Considering 2 x RTX 3090 graphics cards in setup

G

greent

Member
Considering adding a second RTX 3090 graphics card to the below config but was wondering if the Gigabyte Deisgnware motherboard has a better layout for 2 x RTX 3090 (I've read that it's best to have one slot space between two RTX 3090 graphics cards)

Has anyone else here paired two RTX 3090 GPU and on which motherboard? Thanks in advance.

Case
COOLERMASTER MASTERCASE H500M GAMING CASE
Processor (CPU)
AMD Ryzen Threadripper 3960X 24 Core CPU (3.8GHz - 4.5GHz, 142MB CACHE)
Motherboard
ASUS® ROG STRIX TRX40-E GAMING (DDR4, Wi-Fi 6, CrossFireX/SLI) - ARGB Ready!
Memory (RAM)
128GB Corsair VENGEANCE DDR4 3000MHz (4 x 32GB)
Graphics Card
24GB NVIDIA GEFORCE RTX 3090 - HDMI, DP
1st M.2 SSD Drive
2TB SAMSUNG 970 EVO PLUS M.2, PCIe NVMe (up to 3500MB/R, 3300MB/W)
Power Supply
CORSAIR 1600W PRO SERIES™ TITANIUM AX1600i DIGITAL MODULAR PSU
Power Cable
1 x 1 Metre IEC C19 UK Power Cable
Processor Cooling
Corsair H150i RGB PRO XT Hydro Series High Performance CPU Cooler
Thermal Paste
ARCTIC MX-4 EXTREME THERMAL CONDUCTIVITY COMPOUND
Sound Card
ONBOARD 6 CHANNEL (5.1) HIGH DEF AUDIO (AS STANDARD)
Network Card
10/100/1000 GIGABIT LAN PORT (Wi-Fi NOT INCLUDED)
USB/Thunderbolt Options
MIN. 2 x USB 3.0 & 2 x USB 2.0 PORTS @ BACK PANEL + MIN. 2 FRONT PORTS
Operating System
Windows 10 Professional 64 Bit - inc. Single Licence [MUP-00003]
Operating System Language
United Kingdom - English Language
Windows Recovery Media
Windows 10 Multi-Language Recovery Image - Supplied on USB Drive
Office Software
FREE 30 Day Trial of Microsoft 365® (Operating System Required)
Anti-Virus
NO ANTI-VIRUS SOFTWARE
Browser
Microsoft® Edge (Windows 10 Only)
Warranty
3 Year Silver Warranty (1 Year Collect & Return, 1 Year Parts, 3 Year Labour)
Delivery
STANDARD INSURED DELIVERY TO UK MAINLAND (MON-FRI)
Build Time
Standard Build - Approximately 17 to 19 working days
 
G

greent

Member
None of the above!
Deep learning on GPU, my friend (2 x RTX 3090 is 20k Cuda cores). Main concern is whether the two graphics cards will stay cool and work well with this motherboard (or even fit on it!).
 
G

greent

Member
Not Bitcoin, but 48GB (2x24GB) of graphics card memory means you can compete in machine learning competitions (e.g. Kaggle) and do lots more experimentation with data on GPU memory rather than having to move data between the CPU and GPU, which slows things down a lot. Allied with the right software tools (Dask / RAPIDS) you can perform some computational tasks 10x or more faster on the GPU cores instead of CPU. Most machine learning algorithms are doing lots of matrix multiplication, which GPUs are ideal for. And with the popular data science programming languages such as R and Python you're might be a bit restricted if your data doesn't fit in regular memory, so getting lots of GPU memory opens up other possibilities.

But I don't want to sell the idea too hard because unsure if adding a 2nd graphics card would move me to the back of the queue again so I'm trying to verify this first!

Happy to share photos and confirm how it's all working when the rig is eventually built and running.
 
jamiephillips909

jamiephillips909

Godlike
greent said:
Not Bitcoin, but 48GB (2x24GB) of graphics card memory means you can compete in machine learning competitions (e.g. Kaggle) and do lots more experimentation with data on GPU memory rather than having to move data between the CPU and GPU, which slows things down a lot. Allied with the right software tools (Dask / RAPIDS) you can perform some computational tasks 10x or more faster on the GPU cores instead of CPU. Most machine learning algorithms are doing lots of matrix multiplication, which GPUs are ideal for. And with the popular data science programming languages such as R and Python you're might be a bit restricted if your data doesn't fit in regular memory, so getting lots of GPU memory opens up other possibilities.

But I don't want to sell the idea too hard because unsure if adding a 2nd graphics card would move me to the back of the queue again so I'm trying to verify this first!

Happy to share photos and confirm how it's all working when the rig is eventually built and running.
Click to expand...
Would love to see that I would imagine your queue position would remain the same based of the original order but that would be speculation
 
Durakus

Durakus

Bronze Level Poster
I don't have the enthusiast / tech level to comment on how effective that is. In theory it sounds like it would work, but in practice SLI on graphics cards has been seen to be a pretty big waste. Maybe for that price point a processing oriented GPU might be more effective?

There are a few tech-tubers that have covered Dual 3090's.
Jayztwocents is the only one coming to mind right now, He's done a few SLI builds with 3090, but I can't remember if he benched/tested rendering/processing.

 
SpyderTracks

SpyderTracks

We love you Ukraine
Durakus said:
I don't have the enthusiast / tech level to comment on how effective that is. In theory it sounds like it would work, but in practice SLI on graphics cards has been seen to be a pretty big waste. Maybe for that price point a processing oriented GPU might be more effective?

