Thanks for the link. It will be a squeeze, though - especially if any bus-powered devices are in use with a Vega 64 with 16 GiB HBM2.
[doublepost=1501858248][/doublepost]
RJ45 https://www.apple.com/imac-pro/specs/
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.
The Vega RX Thread (Rumors and Info)
- Thread starter fendersrule
- Start date
- Sort by reaction score
[doublepost=1501858248][/doublepost]
If you are thinking something like the 60+ W required for charging laptops over thunderbolt 3, I don't think this will be supported. This is not a requirement for TB3.
I haven't been able to figure out if the 11 TFLOP number that Apple advertises for the iMac Pro is for Vega 56 or Vega 64. Everything other spec on that page says "up to" but not for the GPU.
It does seem to be a USB-C requirement, though, to support bus-powered 3.5" drives and similar medium wattage devices.
That's assuming that the USB-C connectors are true USB-C, and not just T-Bolt3 using the C connectors.
USB-C requirements are a subset of TB3. I'm sure there will be enough power for bus powered HDDs and SSDs.
If you think we can get our hands on that beast for less than $3k, I want to live in your world! LOL
[doublepost=1501868807][/doublepost]You guys have seen this video, correct? I believe this is what Koyoot is arguing. Give the drivers and Bios's time to mature. AMD is NOTORIOUS for having horrendous drivers at launch. I for one, would love for to see Vega improve over time. I plan to purchase one for an eGPU box for my son and I believe this will be the card shipping with Apple's modular Mac. I DO, however, hope that modular means PCIe slots because I'd like to also be able to put a GTX 1080 in my Mac as well. Why because choice is king.
Consumer volta is not going to be far behind. Its reasonable that GV102/Titan V/GTX 2080 Ti could be a 15+ TFLOP card given that GP102 is only a 470 mm2 die and they could probably get up to 600 mm2.
This AMD 'fine wine' never appears to come to macOS. Maybe because we have only had lower end chips. But normally basic performance is locked in at an early stage and no more optimisations happen after.
Do you have any evidence of this? Often time's gaming websites will compare driver performance over a period of months but I have never seen something like this for macOS. AMD's drivers do get updated occasionally with macOS point releases. I doubt these releases are decreasing performance. Also APIs like metal 2 certainly offer enhanced performance potential.
Maybe because Apple plays a big part in implementing OpenGL under OSX.
What I am saying has nothing to do with this. Its time to give up believe that Drivers have anything to do with Vega performance right now.
SOFTWARE has to be rewritten, or simply have to have added specific code, that allows features of Vega hardware to be used, and increase its Geometry Throughput, and allow for higher load balancing, and better overall scheduling.
Obviously, AMD Vega still requires optimization of the drivers, because the pipeline is longer and the instructions have to be dispatched with this in mind, and software has to "know this", but this is job of drivers. Drivers alone can make the Vega GPUs at best 15-20% faster. Hardware features: Primitive Shaders, Programmable Geometry Pipeline, etc can make the GPU 50% faster, on top of that, because Primitive Shaders increase the Geometry Throughput twice, compared to "normal circumstances".
Understood. I think I understand what you're saying now.
[doublepost=1501872724][/doublepost]
COULD be...
I have not seen so far the white paper of Vega, nor ISA instruction set, of Vega. When I will see it, I will tell.
On Barefeats we have same cards tested again and again against newer cards and newer macOS. The numbers appear to stay the same until end of support.
Always test the same API across time to see if there are driver improvements. Never switch from one API to another otherwise it's a deceiving result.
You're right - USB-C requires only 15 watts (3A @ 5V) per port. Good enough for 2.5" spinners but a bit short for 3.5" spinners.
The optional "Power Delivery" specification raises that to 60 or 100 watts (3A@20V or 5A@20V). https://en.wikipedia.org/wiki/USB#Power_related_specifications
Let me make a helpful table to illustrate where Nvidia's GPUs will likely end up.
