Compute with the Surface Book

When discussing Ultrabooks, the word Compute doesn’t get thrown around very often, and for good reason. Even the MacBook Pro 13 only comes with Intel Iris graphics (no GT3e yet) and although Intel’s GPUs have been a priority over the last couple of generations, just like in gaming there is only so much you can do when your TDP is shared with the processor.

With Surface Book, there is more of an opportunity here. If you opt for the model with the NVIDIA GPU, you gain access to CUDA, which is NVIDIA’s parallel computing platform. Quite a few applications that need strong parallel processing have CUDA available as an option. Adobe, for instance, has CUDA support in many of their professional products like Photoshop, After Effects, Premier Pro, and more. NVIDIA lists hundreds of applications on their site which can benefit from GPU compute power, and there are also OpenCL applications as well which would benefit from the more powerful dGPU.

Expectations need to be put in check of course, because the GPU available in the Surface Book is not a workstation class GPU, so we shall see how it compares on these types of tasks. This is not an area where we have an extensive database of other devices, and normally compute is not a heavy focus for Ultrabook reviews, but I feel the Surface Book may find a niche with content creators so it’s worth examining.

Compubench

From the makers of GFXBench is Compubench, and like GFXBench, there are a number of tests which can be completed with either the CPU only, or by choosing a GPU.

CompubenchCL Face Detection

CompubenchCL TV-L1 Optical Flow

CompubenchCL Ocean Surface Simulation

CompubenchCL Particle Simulation 64K

CompubenchCL TRex

CompubenchCL Video Composition

CompubenchCL Bitcoin Mining

The results are a bit mixed. Some of the tests respond very well to having the NVIDIA GPU, but some of the others don’t get as much of a benefit. But where the GPU helps, it can help a lot. Several of the tasks are 50% faster, and the Video Composition sub-test is 212% faster on the discrete GPU.

Agisoft Photscan

This software performs photogrammetric processing of images, and it has an option to use the GPU or just standalone with the CPU. Of the entire benchmark, only one section actually leverages the GPU functions so that test has been highlighted.

Agisoft Photoscan Stage 2

Even the one accelerated test still only shows a 5% decrease in time with the GPU being used. This highlights that even though a task may be accelerated with the GPU, the overall impact may not always be what you are expecting, since not all tasks can be done in parallel.

Using the Surface Book NVIDIA GPU for Compute

There is no doubt that if you are performing work that supports CUDA, the NVIDIA option on the Surface Book is going to make an impact. The question of course is how much. Applications such as those from Adobe do leverage CUDA, but it’s not for all tasks. This is kind of the issue with considering the GPU for compute. If you are someone who uses Adobe Premiere on the go, and need something smaller than a typical workstation class notebook, the GPU is going to help out, but since it doesn’t get leveraged for all tasks, it is very dependent on the exact task that you are performing.

GPU Gaming Performance The PixelSense Display
Comments Locked

249 Comments

View All Comments

  • Brett Howse - Tuesday, November 10, 2015 - link

    We haven't had anything that is recent enough to include. It would be disingenuous of me to include a MacBook Pro from 3 years ago, or a MacBook Air from 2 years ago.
  • PalmKing - Tuesday, November 10, 2015 - link

    Thanks Brett. What a wonderful review. Very crisp and precise, yet not too technical. Great job! Best review i have read for SB - worth the wait. Thanks.

    I am sold on buying an a SB! Though I wish I could afford the i7 with 512GB, I may have to settle with i7 with 256GB model.

    2 Questions:
    1) Is the hard drive significantly slower on 256GB than 512GB (should I be concerned)?
    2) Is there a way to tell if I am getting a Samsung versus Toshiba hard drive when I go buy from the MS store?
  • Brett Howse - Tuesday, November 10, 2015 - link

    With half of the dies of a 512 GB drive, write speed is about half. See the SP4 review here:
    http://anandtech.com/show/9727/the-microsoft-surfa...

    Read speeds are still very good.

    As for Toshiba vs Samsung, the only way I know to tell is to go into Device Manager and see what you get.
  • pixelpusher - Tuesday, November 10, 2015 - link

    Something I have found as a great benefit to having the intel part in the middle of the tablet is the lack of heat in the bottom of the keyboard that normally comes from my retina macbook. MS did a great job of getting heat dispensation away from your lap, which makes it comfortable to use in bed or when laying on a couch. The trackpad rocks!!!
  • tipoo - Tuesday, November 10, 2015 - link

    The thermals impress me too. If you're not using the dGPU, the keyboard may well stay perfectly cool the whole time. Plus no throttling, in an ultrabook size, with dGPU performance. That large passive heat block for fanless cooling is neat too.

    Quite the thermal marvel really.
  • id4andrei - Tuesday, November 10, 2015 - link

    If you think about it, the detachable hybrid factor is starting to make more sense than the traditional notebook factor.

    I mean, first it allows for hybrid use. That's a plus. Then, it allows, no, it begs for the thermal load to be split in two. This leads to efficient cooling, more battery space, and more "powah" in the form of a dGPU otherwise not possible in such a small frame. All similarly sized machines have Intel graphics.
  • chlamchowder - Tuesday, November 10, 2015 - link

    Does anyone else find it fascinating that the GPU is hot pluggable?

    Getting a PCIe device to be hot swappable sounds like a challenge, and is likely even more of a challenge with a GPU. What would happen if a program was using the discrete GPU, and then the user decides to detach the tablet? Seamlessly migrating program state over to the IGP sounds...hard.

    But maybe that's part of why the software isn't 100% solid yet.
  • vithrell - Tuesday, November 10, 2015 - link

    Yes, device alone - as unique as it is - is overpriced for what it has to offer. Technology used to make it work could be used in many other cases like GPU powered docking stations for laptops and tablets or additional displays with built-in GPU. I hope OEMs will use it in their constructions.
  • ToTTenTranz - Tuesday, November 10, 2015 - link

    The GPU is using PCIe x4. Together with the ports that are present in the keyboard, I suspect that connection is actually Thunderbolt v3 under disguise.
  • DanNeely - Tuesday, November 10, 2015 - link

    It's not fully hot plugable. You can only separate the two halves if nothing is using the GPU. This has been an Optimus feature from the start; I'm not sure if anyone else has sold a product that uses it, but there are nVidia demos on youtube where they stop their GPU in software and pull an MXM card out of a dev board.

Log in

Don't have an account? Sign up now