[SOLVED] Need Help/Suggestion on PSU for 4090FE Build.

DX DO

Distinguished
Jul 10, 2012
16
0
18,510
Hi,

I'm Gaurav , doing most of 3D Architecture GPU Rendering works on my workstation. I would like to ask you few question if you don't mind, which will help me greatly to plan accordingly for my extended rig which will be used along with my workstation.

Since many of them, Arguing about to wait/use Atx3.0, NVIDIA shared few words on the topic. Which help me to plan with existing lineup of PSU.
I'm looking for two key factors in PSU that is Quality/Performance with Values. I'll really appreciate your input/suggestion on the topic.

Note: Kindly note that this is going to use as rendering rig and Should be comfortable for few days straight 100% GPU uses.



My current Workstation Configs are following:

PSU: Rm1000x
CPU: AMD 3590x
MOBO: Ws x570 ace
GPU: 3090FE + 2070S + 2070
RAM: GKILL Ripjaw 128GB
HDD: ADATA SX8200PNP 512GB + WD Blue SN570 1TB
Monitor: BenQ PD3200U + Dell P2414H + Acer EK220Q






Q1- I'm looking for Two 4090FE cards which will goes with 2x CM G800 (Mid quality) or Corsair Rm650, XPG Core Reactor 650W (High quality) connected externally with the help of 1x raiser to the Workstation Motherboard since 1x speed does the job. They will be at stock speed without any tweaking/OC.

Few Other PSU which is available in local store - Deepcool DQ750-M V2L, Msi MPG A750GF Or would ask for any recommendation.

Q2- Do i really need double headroom (power) as 4090 card is more balanced than 3090Ti in Power spikes?

Q3- Since I'm doing this first time, To connect extra PSU to the Main PSU Rm1000x, I will use Add2PSU Card, This is correct way/Hardware?


Please share your comments on the topic, Thanks.

Cheers!
Gaurav N.
 
Solution
I have never noticed above 365W on my 3090Fe (No OC) Card while rendering. These readings are from GPU-Z -Board Power draw.

After all Dear Eximo, with your experience, Do you think EVGA SuperNova 1300 GT is suitable for workload rendering even for weeks or AX1600i is the only recommendation.

GPU-Z is not likely to catch the transient spikes. Not sure what the polling rate is, but the transients are up to 200ms only, it is the over current protection on the PSUs you have to worry about, and that will scale with the max power output of the PSU. If you target the worst case of 500-550W per GPU, then you need a 1000+ watt PSU.

My first suggestion of an EVGA 1600 P2 is what I would get based on the cost...
Those PSUs you listed are not enough, at all. Even the RM1000x would only really be suitable for one of these cards. I would think a single larger power supply would do fine in this scenario.

You are looking at 450W each, so something like a 1600W would be a pretty big buffer. You will have to double check you have the appropriate circuit to run such a PSU, in the US that means a 20 amp breaker and circuit (typical 15 is amp)

https://pcpartpicker.com/product/Lb...fully-modular-atx-power-supply-220-p2-1600-x1

Obviously you are not worried about cost, so a $300 PSU is a fine investment, even if you have to pay to import one.
 
Thank you Eximo for your comment.

In such topic, I've not enough experience to understand things straight. Can you please elaborate a bit in details?

As you have mention, 450W+450W would need 1600W PSU when these two card will be only connected to GPU. No CPU load, MOBO or other connection will connect to PSU?

Those PSUs you listed are not enough, at all. Even the RM1000x would only really be suitable for one of these cards. I would think a single larger power supply would do fine in this scenario.

Okay, But Please let me know why they are (650W PSU) insufficient? Any details would be helpful.
 
Hi Sizzling,

Yes, I have seen Jayz comments on the topic. Since those burns were from in between development as Nvidia stated in comment.


"PCI-SIG warned of potential overcurrent/overpower risk with 12VHPWR connectors using non-ATX 3.0 PSU & Gen 5 adapter plugs. Why did PCI-SIG issue the warning and how does this affect NVIDIA’s new products?


It does not affect NVIDIA’s new products. During early development and testing we identified a potential issue with a prototype connector manufactured from a sole supplier. As a PCI-SIG member, we informed the supplier, asked them to fix it (which they did), and shared our findings to help suppliers who are implementing the new standard. The PCI-SIG was passing along those learnings to other suppliers."



Also I have strong belief on Corsair and I trust Jonnyguru comments as he stated in the following thread, Kindly have a look. I really don't care what Jayz want to say on this topic. We have seen 3090Ti (450W) is running smoothly with current lineup PSU, Then why they (4090FE, 450W Non OC) make such an issue when they are more stable than 3090Ti on power consumption/requirement!!


Since CORSAIR Announces Compatibility for NVIDIA 40-Series Graphics Cards
https://www.techpowerup.com/forums/...40-series-graphics-cards.299155/#post-4842306
 
Thank you Eximo for your comment.

In such topic, I've not enough experience to understand things straight. Can you please elaborate a bit in details?

As you have mention, 450W+450W would need 1600W PSU when these two card will be only connected to GPU. No CPU load, MOBO or other connection will connect to PSU?

Okay, But Please let me know why they are (650W PSU) insufficient? Any details would be helpful.

A 650W PSU to power a 450W GPU with known transients of up to 200%? The capacitors are simply too small to take the hit and the OCP on something that small is likely to trip.

