong about this, but just about everything I read said that x8x8 PCie 3.0 will run just as fast as x16x16. It says that there is no bandwidth bottleneck to the x8x8 configuration currently so there really isn't an advantage to x16x16, it just cost more money.

    If the above statement is true, then I spent $150 extra to get a higher grade PSU and i7 6700, both of which it seems I would want if I ever do go dual GPU for less than 5% more total cost. Part cheap insurance...part overboard!!! :p
     
  5. SpeedLab

    SpeedLab Registered

    Joined:
    Nov 25, 2012
    Messages:
    86
    Likes Received:
    0
    @ Spinelli that the exact conclusion that I came to after my research. I would need to get at least the i7-5930k for the 40 lanes. Truthfully the system was the exact same price give or take, that's why it was a tough decision. In the end I decided that most likely I will never go multi-GPU so I went Z170 which is supposed to be a bit faster in single threaded performance, run cooler, and use less juice.

    What is your take about the x16x16 vs x8x8? Does it slow down multi GPU setups running in x8, or is that only with rFactor2?
     
  6. DocJones

    DocJones Registered

    Joined:
    Jan 26, 2011
    Messages:
    403
    Likes Received:
    5
    Not sure where I said something about 'poof'. :D You're perfectly fine with both system and psu. Really, there's no need to be concerned about the psu not being sufficient enough if you decide to go dual GPU. The 700W I calculated above is well within the limits of your psu. Btw, 850W is what it is capable of to put out, not what is its limit drawing from the power line safely. Output as in what is available to your PC's components after taking the 90% efficiency at max load in account. What it could safely draw from the power line given it was running at full (850W) load is actually ~944W. With the two different sources I linked to before you'd have to do a little more math because one is measured between powerline and psu and the other between psu and gfx card, but still. (Edit: rough calculation: 760W input to psu instead of 700W).
    Lots of room for overclocking and a bunch of additional parts in your rig. Skylake CPUs in general have great efficiency, and also overclocks don't seem to require a lot of additional power: click (this is only the first review I found when I looked for one a minute ago with both default clock speed and overclock power consumption for a 6700k, but I doubt further investigation would show vastly different results).

    Stop worrying and look forward to / enjoy your new build. ;)
     
    Last edited by a moderator: Nov 14, 2015
  7. SpeedLab

    SpeedLab Registered

    Joined:
    Nov 25, 2012
    Messages:
    86
    Likes Received:
    0
    Cool thanks "Doc" :p "poof" just sounded fun :D
     
  8. DurgeDriven

    DurgeDriven Banned

    Joined:
    Mar 20, 2012
    Messages:
    6,320
    Likes Received:
    43

    Sorry I misconstrued this...........

     
  9. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,294
    Likes Received:
    36
    Only with rf2 it would seem and if you were to go 980Ti SLI, with each card running in PCI-e 3.0 x16 vs 3.0 x8 mode at a resolution of 1920x1080 you can expect to gain around 30-35% (if not even more) extra average fps in rfactor 2 with max graphics settings.

    30-35% is not a small amount of potential performance wasted if you do decide to go sli in the future with your current cpu+mobo combo limiting you to PCI-e 3.0 x8/x8 for rf2 imo.

    If assuming that when/if sli is working in rf2, then provided both cards were able to run at 3.0 x16 mode, the sli solution should give you about 180% the performance of the single card performance (which i think is the typical average for sli). Now based on my estimate that each 980Ti running in pci-e 3.0 x8 will result in 70-75% the performance of each 980Ti running in 3.0 x16 mode, you will end up with around only 33% the performance of a single 980Ti.

    So in summary, 2x 980Ti in pci-e 3.0 x8/x8 = ~133% the performance of 1x 980Ti in pci-e 3.0 x16 mode in rfactor 2.

    Now, these relative performances estimates are based on a single 1920x1080 resolution. If you plan on more pixels, the difference will reduce (i.e. pci-e mode difference affects percentage average fps difference less) but by how much depends ofc on your intended resolution and whether for example if you plan on using mutliview if going triple screen (for which i cannot make a reasonable estimate because i don't know if that affects performance without multiview anymore.)

    Last point, this are my estimates based on extrapolations of numbers i've collected from many yet arbitrary benchmarking tests performed by myself and others in rf2 with different cards running in different pci-e modes and observing the affects on average fps in rf2. I think this is a reasonably accurate estimate for the performance one should/would get if one were to benchmark compare each of these options (that is, 980Ti in pci-e 3.0 x16 vs 3.0 x8/x8 @1080p).

    edit: Reverted last correction as I think i did do it right the first time. Which if correct, it's a pretty big loss of potential performance to be had with a fully capable pci-e 3.0 x16/x16 system if you choose to go sli.

