GTX 1080 compute performance: http://www.overclock.net/t/1600173/compubench-gtx-1080-and-gtx-980ti-compute-performance
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
nVidia GTX 1080
- Thread starter AkuskaUK
- Start date
- Sort by reaction score
You won't see that on the cMP. Older generation Nvidia cards score higher in Windows benchmarks by as much as 30% Pascal now has proper support for Asynchronous Compute, but without a true native driver you won't have that feature on OSX.
Which API exposes asynchronous compute on OS X? I'm confused as to what you think NVIDIA should be doing to expose new functionality in Maxwell or Pascal when they have no control over things like the OpenGL, OpenCL or Metal frameworks (i.e. the things that define the external interface for applications)?
Found a nice new video about how Nvidia develops and tests their GPUs. Each architecture is first emulated entirely in software running on massive servers. The servers themselves are then connected to PCIE slots as an "emulated graphics card" in various test bed systems including Windows and Linux PCs and of course they have cMPs in the lab too.
Interesting.... But that's not the impressive stuff.
I've been in the Nvidia EDA labs. About a hundred racks, full of Dell 2U servers packed 18 per rack. Next to the lab are a pair of power/cooling buildings. Redundant. Each building could power and cool the lab by itself. And by "power and cool", that means each building had diesel backup generators to power the building, A/C and the lab.
These are the computers used to design the chips. The video is about the labs that emulate the design for verification.
That explains the ridiculously high cost of video cards, and the fact that they never seem to truly go on sale until years down the road!
No, those are explained by "people are happy to pay".
And I'm surprised that someone paying the Apple Tax would complain that something is overpriced.
I don't know about happy to pay. Willing perhaps. Begrudgingly, definitely!
Nvidia will be forced to lower their prices the minute AMD produces something competitive.
It's been a few years but you never know.
It's been a few years but you never know.
Interesting that they have Mac Pros there, I wonder if that was a GP/P1xx series card they were testing with it? I assume they'd use them for OS X, though of course they'd work perfectly well for Linux or Windows.
Lends credibility to the hope that they'll release a new OS X web driver to support Pascal cards, possibly coinciding with the release of OS X 10.12.
Good news (potentially) for Mac Pro 2009/2010/2012 owners, though confusing, given Apples recent shift towards AMD. Perhaps they're always developing options for Apple to look into to tempt them away from their AMD exclusivity.
Last edited:
Apple's recent shift towards Nvidia? Did I miss some news?
Most likely the system is there to support the GPUs they have officially licensed for the Mac. Beyond Kepler progress will be slow and can only come if Apple still has an open door for them. I'm stating the obvious.
I will be getting a Pascal GPU to test in the cMP within the next month and report back all benchmark tests (if it works), though I stress that I will likely go for the 1070 as it strikes a great balance for me. Though I'm an Nvidia guy, I'm under no illusion about Pascal. It has great performance due to its high clock speed and efficiency, but clock for clock it is slower than Fury cards. So if Polaris delivers higher clocks and efficiency then AMD could win the crown this year.
If it the 1070 doesn't work in the cMP it goes in my Skylake PC anyway and I will test it against the 680 in the cMP just for fun.
Last edited:
Sorry meant AMD, corrected my post.
[doublepost=1463699135][/doublepost]
I'm not so sure it's that rigid. If you're Nvidia, you're making things like the 1080/1070 anyway; why not test them on OS X, provided you have the hardware. Admittedly, the pre-2013 models form factor is now consigned to history, but there's still Thunderbolt 3 for eGPU things that OS X may well wish to leverage in the near future?
In the very interesting video you provided, they clearly stated that area was a lab they tested new hardware designs in; Kepler has been and gone, why would they be testing that in 2016? I don't think they were at all, I think that's more of a 'Hey Apple, you might not be selling anything with an Nvidia card in but the current ones run pretty well all the same'. That's got to be a good way to win people like Apple and other system builders over, surely?
Admittedly, that's all conjecture; as I said, they could be using those Mac Pros purely as Xeon workstations they had lying around for PCIe 2.0 ports, but who knows, right? I mean at the very least, for them to test their web drivers on the unlisted cards it supports (7xx/9xx series), they'd need hardware to actually test that on? Perhaps that's what we're looking at.
