Great, everybody is Happy with the New Imac Pro , would be Great if a Pro User has a E-GPU via Thunderbold3 connected example Nvidia Titan XP or TITAN V and could share some Infos here about it.
i think the Vega 64 is not the perfect GPU at all.
Great, everybody is Happy with the New Imac Pro , would be Great if a Pro User has a E-GPU via Thunderbold3 connected example Nvidia Titan XP or TITAN V and could share some Infos here about it.
i think the Vega 64 is not the perfect GPU at all.
Have you tried these drivers from Nvidia?The PRO Vega 64 isn't perfect, but it's not a bad GPU either. It's twice as fast as an RX580 in an external enclosure at OpenCL computation, and runs games about 70% faster than a GTX1080ti in MacOS.
I've got a spare 1080ti and an external enclosure (currently housing the RX580). But we don't have eGPU-ready nVidia drivers for High Sierra yet, and even under Sierra the drivers are buggy as hell. I got the RX580 because I was tired of dealing with all the small problems that I kept running into. AMD is what Apple supports and optimizes for now, and they are not cooperating with nVidia's attempts to build a (mostly Hackintosh) MacOS user base.
Does the latest iMac(2017) and iMac Pro support eGPU? I can find successful cases for macbookpro only.
and runs games about 70% faster than a GTX1080ti in MacOS.
great.... Is GTX 1080ti better than titan xp?
i dont game, i allways used TITAN Cards in my builds, the 1080TI is for sure a very good GPU but ill do 4k Video Edit & Cut Daily
And every other video editing app on Mac supports the better cards.Except FCPX, a low end AMD card can do better because of FCPX is very optimised to AMD GPU.
The PRO Vega 64 isn't perfect, but it's not a bad GPU either. It's twice as fast as an RX580 in an external enclosure at OpenCL computation, and runs games about 70% faster than a GTX1080ti in MacOS.
...
great.... Is GTX 1080ti better than titan xp?
Have you tried these drivers from Nvidia?
http://www.nvidia.com/download/driverResults.aspx/125379/en-us
Does the latest iMac(2017) and iMac Pro support eGPU? I can find successful cases for macbookpro only.
Do they have Metal 1 & 2 support yet? They lost that in High Sierra.NVIDIA's Mac drivers work really well right now for all their current cards.
Yes, on a Hackintosh, where they couldn't use Metal, and there was no OpenCL support under High Sierra. I haven't personally tested it, but more than one site has said that those drivers don't work for Pascal cards in external enclosures.
Until Apple decides to play nice with nVidia, don't expect this situation to change much.
VEGA (56 or 64) is only about ⅓ faster than the RX580, not twice.
[doublepost=1514910401][/doublepost]
Xp is indeed faster than 1080TI, but the ROI is small.
If there's the budget for that, go for it.
ProVega64 runs the Heaven benchmark almost exactly twice as fast as an RX580 on the same machine. Though I am using an external enclosure for the RX580 which could be gimping it a bit. Certain computational tasks also run about twice as fast. Others are more borderline.
The price premium for the ProVega64 is indeed high, but this is a workstation GPU with a full 16GB of RAM – which is important for 3D development. If you want a performant GPU under MacOS that is properly supported and optimized for, the ProVega is it. I can wish for nVidia all I want, but it's not a reliable or very performant solution under MacOS, and I already have a Windows PC with two 1080ti's for when I need their power.
Also, comparing dGPU to eGPU is not a very fair comparison.
e.g. My 1080Ti work very flawless under MacOS indeed. Everything is supported, extremely fast. But I am running it internally in a Mac Pro, not as eGPU. For you, the same 1080ti running as eGPU can be completely useless in MacOS.
It's the same card, same OS, but internal / external make all the difference.
In my above post (#17), as you can see from the graph. Running the same GPU internally (41 FPS on a Mac Pro 5,1) can have 2x performance that the same GPU work as eGPU (only 19 FPS, even on a faster / newer Mac Pro 6,1).
So, if you comparing your internal Vega 64 to an eGPU RX580, and then conclude the Vega 64 has 2x performance of the RX580. This is not quite valid.
I built a hackintosh this past year (now only a Windows PC) with 2 GTX1080ti's in it. They were not fully supported (No SLI, I couldn't stream any video from iTunes, needed a 3rd party app to enable retina support on my 5K monitors, etc.), and while they ran games just fine, they did not improve the real-time display in my 3D apps appreciably over the R9 M395X in my late 2015 iMac (yes, really). And when I updated to High Sierra? No more CUDA, no Metal support. Some of these things may now have been fixed by nVidia or the Hackintosh community, but I don't consider nVidia to be an option in MacOS that I can currently depend on when I need it. This is why I finally caved and bought the iMac Pro. I got sick of constantly fixing little things instead of focusing on my work.
As for external vs. internal GPU benchmarks... An external enclosure is the only way I currently have to measure the RX580's performance. Fact is, I'm less concerned about it's raw performance than the fact that pairing it with the ProVega for OpenCL computations caused those computations to take longer than with the ProVega alone. I really hope that changes with the release of "official" eGPU support.
All I can say is both Hackintosh and eGPU are never really that reliable. That's why we (the Mac Pro users) really wish Apple will release another Mac Pro again with the PCIe slots.
SLI is a Windows only features. But even in Windows, it's still cause lots of issue.
Anyway, enjoy you iMac Pro. It's definitely a powerful machine. My last post mainly want to point out Metal and CUDA can work on Pascal card. For internal dGPU, it can work flawlessly. For eGPU, the method is developed, but not yet publish. Also, may be it's more appropriate to clearly state "Internal Vega 64 is 2x faster than eGPU RX580", but not simply "Vega 64 is 2x faster than RX580".
I did say it was running in an external enclosure when I talked about the relative performance. I was even specific in saying it was an OpenCL test. I guess you missed it.