Compute

Shifting gears, we have our look at compute performance.

As we outlined earlier, GTX Titan X is not the same kind of compute powerhouse that the original GTX Titan was. Make no mistake, at single precision (FP32) compute tasks it is still a very potent card, which for consumer level workloads is generally all that will matter. But for pro-level double precision (FP64) workloads the new Titan lacks the high FP64 performance of the old one.

Starting us off for our look at compute is LuxMark3.0, the latest version of the official benchmark of LuxRender 2.0. LuxRender’s GPU-accelerated rendering mode is an OpenCL based ray tracer that forms a part of the larger LuxRender suite. Ray tracing has become a stronghold for GPUs in recent years as ray tracing maps well to GPU pipelines, allowing artists to render scenes much more quickly than with CPUs alone.

Compute: LuxMark 3.0 - Hotel

While in LuxMark 2.0 AMD and NVIDIA were fairly close post-Maxwell, the recently released LuxMark 3.0 finds NVIDIA trailing AMD once more. While GTX Titan X sees a better than average 41% performance increase over the GTX 980 (owing to its ability to stay at its max boost clock on this benchmark) it’s not enough to dethrone the Radeon R9 290X. Even though GTX Titan X packs a lot of performance on paper, and can more than deliver it in graphics workloads, as we can see compute workloads are still highly variable.

For our second set of compute benchmarks we have CompuBench 1.5, the successor to CLBenchmark. CompuBench offers a wide array of different practical compute workloads, and we’ve decided to focus on face detection, optical flow modeling, and particle simulations.

Compute: CompuBench 1.5 - Face Detection

Compute: CompuBench 1.5 - Optical Flow

Compute: CompuBench 1.5 - Particle Simulation 64K

Although GTX Titan X struggled at LuxMark, the same cannot be said for CompuBench. Though the lead varies with the specific sub-benchmark, in every case the latest Titan comes out on top. Face detection in particular shows some massive gains, with GTX Titan X more than doubling the GK110 based GTX 780 Ti's performance.

Our 3rd compute benchmark is Sony Vegas Pro 13, an OpenGL and OpenCL video editing and authoring package. Vegas can use GPUs in a few different ways, the primary uses being to accelerate the video effects and compositing process itself, and in the video encoding step. With video encoding being increasingly offloaded to dedicated DSPs these days we’re focusing on the editing and compositing process, rendering to a low CPU overhead format (XDCAM EX). This specific test comes from Sony, and measures how long it takes to render a video.

Compute: Sony Vegas Pro 13 Video Render

Traditionally a benchmark that favors AMD, GTX Titan X closes the gap some. But it's still not enough to surpass the R9 290X.

Moving on, our 4th compute benchmark is FAHBench, the official Folding @ Home benchmark. Folding @ Home is the popular Stanford-backed research and distributed computing initiative that has work distributed to millions of volunteer computers over the internet, each of which is responsible for a tiny slice of a protein folding simulation. FAHBench can test both single precision and double precision floating point performance, with single precision being the most useful metric for most consumer cards due to their low double precision performance. Each precision has two modes, explicit and implicit, the difference being whether water atoms are included in the simulation, which adds quite a bit of work and overhead. This is another OpenCL test, utilizing the OpenCL path for FAHCore 17.

Compute: Folding @ Home: Explicit, Single Precision

Compute: Folding @ Home: Implicit, Single Precision

Folding @ Home’s single precision tests reiterate just how powerful GTX Titan X can be at FP32 workloads, even if it’s ostensibly a graphics GPU. With a 50-75% lead over the GTX 780 Ti, the GTX Titan X showcases some of the remarkable efficiency improvements that the Maxwell GPU architecture can offer in compute scenarios, and in the process shoots well past the AMD Radeon cards.

Compute: Folding @ Home: Explicit, Double Precision

On the other hand with a native FP64 rate of 1/32, the GTX Titan X flounders at double precision. There is no better example of just how much the GTX Titan X and the original GTX Titan differ in their FP64 capabilities than this graph; the GTX Titan X can’t beat the GTX 580, never mind the chart-topping original GTX Titan. FP64 users looking for an entry level FP64 card would be well advised to stick with the GTX Titan Black for now. The new Titan is not the prosumer compute card that was the old Titan.