    If i were in your position of building a new pc with a 980Ti as an option, i would plan with the intent of just getting the one and not think about sli. I've had perfectly good experiences with sli in the past but i think with a 980Ti you don't need to worry about performance. The card is a beast already.
     
    Last edited by a moderator: Nov 14, 2015
  10. DurgeDriven

    DurgeDriven Banned

    Joined:
    Mar 20, 2012
    Messages:
    6,320
    Likes Received:
    43
    Something people may want to go over is all the other questions he asked which I answered in a PM early.

    What do you think of the 2 SSD ( 512GB - Samsung 950 and 250GB - Samsung 850 )

    Speed posed a couple questions.

    Fast one ( 950) obviously Windows and rF2 must go on that should boot in the blink of a eye, well almost. :)

    But what do we put where and do we partition ?

    This would be a prime concern for me.

    I like to use True Image and do "Full" C:\ partition backups. ( MBR , reserved space etc )


    The larger the partition the more time they take.

    My OSes are 40GB partitions ( Force3 and Samsung Pro ) and they both take about 3 minutes to backup or restore.

    This way all my sims can go untroubled, I do not have to worry about incremental backups for things like garage changes in sims every day.



    Anyone think he may be better with a 256GB-950 and 1TB-850

    Should be close to same cost ?
     
    Last edited by a moderator: Nov 14, 2015
  11. SpeedLab

    SpeedLab Registered

    Joined:
    Nov 25, 2012
    Messages:
    86
    Likes Received:
    0
    @DrR1pper Thank you for all of that. So if I understand you correctly only in rFactor2 is the 3.0 x16x16 important for running multi-GPU setups. Most other games work well 3.0 x8x8 difference would be about 5% to 10% less than a x16 setup but in rFactor2 it would be 50% less?

    As far as resolution right now I'm on a 2560 x 1080 21.9 aspect LG monitor. I plan on waiting until the oculus is fully ready. So the question is will one graphics card be enough to run the oculus properly?

    So the only way to run rFactor2 in a proper 3.0 x16x16 setup is to buy at least the i7 5930k and an appropriate motherboard?
     
  12. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,294
    Likes Received:
    36
    Correct. But just to clarify, this also applies to a single card solution. So if you ran a single 980Ti in 3.0 x8 mode in rf2, you would get ~70-75% the performance of what would otherwise be achievable by a single 980Ti in 3.0 x16 mode. Likewise, when comparing 980Ti SLI setups in 3.0 x8/x8 vs x16/x16, you would get ~70-75% the performance of the latter (x16/x16) with the former (x8/x8) pci-e modes.

    To the part in bold i would say, if even that. I've even seen a video of the latest 3dmark benchmark ran in pci-e 1.0 x16 (equivalent to 3.0 x4 mode) and the performance was basically identical to 3.0 x16 mode. Many games i think are the same. For whatever reason(s), rfactor 2's performance is (at least in my own gaming experience) uniquely and incredibly pci-e bandwidth sensitive.

    Regarding oculus, last i heard (maybe from oculus or nvidia or someone here) is that it should be perfectly suited for SLI because each card would render simultaneously rather than in an alternating manner that can cause micro stutter problems which nvidia address by increasing the frame buffer size for SLI (causing a greater delay from the time the real-time frame is rendered to when you will see it on your display device). Rendering stereoscopic 3d allows each card to render independently of the other which means you don't have any of the potential SLI micro stutter problems so it performs as well in smoothness and lag as a single card solution.

    Spinelli can tell you all about that if you have more questions. He plays rf2 in stereoscopic 3d and says performance and scalability of his two cards is brilliant. But in 2d, i don't know how well it currently works in rf2.

    Unfortunately, yes. If you do, you can expect to get an extra ~30-35% (if not more) extra fps out of your 980Ti solution in rf2. For every other game, it won't make a difference between 3.0 x8/x8 vs x16/x16. Yeah, i know, it sucks to be put in this position.
     
    Last edited by a moderator: Nov 14, 2015
  13. SpeedLab

    SpeedLab Registered

    Joined:
    Nov 25, 2012
    Messages:
    86
    Likes Received:
    0
    @DrR1pper So multi-card in the Z170 system would work good for almost everything except rFactor2.

    Maybe this is a dumb question but why did Intel decide to restrict Sky Lake to only 20 lanes? Will it not be important in the future for multi-card setups to run in full x16 x16?