Pascal Generation:
GP100 - 3584 cores - 610 mm2 - 10.6 TFLOPS SP, 5.3 TFLOPS DP
GP102 - 3840 cores - 471 mm2 - 12.1 TFLOPS SP, 0.17 TFLOPS DP
Volta Generation:
GV100 - 5120 cores - 815 mm2 - 14.9 TFLOPS SP, 7.4 TFLOPS DP
GV102 - (Estimated) 5120 cores - ~600 mm2 - 17.2 TFLOPS SP, ~0.25 TFLOPS DP
We have a 14% increase in SP TFLOPS between GP100 and GP102, applying the same to GV100 we get 17.2 TFLOPS for GV102. See, not too much of a stretch given what we know about Nvidia's past behavior and what we know about Volta.
I don't see any evidence of that. Their methods change settings and games too much to see any trend. We need something like this.
16 nm process, marketed by TSMC and Nvidia as 12 nm does not allow for increased core clocks. It is process node optimized for density.
At 600mm2 you can expect at best 14 TFLOPs of FP32. 815 mm2 with just 4 HBM2 memory controllers, which are smaller than GDDR5/X/6 memory controllers. GV102 will have GDDR6, with 384 bit memory bus, so you have to take account for this. So they will also take out the space, for CUDA cores. Most likely 600 mm2(if it will be this "small") will bring you 13-13.5 TFLOPs of FP32.
There is obviously possibility that Nvidia will make on standard TSMC 16 nm Finfet process, consumer Volta GPUs. But I would not expect than bigger than 600 mm2 die size, and 13 TFLOPs of compute power in slightly lower than 300W TDP, with it.
In Nvidia Volta thread I have posted one link to that proved this.
https://www.semiwiki.com/forum/content/6662-tsmc-talks-about-22nm-12nm-7nm-euv.html
https://www.extremetech.com/computing/245880-rumor-nvidias-volta-built-tsmcs-new-12nm-process
[O]ur strategy is continuously to improve every node in the performance, such as 28 nanometer. And are continuing to improve the 16 nanometers technology. And we have some very good progress, and you might call it the 12 nanometer because we’re improve in the density, classical density, performance and power consumption. Yes, we have that.
http://www.eetimes.com/document.asp?doc_id=1331489&page_number=4
Few quotes from the article:
TSMC plans a six-track 12nm FinFET process that falls between its 22nm planar and 10/7nm FinFET nodes. It sees it as an upgrade for midrange mobile and video processors and high-end IoT devices now using its 16FFC process. High end GPUs and FPGA are using a 16FF+ variant and will migrate to 10/7nm.
Compared to 16FFC, 12nm chips running at less than 2.4 GHz could see 20 percent area shrinks. Those running faster than 2.4 GHz could be optimized for an additional six percent in speed gains.
They used the same Luxmark and OpenGL titles such as Dirt, X Plane and Tomb Raider for 2-3 years. They're still using most of the same titles.
That covers your first paragraph - I'd like to see links for the speculations in your second paragraph.
The third paragraph is clearly opinion, no explicit need for links - although it would be nice to explain how you were reasoning..
Fine, lets try another way to estimate performance.
GP106 - 200 mm2 - 5120 GFLOP - 25.6 GFLOP/mm2
GP104 - 314 mm2 - 9216 GFLOP - 29.4 GFLOP/mm2
GP102 - 471 mm2 - 12150 GFLOP - 25.8 GFLOP/mm2
So without thinking about GP100 or GV100, lets see what we get with a potential GV102.
First, lets be conservative and say we keep the ~25 GFLOP/mm2 of GP102
GV102 - 600 mm2 - 25 GFLOP/mm2 - 15000 GFLOP
Or lets say that Nvidia does a little better and manages 27.5 GFLOP/mm2
GV102 - 600 mm2 - 27.5 GFLOP/mm2 - 16500 GFLOP
Nvidia is relatively predictable in what they do with their GPUs. Except it was very surprising to see them announce a gigantic 815 mm2 GPU. Don't forget that Nvidia could go bigger with GV102, since obviously GV100 is huge.
Math, not fanboyism, is the best way to project their next generation GPUs.