Trending

Hands-On With Nvidia's Titan X (Pascal) In SLI

By

We test VR, SLI, Nvidia's new high-bandwidth bridges, and the concern that HBM2 might have been a better choice for Titan X.

The Lack Of HBM2 Is No Big Deal

Given AMD’s use of HBM on its Fiji-based cards, hardware enthusiasts are quick to question Nvidia’s lack of HBM2 on the Titan X. The company instead chose to give this card a massive 12 GB of GDDR5X with a nominal memory bandwidth of 480 GB/s. That's a 42% increase over the previous-generation Titan X that offered 337 GB/s.

While it is true that HBM2 facilitates a bandwidth advantage over GDDR5X, the benefits of more throughput would only become apparent in situations where on-die resources aren't being fed fast enough. The concept is akin to PCIe 3.0/2.0/1.1 links between CPU and GPU, and you can already run any modern card with four lanes of third-gen PCIe with little or no performance degradation. Bumping up that bandwidth with eight or 16 lanes yields a 1% to 2% performance increase on average, and anything beyond confers negligible gains.

Below you can see a short run of The Witcher 3 running on two Titan Xes in SLI at 5K (5120x2880p).

Notice that the bottleneck (as you would expect) is on the GPU’s front end. The memory controllers are busy 50-60% of the time, indicating plenty of available headroom with these GDDR5X modules. Would this card be faster if equipped with HBM2 instead of GDDR5X? We down-clocked the memory by -20% and observed a -4% impact on performance in 3DMark. So, more bandwidth might be good for another percentage point or two, at best.

22 Comments Comment from the forums
  • ryguybuddy 19 October 2016 14:05
    I would like to see another follow-up with Titan X's in a watercool loop and a 6950X or something to see how much of an improvement that would make.
    Reply
  • clonazepam 19 October 2016 15:00
    I vaguely remember reading a different article where the two types of SLI bridges were x-rayed and examined much more closely. IIRC, there's very little difference. The newer one has the traces all set to equal lengths (the click of a button in design software), and one pin I think is changed to Ground which allows the driver to detect its presence. I think they're physically equal in otherwise, besides trace length, LED, and the change to ground.

    Anyway, I've done the multiple card setups for many generations, but game development is going in a direction where that's no longer a worthwhile endeavor as support dwindles more and more.

    Thanks for the read.
    Reply
  • Realist9 19 October 2016 15:26
    This is an interesting experiment 'for science'. But that's all. SLI support in games is so spotty, unreliable, and sometimes non-beneficial, that it is now, IMO, irrelevant.
    Reply
  • Compuser10165 19 October 2016 15:50
    I think going X99 with such a system is a requirement in order to reduce the cpu limit factor. Desite the 10 core 6950X, an OCed 5960X (to about 4.4-4.5ghz) should be enough for such a setup. That is because Haswell -E is easier to OC at higher frequencies versus Broadwell-E.
    Reply
  • ledhead11 19 October 2016 17:31
    Thanks for the review. It's nice seeing these SLI reviews especially since there are so few for either the Titan X or 1080's.

    When I got my 1080's I tried the EVGA bridge and had problems with getting full contact on my cards. Some boots would show the cards, some didn''t, so for a time I used dual ribbons until I got the NVidia HB a week or two later. The NVidia worked for me no problem. The main difference I noticed was, a few more FPS here and there but really a more stable, consistent frame rate. I read the same article about the x-ray comparisons as well before purchasing and have to say all this info is getting pretty consistent.

    I can tell you that 1080SLI has very similar performance behavior as to the reviews of Titan X SLI I've seen. Both SLI setups seem to really shine in the 4k/60hz or 1440p/144hz. When I tried DSR 5k on my 4k display the frames quickly dropped to around 40fps.

    I'm not really seeing the CPU bottleneck you mention except for the Firestrike tests. Whether 4k/60hz or 1440p/144hz my 4930k @ 4.10ghz rarely goes above 40%.

    I completely agree with you about what to use the Titan's for- 4k/5k all the way. 1080SLI just starts to hit a ceiling at around 60-80fps in 4k and averages 100-150fps in 1440p depending on the game.
    Reply
  • ledhead11 19 October 2016 17:35
    Oh and I forgot to mention, the most noticeable FPS increases with the HB bridge also happened at 4k for me.
    Reply
  • niz 19 October 2016 19:14
    I'm very pleased with my (P) Titan X, I upgraded from a 980GTX mostly for VR (Vive) and to be able to play Elite Dangerous at VR Ultra with superscaling set to 2.0, which it does fine now and is so rock solid I just leave SS set to 2.0 for everything in VR. With E:D in VR there's an occasional chugging but it seems to only be during loading screens (i.e. as a result of file I/O) so it seems that theres a real chipset issue that I/O can steal damaging amounts of bandwidth from something (PCIe bus?) rather than the CPU or GPU being maxxed out.
    I have an acer X34 monitor so 3440 x 1440, my PC is running a i7-6700k at stock speed. I keep thinking about blowing another $1200 just to go SLI just because "moar is moar" and just the thought of Titans in SLI give me a nerd boner, but it honestly seems like I'd see no noticeable benefit.
    Reply
  • RedJaron 19 October 2016 20:56
    18749995 said:
    I think going X99 with such a system is a requirement in order to reduce the cpu limit factor. Desite the 10 core 6950X, an OCed 5960X (to about 4.4-4.5ghz) should be enough for such a setup. That is because Haswell -E is easier to OC at higher frequencies versus Broadwell-E.
    Just because a CPU can be clocked higher doesn't mean it performs better. Improvements in IPC and efficiency more often than not make up for the lower clock speeds.
    Reply
  • somebodyspecial 20 October 2016 05:38
    Skipped the article after looking at contents. Still waiting on PRO benchmarks for this card to see how it does vs. other top cards in stuff like adobe (cuda vs. amd's OpenCL or whatever works best for them), 3dsmax, Maya, Blender etc. These cards are not even aimed at gamers, so wake me when you test what they were designed to do (pro stuff on the cheap), so content creators have some idea of the true value of these as cheap alternatives to $5000 NV cards (P6000, etc).
    Reply
  • rcald2000 20 October 2016 15:23
    + Filippo L. Scognamiglio Pasini
    Since you noticed two Titan XP's achieving only 50% utilization when paired with the i7-6700k, then which overclock CPU would diminish or eliminate that condition? i7-5820k (CPU I own) or an i7-5960X? Any other recommendations would be appreciated. Thank you for the great article.
    Reply