    If rFactor2 improves whatever it needs to do to run properly in x8 x8 going with regular SATA SSD's you could buy a Z170 system with two 980 ti for $600 less than it would cost you to get a compatible x99 system with two 980ti running in x16 x16.

    Another way to say it is:

    Z170 i7 6700k with two 980 ti in x8x8 $2500

    X99 i7 5930K with two 980 ti in x16x16 $3000

    I didn't really understand your answer about the oculus. Will it be better with one card or two? Will one 980 ti will be powerful enough?
     
    Last edited by a moderator: Nov 14, 2015
  14. SpeedLab

    SpeedLab Registered

    Joined:
    Nov 25, 2012
    Messages:
    86
    Likes Received:
    0
    @ DD You bring up some good points. I was told that Windows 10 installation is about 160 GB. I have no idea how to reduce the size of that or how much you could realistically reduce it.

    I know the bigger solid-state drive the faster it works but why wouldn't it be better to get 2 500gb 850 Evo's. One of them with the OS and then a partition with rFactor2 on the other part. Then you would have a separate 500 GB drive for all your other programs and storage. This would be a Raid 0 setup correct? I didn't think it was a good idea to put anything but the OS and rFactor2 on one drive.

    I could get two of the 500 GB 850 Evo's or a 1 GB 850 Evo for the same price as one of the 500 GB NVMe drives (Samsung 950 Pro)
     
  15. DurgeDriven

    DurgeDriven Banned

    Joined:
    Mar 20, 2012
    Messages:
    6,320
    Likes Received:
    43
    160 GB !@! lmao

    My W10 is well under 20GB mate or 50% free on the partition :)

    Actually W7 is bigger at 19.6GB and both have the identical tools and software added.


    I just thought maybe 256GB would be big enough for your OS and your favourite sims.

    Not the ISI packages you could put those on even a Black Caviar ( Sata ) and go do something if installing alot.

    Put it this way look how long it takes to install content from " remote" no matter what hardrive you have.


    My 250GB Evo ( Disk 1 ) is Partitioned like this.

    [​IMG]


    BACKUP is small partition for most important tools , drivers, docs, photos, screenshots, small playlist all small stuff plus a trueimage of the OS.

    SIMS_SSD is rF2 and other sims

    Most sims I have many multi folders of GTR/GTL/GPL and rF are on the 1TB WD Black Caviar


    My W10 is on a partitioned 60GB force 3 split into 2, it takes up 15GB I think, I will have a look.

    turning off hibernation and system restore, other stuff saves heaps.

    SSD do not need the same free space as a mechanical drive.

    That being said more free space is better then less.


    P.S > it really depends what you want, to add to the OS, is it a sim OS or a "all round" OS ;)

    All other drives ( archives ) are bare which I hotplug into a dock when needed.
     
    Last edited by a moderator: Nov 14, 2015
  16. WhiteShadow

    WhiteShadow Registered

    Joined:
    Feb 16, 2015
    Messages:
    681
    Likes Received:
    3
    If you are going for 60hz monitor(s) single 980ti is all you need to rFactor2.

    I think you find this video intresting. i7-5820K PCIe 8x in SLI vs i7-5930k 16x in SLI
     
  17. SpeedLab

    SpeedLab Registered

    Joined:
    Nov 25, 2012
    Messages:
    86
    Likes Received:
    0
    White shadow I appreciate you for posting this. I'm pretty sure this was covered in the above posts though. For the majority of games and purposes there's no reason to run multi-GPU's in x16x16.

    This was one of the main reasons for moving towards the Z170 platform. The unknown question is will this still be true in the future.

    What I would really like to see is someone from ISI to comment on this. Maybe they have before but I haven't seen it anywhere.

    The big questions seem to be:

    Why does rFactor2 not run well with two cards in x8x8 compared to other games?

    Will this be addressed in the near future or is this what we can expect for the long run?
     
  18. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,294
    Likes Received:
    36
    Ok, so let's start fresh…

    The interesting thing about SLI performance in rf2 if you plan on using an oculus rift is that it should perform exactly the same as it currently does the in rf2 in stereoscopic 3d mode (i.e. using a 3d monitor with 3d glasses) with SLI. Spinelli (who only plays rf2 in stereoscopic 3d with triple screen) has reported near perfect scaling of performance with SLI in rf2 when rendering in 3d mode (which should be equally true if using the oculus rift when it arrives and is natively supported in rf2 since they both rely on the exact same 3d stereoscopic rendering technique). What "perfect scaling of performance with SLI in 3d mode" means is that the performance (i.e. the average fps in game) that you get with two graphics cards (i.e. SLI) in 3d mode is pretty much equal if not identical to the performance you receive with just a single graphics card rendering the game in 2d mode. If you pair two cards (i.e. SLI) in 2d mode, "perfect scaling" would result in twice the performance (i.e. double the average fps) than a single card in 2d mode. However in practice this rarely happens for SLI in 2d mode which usually achieves around an extra ~80% fps when adding a second card. This is typical of the performance gains you get with SLI in 2d mode across the majority of games with functional SLI compatibility (which is a lot these days).