You are still stuck on the idea of using multiple PSUs, don't. Get a single decent PSU to power a pair of high end cards. It is safer and simpler. If you want to go down that route, then I still suggest something like 750W GPU per RTX4090. Or maybe something like a 1200W PSU (and that might not be enough)
 
A 650W PSU to power a 450W GPU with known transients of up to 200%? The capacitors are simply too small to take the hit and the OCP on something that small is likely to trip.

Yes, Thank you. I know, PSU required some headroom that why i asked about it in my 2nd question. I understood completely.


I'm not a fan of multiple/small setup in the line but since this does the job, I'm comfortable with the solution. I really want to go for 1000/1200/1300 PSU if they solve the purpose. Above that they (local vendors) are kind of over priced.

I have a question for you (I mean for Head room), Since I have almost 800W+ total consumption on current (Rm1000x PSU) while rendering which is working great since last few years, If Head room is really a concern why there are no trips?

Some insight on more detailed answer would be appreciated. Thanks.
 
Hi Sizzling,

Yes, I have seen Jayz comments on the topic. Since those burns were from in between development as Nvidia stated in comment.


"PCI-SIG warned of potential overcurrent/overpower risk with 12VHPWR connectors using non-ATX 3.0 PSU & Gen 5 adapter plugs. Why did PCI-SIG issue the warning and how does this affect NVIDIA’s new products?


It does not affect NVIDIA’s new products. During early development and testing we identified a potential issue with a prototype connector manufactured from a sole supplier. As a PCI-SIG member, we informed the supplier, asked them to fix it (which they did), and shared our findings to help suppliers who are implementing the new standard. The PCI-SIG was passing along those learnings to other suppliers."



Also I have strong belief on Corsair and I trust Jonnyguru comments as he stated in the following thread, Kindly have a look. I really don't care what Jayz want to say on this topic. We have seen 3090Ti (450W) is running smoothly with current lineup PSU, Then why they (4090FE, 450W Non OC) make such an issue when they are more stable than 3090Ti on power consumption/requirement!!


Since CORSAIR Announces Compatibility for NVIDIA 40-Series Graphics Cards
https://www.techpowerup.com/forums/...40-series-graphics-cards.299155/#post-4842306
Ultimately you want a psu with 12VHPWR for a 4090 which is recommended in that video
 
The 20 series cards do not have the same extreme transient loads as the 30 series, and the 40 series is just an even larger GPU.

The RM1000x is a fine PSU. I run the RM850x myself. The chances of all three cards giving you a transient at the same time is low. They are all slightly different. A pair of 4090 doing the same job on the hand, are likely to have similar power spikes, and given the size of the GPUs and their expected wattage it is simply prudent to have the extra headroom.

900W + 50% = 1350W + the rest of the system. 3090 had spikes as high as 550W possible, the 3090Ti would be a little higher, and the 4090 is likely similar to the 3090Ti in power draw.

If you like Corsair, there is always the AX1600i.
 
Yes, Agree as per your comment.

I have never noticed above 365W on my 3090Fe (No OC) Card while rendering. These readings are from GPU-Z -Board Power draw.

900W + 50% = 1350W + the rest of the system

There won't be any +the rest of the system load, As this will be connected to motherboard via separate 1x cable to the motherboard.

In Summery:This Rig is only One/Two PSU + 2x RTX4090 + Attached riser with 2x Pci lane cable in a separate rig.



Also Nvidia spec sheet recommend 850W with a comment for 3090Ti .

"4 - Requirement is made based on PC configured with an Intel Core i9-10900K processor. A lower power rating may work depending on system configuration.
Kindly check this here. Specs-> View Full specs and scroll to the bottom."

This is all my base calculation to go for 650W if possible or 750W as you recommend to have extra headroom for spikes which is a great advise.

After all Dear Eximo, with your experience, Do you think EVGA SuperNova 1300 GT is suitable for workload rendering even for weeks or AX1600i is the only recommendation. Or should wait more to understand the market and reviews. I'm only afraid from Scalpers who's burn the reasonable market.

I really appreciate your inputs on the topic. Thank you.
 
I have never noticed above 365W on my 3090Fe (No OC) Card while rendering. These readings are from GPU-Z -Board Power draw.

After all Dear Eximo, with your experience, Do you think EVGA SuperNova 1300 GT is suitable for workload rendering even for weeks or AX1600i is the only recommendation.

GPU-Z is not likely to catch the transient spikes. Not sure what the polling rate is, but the transients are up to 200ms only, it is the over current protection on the PSUs you have to worry about, and that will scale with the max power output of the PSU. If you target the worst case of 500-550W per GPU, then you need a 1000+ watt PSU.

My first suggestion of an EVGA 1600 P2 is what I would get based on the cost. Corsair AX1600i is very good, but it is usually very expensive. $600 right now.

Rather than the EVGA GT I would stick with the known quantity of the EVGA G2 1300W. EVGA P2 1200W, Corsair HX1200, HX1500i, EVGA T2 1600W. Plenty of choices out there from trustworthy model lines.
 
Solution
From what I have found there are new units about to hit the market from several brands. In a Linus video I watched earlier they used a new Gigabyte unit that had 12VHPPWR

True, but those were press kits. (Actually Gigabyte sent that kit around to a few Youtubers I noticed) I'm sure they will be on the market soon, but might take a while to make their way to India.