GPU Installation on R730

ThomasTheTech,

A single CPU configuration wouldn't work, as the R730 requires 2 CPU's for GPU use.

I believe the part number for the cable is N08NH.

Regarding the power supplies, what wattage are they? I ask as they would have to be 1100 watt for GOU use as well.

Let me know if this helps.

Hi 

how many P100 did you manage to power up in your R730 ?  do you have any blog describing this ?

Good evening, I want to know if r730 is full compatible with tesla p100, or I need some connector or fan kit.

Esx 6.7 and gpu share.

thanks

@DELL-Chris H​ ,

You mentioned that you can use 4 150W single width GPUs. How can this be if you can only connect a single GPU cable per riser? Say I have 4 RTX 4000 Ada Generation GPUs (single-width and under 150W), where will I connect the extra 2 power cables?

Hello, would this help?

https://dell.to/4acSbz5

Respectfully,

Hello @DELL-Young E ,

Thank you for the prompt response. Let me explain the problem a little more clearly as I don't think I properly communicated the problem I am facing. The Dell PowerEdge R730 Datasheet (linked here: https://i.dell.com/sites/doccontent/shared-content/data-sheets/en/Documents/Dell-PowerEdge-R730-and-R730xd-Technical-Guide-v1-7.pdf) mentions that you can use 2 x 300W (double-wide) or 4 x 150W (single-wide) graphics cards. I am trying to accomplish this with 4 - RTX 4000 Ada Generation GPUs that are single-wide and 130W. According to the datasheet, using 4 GPUs should be possible if you have riser 2 and riser 3. Riser 2 and riser 3 gets you two 8-pin connectors total to use.

In the link you sent, the figure states that you can use the power connector (labeled 4 on the figure and d in the legend) for GPUs. The RTX 4000 Ada Generation uses a single 8-pin connector for power. Note that the Tesla K80 and others recommended also use a single 8-pin connector. The recommended power cable by @DELL-Chris H above is the N08NH. That cable is an 8-pin to (6-pin + 8-pin).

If this server is intended to work with 4 GPUs total, how am I supposed to connect the power cables to the 4 GPUs?

My concerns are:

A) Is connecting a N08NH cable to the 8-pin on riser 2/riser 3, and then connecting each end of the split end of the N08NH cable to two different GPUs safe to do?

B1) Since these GPUs use an 8-pin connection, is there a different cable than the N08NH cable to connect to two 8-pin GPUs?

B2) If there isn't such a cable by Dell, is connecting the 6-pin from the N08NH safe to connect to a 6-pin to 8-pin adapter and which would you recommend?

Thank you

Hello,

A) to answer your question, we can not guarantee it's going to be safe.

Respectfully,

@DELL-Young E​,

Well your documentation states that it is supported, so please let me know how it is physically possible to connect 4 single-width GPUs under 150W each to an R730. Where is the power supposed to be connected to and how?

Thanks

Hello,

For four 150W single width you would need riser 2 and 3 and two of the split power cables np#N08NH.

This image is of a validated AMD, FirePro S7150, single wide, (150W) and uses 6pin power:

Dell has not validated the RTX 4000 Ada so it may not work.

Maybe someone in the community has tried it and can share their experience adding four of those GPU.

@DELL-Charles R​,

The 6-pin requirement is not stated anywhere in Dell's documentation. That is very disappointing and misleading.

Since we’re talking about the AMD FirePro S7150, let's say I wanted use 4 of them in my servers. How am I supposed to connect the power to each of the GPU? Is the power supposed to be shared for two GPUs with one of those N08NH cables? Does this cause any issues with power draw on the GPUs or the PSU or the riser?

Thanks

Hello could we go through these guidelines one more time? I am not sure about the server supporting 4 GPUs.
https://dell.to/3TShjnv
Respectfully,

@Ramdisk01221​ Basically you need to make your own power cable. The chances of Dell selling you something that will just "work" are next to zero. What I did was start with the N08NH cable remove the plugs from the GPU end and put my own plug on. Even as they come the pin outs are not suitable for a P100/V100/A100 card. A suitable finest Chinesium crimp tool can be had for around 20 USD that will suffice for small volume work.

The original plan had been to run four R730's with A100 cards in and back in autumn 2020 I even had a loner A100 that I tested out. However when we placed the order for the A100's we had a call from the vendor saying they had a special on and could offer the A100's in servers for no extra cost!!! I am not stupid enough to look a gift horse in the mouth. We ended up putting out old V100's that had been in a C410X attached to C6220's in them instead.

Basically if you are not confident in making up your own power cables then forget about it. If you are then it will work just fine. Nvidia provide PDF's with pin out information on the power plugs and the wiring on the N08NH cable is as expected, black is ground and yellow is 12V.

Apart from that you need the 1100W PSU's and replace the low profile PCI blanks with solid ones for airflow reasons. I have had V100's in four R730's for two years now with no problems.