    Now the reason why SLI in 3d mode tends to give you near perfect (i.e. almost 100%) scaling of performance is because to produce 3d you need to render two different and independent virtual viewpoints in game (i.e one for the left eye and another for the right eye, which is like having to render two games at the same time = literally double the work load). SLI is a perfect solution here because each card can be dedicated to the job of rendering one virtual viewpoint only (which is then like having each card run as a single card in a 2d mode of rendering). This means the performance and smoothness will feel as good as having to render only a single viewpoint (i.e. 2d mode) on a single card.

    On the other hand, one of the pitfalls of SLI in 2d mode is that both cards are used to render the same viewpoint which therefore requires alternating the job of rendering the next frame from one card to the other over time. This can produce a good increase in performance (e.g. typically ~80% extra fps) but with it can come other problems such as microstutter and if you've never experience microstuttering stuttering before, just know that microstutter really really sucks. The main way Nvidia addresses microstutter is to increase the frame buffer size so that they can ensure a smoother and more consistent flow of output frames to your display device when using SLI in 2d mode. The big negative of doing this is that increasing the frame buffer size will in turn increase the amount of visual input lag (i.e. the time delay in seeing the real-time frame that you should be seeing on screen right now). When playing on a monitor and in a not so competitive game or one where the amount of input lag caused by 2d mode SLI does not really affect the experience, it can quite easily be forgiven. However, for applications such as the oculus rift, we've been told (or for those of us who have first hand experience will know) this is a big no-no as input lag has a significant impact on the comfort of using the oculus rift (e.g. more than 5 mins in the dk1 and i wanted to puke, dk2 was great though….absolutely no issue of discomfort or long continuous use in live4speed for example). Think turning your head and the visual world updating just a little later than it should. This can induce nausea and discomfort if the amount of delay is more than the threshold of human perception. Oculus have said this threshold is around 20ms, below which the most sensitive people in the population no longer feel discomfort.

    So if you want the best possible performance with oculus (or any other virtual reality headset) in rf2 (or any game), then 980Ti in SLI should definately give you that.

    Where PCI-e mode comes into play is that if you use 2x 980Ti's in 3.0 x16/x16 instead of x8/x8, then you will gain around 30-35% more fps in rf2.

    The question is whether the extra $500 cost so that you can run SLI 980Ti's in 3.0 x16/x16 mode is worth getting the extra 30-35% higher average fps in rf2 compared with performance in 3.0 x8/x8 mode. If I were in your shoes, happy and able to pay the difference and after the best possible experience in rf2 come oculus rift, then i would go with the x99 - i7-5930k option.

    With those cards, you'll most probably be able to run rf2 in max graphics with some heavy downscaling (i.e. render at twice the resolution and then downscale to pixel count of oculus rift) to really reduce the aliasing in the images. This is recommended by Oculus.
     
    Last edited by a moderator: Nov 14, 2015
  19. WhiteShadow

    WhiteShadow Registered

    Joined:
    Feb 16, 2015
    Messages:
    681
    Likes Received:
    3
    I have better solution to all problems with SLI in rFactor and it is much cheaper to all SLI users :)
    http://isiforums.net/f/showthread.php/26985-New-NVIDIA-SLI-profile
     
  20. DrR1pper

    DrR1pper Registered

    Joined:
    Apr 29, 2012
    Messages:
    3,294
    Likes Received:
    36
    It's a good idea if Nvidia are the cause of the/any SLI performance issues in 2d mode in rf2. However, according to Spinelli, SLI in 3d mode in rf2 has perfect scaling and is working fine. Whether that is still true with the latest drivers though, I don't know. MAybe Spinelli can tell us.

    But even if an SLI issue exists in rf2, it won't resolve the pci-e bandwidth issue that rf2's performance is sensitive to (especially with the newer and more powerful the card). Don't quote me on it but if i recall correctly, the same holds true for AMD card owners as well which would suggest it is due to the game engine in some way.

    (p.s. just added one more request for rf2 on the nvidia sli profile request thread you provided :))
     
    Last edited by a moderator: Nov 14, 2015

Share This Page