All we need to do now is convince an OEM to actually make a new Mac Edition; EVGA for example. I've emailed them before about the 980, and recently emailed them about the 1080, and their response was typically vague and non-commital, but still.
Won't happen at all unless Apple listens and brings back a system with PCIe slots.
Which means "Won't happen".
Unlikely, certainly. Impossible, not so sure. Don't count eGPU out. A new Thunderbolt Cinema Display could have an embedded eGPU for Laptops/Mac minis to benefit from if bought in tandem. eGPU support would give an avenue for people without available PCIe slots room to manouver, which in turn would require a Mac EFI, and could also possibly work with Razors eGPU container as an example.
Because the Kepler is still being supported in driver updates and many still have warranties. I have no doubt they would be testing newer architectures but they don't appear to have assigned much budget or manpower to Mac driver development as they haven't fixed some simple but crucial OpenCL and GUI bugs for many months now.
For testing things like Thunderbolt 3 GPU, the OS has to support hot plugging and true plug n' play. OS X is still far behind in that respect. Even on Windows it isn't quite mature. Disconnecting a GPU can often freeze a system.
I think you missed my point; why would you have Mac Pros in a room to test new hardware, when officially the only Nvidia Mac card is Kepler. It's been around since 2012, I'm sure they'd have some 680s lying around.
I think it's more likely those machines were testing Pascal on OS X. Obviously I could be wrong, but I've no idea why you'd bother wasting so much computing power emulating a product that was finished and released 4 years ago.
If I get it right, Soy means Nvidia is testing the new hardware, since they have to test the Kepler anyway, the Mac Pro is already there, why not? But they won't really spend lots resource to build the required software.
So, what he said is not contradicting to what you said (both agree that Nvidia test the new hardware on Mac Pro).
IMO, no matter will they release anything for OSX now, it make sense to test the system. Nvidia is a GPU manufacture, Apple is one of their protential customer. It make perfect sense to build and test the new hardware on the OSX side. And use the data on the next business meeting / negotiation.
If they have zero data on how Pascal GPU perform in OSX. How can they get the contract from Apple?
On business point of view, Apple is a big business partner. Why not do something to collect the useful data and get this protential contract? The testing facility is there anyway.
However, it doesn't mean that they will release anything for the OSX user (even though I personally believe that the coming web driver will support the Pascal card).
Given Apple's history of maintaining prices, while upping specs as technology advances, I wouldn't be surprised if the eGPU-based Retina display comes out at a comparable cost to the current Thunderbolt display.
*edit* The market size would be any Mac with a TB3 port, which can't drive a dual cable solution like the Dell 5k due to a lack of ports, or intel graphics. TB3 doesn't have a Displayport version for single cable 5k (someone please correct me on that, because I'm hazy on it), which could mean the only way to drive that display on a single cable is by the display being an eGPU.
The real question is whether eGPUs over TB3 will be a whitelist-only OS feature.
Last edited:
Yeah there has to be one branch of driver development that is always for testing. The other branch is to support officially released and licensed products. But the web driver has to be ported from the Windows code base. Unless the web driver makes a sudden big jump from version 346 to 368 (official support for Pascal in Windows) then it's a long wait, if ever.
If you watch the video again, he mentions testing on PC and Linux before finally and kind of reluctantly mentioning the Mac as an after thought. It's not that hopeful.
As usual, you don't know what you are talking about. Hope you'll admit it this time when we have Pascal running on Macs before Polaris.
Why not just post about stuff you actually know?
Given the way things have panned out historically, I think that this is likely. With Pascal's (paper) release next week, we could potentially see web drivers in the near future (a month or so perhaps?) or Nvidia might just wait for 10.12, which would probably be around September.
Polaris on the other hand will only be supported if (or perhaps when) Macs are released with it. Given how Apple have been with the Mac Pro in the past, I think it's far more likely the new GPUs will be custom Fiji boards, which would realistically put Pascal support to next year at least.
