Thread replies: 21
Thread images: 1
Thread images: 1
File: 1495147119764.jpg (390KB, 1365x2048px) Image search:
[Google]
390KB, 1365x2048px
I'd like to build a Deep Learning Rig using a Ryzen X1800 and 4 GPU cards (either Nvidia 1080ti or AMD RX580).
My question is, how do I get four cards working on an AMD Ryzen board that may have enough PCIE slots but not enough room, i.e. using risers?
Thanks in advance.
EVGA DG-86 Gaming Case, 100-E1-1014-K0, Full Tower, K-Boost, Software Fan Controller
EVGA DG-86 Gaming Case, 100-E1-1014-K0, Full Tower, K-Boost, Software Fan Controller
Model #:100-E1-1014-K0
Item #:N82E16811205017
$199.99 -$50.00 Instant $149.99
4
Update
GIGABYTE AORUS GeForce GTX 1080 Ti DirectX 12 GV-N108TAORUS-11GD 11GB 352-Bit GDDR5X PCI Express 3.0 x16 ATX Video Card
GIGABYTE AORUS GeForce GTX 1080 Ti DirectX 12 GV-N108TAORUS-11GD 11GB 352-Bit GDDR5X PCI Express 3.0 x16 ATX Video Card
Model #:GV-N108TAORUS-11GD
Item #:N82E16814125954
Return Policy:Replacement Only Return Policy
In Stock
Note (Add)
$719.99 $2,879.96
1
Update
CORSAIR AXi Series AX1500i Digital 1500W 80 PLUS TITANIUM Haswell Ready Full Modular ATX12V & EPS12V SLI and Crossfire Ready Power Supply with C-Link Monitoring and Control
CORSAIR AXi Series AX1500i Digital 1500W 80 PLUS TITANIUM Haswell Ready Full Modular ATX12V & EPS12V SLI and Crossfire ...
Model #:CP-9020057-NA
Item #:N82E16817139079
Return Policy:Standard Return Policy
In Stock
Note (Add)
$449.99 -$40.00 Instant $409.99
1
Update
AMD RYZEN 7 1800X 8-Core 3.6 GHz (4.0 GHz Turbo) Socket AM4 95W YD180XBCAEWOF Desktop Processor
AMD RYZEN 7 1800X 8-Core 3.6 GHz (4.0 GHz Turbo) Socket AM4 95W YD180XBCAEWOF Desktop Processor
Model #:YD180XBCAEWOF
Item #:N82E16819113430
Return Policy:Replacement Only Return Policy
In Stock
Note (Add)
$459.99 $459.99
2
Update
ADATA Ultimate SU800 512GB 3D NAND 2.5 Inch SATA-III Internal Solid State Drive (ASU800SS-512GT-C)
ADATA Ultimate SU800 512GB 3D NAND 2.5 Inch SATA-III Internal Solid State Drive (ASU800SS-512GT-C)
Model #:ASU800SS-512GT-C
$179.99 $359.98
>>
Thanks Anon!
>>
>>60491013
Enthusiast products and professional products are not the same. The Ryzen 7 1800x does not have enough PCIe lanes for that many GPUs.
I would consider looking at what tools you're going to be using, and what hardware is recommended.
Or just use Jewgle's Cloud TPUs ;-)
>>
>>60491013
Unironically wait for Threadripper.
>>
>>60491108
This. Also Vega.
>>
>>60491013
Depending on what framework you know, go with 1080ti or a Vega card.
Sli or CF aren't that great on most am4 motherboards
>>
>>60491013
don't buy ryzen and just buy 2x xeons or 4x xeons for cheap af, $50-100 each, so you can get more pcie lanes
>>
>>60491013
youre crazy, dont do it
>>
>>60491013
What are you building? A gf AI?
>>
>>60491013
You need cuda for deep learning so I'm not quite sure here what you want with amd cards. Open cl isn't as widely supported as cuda is.
>>
>>60495042
>what is RocM
>>
>>60496141
What is rocm?
>>
>>60496492
AMDs mind of alternative to CUDA
It's still early as fuck but seeing how they make their cards in this industry they might end up with something good. It mostly depends on how many people adopt our switch to it
>>
>>60497826
If there is a standard offered by amd I think most frameworks might adopt it. Although I don't really know why people don't like open cl. But I'm a noob on this field.
>>
>>60491013
Fuck off
>>
>>60498851
Same, simple put open cl is on the side of more broader things, while cuda for example is really optimized for certain tasks on certain hardware.
>>
>>60499223
If amd puts some effort in it then maybe we are lucky and nvidias grasp on the deep learning industry weakens a bit. But I'd wait for now if you really want amd cards. I thought about getting an amd card too but the lack or cuda annoys me. I want to do some simple stuff with deep learning too so I'll probably go for a weak nvidia GPU.
>>
>>60499448
You don't get CUDA on amd hardware, but yeah Vega instinct cards are interesting as fuck for DL
>>
>>60497826
are there any frameworks that already use rocm?
>>
>>60500213
Idk check out their page and github
Thread posts: 21
Thread images: 1
Thread images: 1