original post edited to reflect some updates (check the bottom of the post for an explanation)
Hi All,
I'm planning a new PC with an AMD Ryzen Threadripper 3990X and 4x Nvidia3080 3090 and I'm undecided about getting a single 2000 W PSU or 2 smaller PSUs in a dual-PSU setup (using the Add2Psu adapter).
The OuterVision Power Supply Calculator recommends 1908 W for my build (with x4 3080): https://outervision.com/b/9Of8MZ
The OuterVision Power Supply Calculator recommends 2036 W for my build (with x4 3090): https://outervision.com/b/YiYgQ6
This instead is the list of the components (it should be 4x 3080 but PCPartPicker doesn't allow me to add more than 1 GPU): https://pcpartpicker.com/list/98kRfP
This instead is the list of the components (it should be 4x 3090 but PCPartPicker doesn't allow me to add more than 1 GPU): https://pcpartpicker.com/list/rgYqYg
The PC case is yet to be decided depending on the PSU choice too, in case I opt for the dual-PSU.
For the cooling I'm going to use IceGiant ProSiphon Elite for the CPU, meanwhile for the GPUs I will setup a custom water-cooling system once I can get the 3rd/4th GPU (right now I cannot get even a single Founder Edition).
This machine is going to be used for my work-from-home, to train deep learning models for work (hence the 4x GPUs), machine learning models for data science competitions like Kaggle (hence the 64 cores for training in parallel with Scikit-Learn and similar Python libraries), probably some crypto mining (which could eat most of the computing power while I'm not working) and no gaming whatsoever.
Budget is not a huge concern as I use it for work but I have another blue print for a cheaper machine with an AMD 3970X in case I decide to spend less.
My goal is to get a great workstation for a non-business price (less than pre-built deep learning machines) and keep it for 5+ years.
At the following link they say that a 1600W/1800W PSU is ok for 4x3080 and a 2000W one is ok for 4x3090: https://timdettmers.com/2020/09/07/which-gpu-for-deep-learning/
For my case Itrust more think the OuterVision calcs are more reliable.
I've searched the websites of many PSU producers (https://www.ranker.com/list/the-best-power-supply-manufacturers/computer-hardware up to Zalman) and these are all the 2000W+ PSUs I've found:
I've also read the following thread: https://linustechtips.com/topic/300961-high-wattage-psus-useless-2000w-80-platinum-psu/
My main goals are the following:
Edits:
Hi All,
I'm planning a new PC with an AMD Ryzen Threadripper 3990X and 4x Nvidia
The OuterVision Power Supply Calculator recommends 2036 W for my build (with x4 3090): https://outervision.com/b/YiYgQ6
This instead is the list of the components (it should be 4x 3090 but PCPartPicker doesn't allow me to add more than 1 GPU): https://pcpartpicker.com/list/rgYqYg
The PC case is yet to be decided depending on the PSU choice too, in case I opt for the dual-PSU.
For the cooling I'm going to use IceGiant ProSiphon Elite for the CPU, meanwhile for the GPUs I will setup a custom water-cooling system once I can get the 3rd/4th GPU (right now I cannot get even a single Founder Edition).
This machine is going to be used for my work-from-home, to train deep learning models for work (hence the 4x GPUs), machine learning models for data science competitions like Kaggle (hence the 64 cores for training in parallel with Scikit-Learn and similar Python libraries), probably some crypto mining (which could eat most of the computing power while I'm not working) and no gaming whatsoever.
Budget is not a huge concern as I use it for work but I have another blue print for a cheaper machine with an AMD 3970X in case I decide to spend less.
My goal is to get a great workstation for a non-business price (less than pre-built deep learning machines) and keep it for 5+ years.
At the following link they say that a 1600W/1800W PSU is ok for 4x3080 and a 2000W one is ok for 4x3090: https://timdettmers.com/2020/09/07/which-gpu-for-deep-learning/
For my case I
I've searched the websites of many PSU producers (https://www.ranker.com/list/the-best-power-supply-manufacturers/computer-hardware up to Zalman) and these are all the 2000W+ PSUs I've found:
- EVGA SuperNOVA 2000 G1+: https://asia.evga.com/products/product.aspx?pn=220-GP-2000-X6
- FSP CANNON 2000W: https://www.fsplifestyle.com/en/product/cannon2000w.html
- FSP CANNON PRO 2000W: https://www.fsplifestyle.com/en/product/cannonpro2000w.html
- SilverStone HELA 2050 Platinum (not yet available): https://www.silverstonetek.com/product.php?pid=1023
- Super Flower LEADEX PLATINUM 2000W: https://www.super-flower.com.tw/product-data.php?productID=67&lang=en
- GameMax GM-2000: https://www.gamemaxpc.com/productkkk/549-en.html
I've also read the following thread: https://linustechtips.com/topic/300961-high-wattage-psus-useless-2000w-80-platinum-psu/
My main goals are the following:
- get a reliable and safe power supply to run my machine heavily (the plan is to use it 24/7 across deep learning, machine learning, crypto mining)
- "reliable" to avoid any downtime given that I'll use the machine for work
- "safe" to avoid any potential damage to the expensive hardware
- keep the noise at a minimum given that I'll keep it in my bedroom and run it while sleeping
- I'm not really planning of overclocking the CPU, GPUs, etc., but I could test it in case I see fit
- Shall I buy a 2000W PSU or 2 PSUs for a dual-PSU setup?
- For the 2000W PSU, which one you suggest from the above list?
- For the dual-PSU setup what wattage and PSUs you suggest?
- I'm checking the list here as well: https://linustechtips.com/topic/1116640-psu-tier-list/
Edits:
- 10-Dec-2021
- today I've been able to make an online order for an NVIDIA FE 3090 on a GPU drop as it popped up on the NVIDIA website (sold via an European partner online shop). I was hoping to get a 3080 Ti and/or a 3080 as well and re-sell the 3090 later on but I haven't been able so I'll probably stick to the 3090 and modify my build. I've been quite lucky to get an FE as the website was mostly down during the process.
- I've found this interesting blog post from Lambda Labs (a maker of deep learning workstations) which gives some good information on multi-GPU build with NVIDIA RTX 30 series cards: https://lambdalabs.com/blog/deep-learning-hardware-deep-dive-rtx-30xx/
- The above 2 points are skewing my decision towards a dual-PSU setup.
Last edited:
Facebook
Twitter