There are a few tech-tubers that have covered Dual 3090's.
Jayztwocents is the only one coming to mind right now, He's done a few SLI builds with 3090, but I can't remember if he benched/tested rendering/processing.

Click to expand...
SLI is only a waste for gaming. For rendering or machine learning it’s extremely beneficial.

As to the OP’s question, I would recommend upgrading the motherboard as you suggested. It’s more of a pro use board whereas the ASUs strix is more designed for the lesser workload of gaming.

Have you already ordered the build? If not, drop the thermal paste as the cooler comes pre applied with high quality paste.
 
Durakus

Durakus

Bronze Level Poster
SpyderTracks said:
SLI is only a waste for gaming. For rendering or machine learning it’s extremely beneficial.

As to the OP’s question, I would recommend upgrading the motherboard as you suggested. It’s more of a pro use board whereas the ASUs strix is more designed for the lesser workload of gaming.

Have you already ordered the build? If not, drop the thermal paste as the cooler comes pre applied with high quality paste.
Click to expand...
Aye, I am just concerned that due to SLI features being essentially dropped and unoptimized, the "Works in theory" statement is that It SHOULD work but it may not work as well as intended for the price point of the 3090 if they don't communicate or load data correctly in the SLI configuration.
 
SpyderTracks

SpyderTracks

We love you Ukraine
Durakus said:
Aye, I am just concerned that due to SLI features being essentially dropped and unoptimized, the "Works in theory" statement is that It SHOULD work but it may not work as well as intended for the price point of the 3090 if they don't communicate or load data correctly in the SLI configuration.
Click to expand...
Don't get bogged down by SLI, SLI is strictly a gaming term.

The 3090's use NVLink which is the modern form of SLI and it works extremely well. Remember, the 3090 is not a gaming card which is why it still supports NVLink. Any workstation card such as the 3090 benefits greatly from NVLink for any compatible usage like the OP's.
 
jamiephillips909

jamiephillips909

Godlike
SpyderTracks said:
Don't get bogged down by SLI, SLI is strictly a gaming term.

The 3090's use NVLink which is the modern form of SLI and it works extremely well. Remember, the 3090 is not a gaming card which is why it still supports NVLink. Any workstation card such as the 3090 benefits greatly from NVLink for any compatible usage like the OP's.
Click to expand...
They should of given it the titan name for what it is its pritty much a titan class card ive never seen it as a appropriate gaming card
 
FerrariVie

FerrariVie

Super Star
If you're looking for the absolute best in machine learning, you should give the Quadro RTX 8000 or A6000 a try, 48Gb of ECC VRAM in just one card and also supports NVlink, so you can add up to 2 cards and get 96Gb or ECC VRAM, if your budget is unlimited :p
 
G

greent

Member
SpyderTracks said:
As to the OP’s question, I would recommend upgrading the motherboard as you suggested. It’s more of a pro use board whereas the ASUs strix is more designed for the lesser workload of gaming.

Have you already ordered the build? If not, drop the thermal paste as the cooler comes pre applied with high quality paste.
Click to expand...
Thanks for both of these recommendations, and to all other posters for chipping in, which is appreciated.

Unfortunately, budget is limited, but this is a big spend for me. I'm leaning towards the Gigabyte Designware MB unless someone else suggests otherwise. I'm also still waiting for support to reply as they need to manually add the 2nd RTX 3090 (it's not one of the cards available in the dropdown) and get back to me with a price and answer re the motherboard.

SpyderTracks - are you doing something similar, and if so, on Linux, Windows or WSL2? I've been "practicing" GPU compute on my current notebook using Ubuntu 20.04 via WSL2 on Windows but I'm wondering if I should go straight to Ubuntu and ditch Windows / WSL2.
 
G

greent

Member
FeVieira said:
If you're looking for the absolute best in machine learning, you should give the Quadro RTX 8000 or A6000 a try, 48Gb of ECC VRAM in just one card and also supports NVlink, so you can add up to 2 cards and get 96Gb or ECC VRAM, if your budget is unlimited :p
Click to expand...
Thanks for this info. Budget doesn't quite stretch that far. Is that something you've tried?
 
SpyderTracks

SpyderTracks

We love you Ukraine
greent said:
Thanks for both of these recommendations, and to all other posters for chipping in, which is appreciated.

Unfortunately, budget is limited, but this is a big spend for me. I'm leaning towards the Gigabyte Designware MB unless someone else suggests otherwise. I'm also still waiting for support to reply as they need to manually add the 2nd RTX 3090 (it's not one of the cards available in the dropdown) and get back to me with a price and answer re the motherboard.

SpyderTracks - are you doing something similar, and if so, on Linux, Windows or WSL2? I've been "practicing" GPU compute on my current notebook using Ubuntu 20.04 via WSL2 on Windows but I'm wondering if I should go straight to Ubuntu and ditch Windows / WSL2.
Click to expand...
I’m not doing likewise, just have looked into it for a friend who I designed a ML build for.
 
G

greent

Member
barlew said:
Haha the budget is limited! To what the GDP or a small nation!?
Click to expand...
The wife wasn't too happy when I dropped a few k short selling the stock markets in my spread betting account. I think she's hoping that this will keep me busy so that it will end up costing less.

I'm sure she'll be delighted when I start dabbling in bitcoin on account of my new found expertise in graphics cards..
 
You must log in or register to reply here.
Top