Wrapping things up, our final compute benchmark is an in-house project developed by our very own Dr. Ian Cutress. SystemCompute is our first C++ AMP benchmark, utilizing Microsoft’s simple C++ extensions to allow the easy use of GPU computing in C++ programs. SystemCompute in turn is a collection of benchmarks for several different fundamental compute algorithms, with the final score represented in points. DirectCompute is the compute backend for C++ AMP on Windows, so this forms our other DirectCompute test.

Compute: SystemCompute v0.5.7.2 C++ AMP Benchmark

With the GTX 980 already performing well here, the GTX Titan X takes it home, improving on the GTX 980 by 31%. Whereas GTX 980 could only hold even with the Radeon R9 290X, the GTX Titan X takes a clear lead.

Overall then the new GTX Titan X can still be a force to be reckoned with in compute scenarios, but only when the workloads are FP32. Users accustomed to the original GTX Titan’s FP64 performance on the other hand will find that this is a very different card, one that doesn’t live up to the same standards.

Synthetics Power, Temperature, & Noise
POST A COMMENT

276 Comments

View All Comments

  • stun - Tuesday, March 17, 2015 - link

    I hope AMD announces R9 390X fast.
    I am finally upgrading my Radeon 6870 to either GTX 980, TITAN X, or R9 390X.
    Reply
  • joeh4384 - Tuesday, March 17, 2015 - link

    I do not think Nvidia will have that long with this being the only mega GPU on the market. I really wish they allowed partner models of the Titan. I think a lot of people would go nuts over a MSI Lightning Titan or something like that. Reply
  • farealstarfareal - Tuesday, March 17, 2015 - link

    Yes, a big mistake like the last Titan to not allow custom AIB cards. Good likelihood the 390X will blow the doors off the card with many custom models like MSI Lightning, DCU2 etc.

    Also $1000 for this ??! lol is the only sensible response, none of the dual precision we saw in the original Titan to justify that price, but all of the price. Nvidia trying to cash in here, 390X will force them to do a card probably with less VRAM so people will actually buy this overpriced/overhyped card.
    Reply
  • chizow - Tuesday, March 17, 2015 - link

    Titan and NVTTM are just as much about image, style and quality as much as performance. Its pretty obvious Nvidia is proud of the look and performance of this cooler, and isn't willing to strap on a hunking mass of Al/Cu to make it look like something that fell off the back of a Humvee.

    They also want to make sure it fits in the SFF and Lanboxes that have become popular. In any case I'm quite happy they dropped the DP nonsense with this card and went all gaming, no cuts, max VRAM.

    It is truly a card made for gamers, by gamers! 100% GeForce, 100% gaming, no BS compute.
    Reply
  • ratzes - Tuesday, March 17, 2015 - link

    What do you think they give up when they add DP? Its the same fabrication, was for titan vs 780ti. If I'm mistaken, the only difference between cards are whether the process screwed up 1 or more of the smps, then they get sold as gaming cards at varying decreasing prices... Reply
  • MrSpadge - Tuesday, March 17, 2015 - link

    Lot's of die space, since they used dedicated FP64 ALUs. Reply
  • chizow - Wednesday, March 18, 2015 - link

    @ratzes, its well documented, even in the article. DP/FP64 requires extra registers for the higher precision, which means more transistors allocated to that functionality. GM200 is only 1Bn more transistors than GK210 on the same process node, yet they managed to cram in a ton more functional units. Now compare to GM204 to GK204 3.5Bn to 5.2Bn and you can see, its pretty amazing they were even able to logically increase by 1.5x over the GM204, which we know is all gaming, no DP compute also. Reply
  • hkscfreak - Wednesday, March 18, 2015 - link

    Someone didn't read... Reply
  • nikaldro - Tuesday, March 17, 2015 - link

    fanboysm to the Nth p0waH.. Reply
  • furthur - Wednesday, March 18, 2015 - link

    which meant fuck all when Hawaii was released Reply

Log in

Don't have an account? Sign up now