yes but single slot and only 150mm long
i find this good because the powercable is pointing back instead of upward so even with the power cable plugged in it won't extend beond the motherboard
As HBM has been increasing bandwidth per pin, so has GDDR. Currently, you need to be packing 3 or more HBM stacks onto your interposer before you actually have more bandwidth available than a GDDR array. Any fewer, then you're still paying the premium for more expensive die stacks and fabbing an interposer, but not gaining any performance.
for power consumption you could take th nvidia quadro gp100 (235w) that used hbm2 memory and compare it against the gtx 1080ti (250w) wich has the same amount of cuda cores BUT has only 11gb of gddr5 memory. But there might be other differences that contribute to the increased powerdraw. So still no apples to apples comparison.As HBM has been increasing bandwidth per pin, so has GDDR. Currently, you need to be packing 3 or more HBM stacks onto your interposer before you actually have more bandwidth available than a GDDR array. Any fewer, then you're still paying the premium for more expensive die stacks and fabbing an interposer, but not gaining any performance.
In terms of power consumption we can't really do an apples-to-apples comparison (Vega and Fury are far behind Pascal in perf/watt, but we don't know how much better/worse that situation would be without a non-HBM Vega to test).
In terms of PCB area, the advantage is present but not huge. Compare the Vega Nano mockup to the 1080 Mini:
(Alignment done visually using the PCIe card-edge and retention bracket screw. Don't use it to design a case!)
The height increases by ~7mm (though unlike Fury Nano, the Vega Nano has lost the end-positioned PEG connector), and the length by ~20mm, but the 1080 Mini is still the same length PCB as an ITX board. The Nano's cooler is smaller, but also rated to dissipate 40W less heat (150W vs. 180W). Depending on how well Vega scales down in performance as it's undervolted and downclocked (and by how much the finalised Vega Nano is cut down on-die), it may be that a more realistic comparison is a 1070, 1060 or even 1050Ti. The 1050Ti is available in a half-hieght form factor, though at only 4G ram (though at those lower performances it may not make a practical difference), but I don;t recall there being any 1070s or 1060s all that much msaller than the 1080 Mini (the stock 1060 is slightly shorter, the PCB being flush with the top of the bracket, but that's about it).
The PCIe Quadro GP100 has a boost clock of 1480MHz, while the 1080Ti has a boost of 1582MHz. 235W:250W is roughly the same ratio as 1480MHz:1582MHz (1.064 and 1.070), so that could cover all the power delta. It's still not quite a perfect comparison: the 1080Ti will clock even higher when it has the headroom, and the GP100 has a bunch of extra die area for to FP64.
The GPU package itself is too big for that I think. Like, physically wouldn't fit on a low profile PCB.
Made a 250mm long Vega Nano mockup.
No more than a modified Fury X model, but assuming that the Vega Nano has the same dimensions, this should be very doable. Of course, it uses a Vega 56 chip and I put 4x8pin for a dual 64 chip because why not
oh nuts, I hope one doesn't have to undo the block just to get to the SSD part of the SSG o_o
Everything I have read says the AMD SSG cards are Polaris or Vega cards with 2 M.2 NVMe slots supporting up to 2TB of storage.
I meant like if a full block also cools the m.2 drives, would it (most probably) restrict access to the SSDs because a waterblock is in the way?
I remember seeing a pic of a prototype with Samsung Pros in them, but apparently google lost the image ._.
AMD is a strong proponent of what they call a Unified Memory Architecture where the CPUs and GPUs all share memory addresses, so any of them can access the memory attached to other devices (rather than how it is traditionally done where the blocks are copied from one memory space to another). Basically it's like how multi socket CPUs can access each other's memory, but extended to other types of processors, so it doesn't make sense for them to restrict that access behind some sort of proprietary protocol.
