Can somebody help me understand what you do with a 50W TDP "professional" card? Either it sounds underpowered to me, or overpowered if you buy something like that with RT cores. Or does one use this to test if something even runs properly, no matter the performance (e.g. proof of concept)

My guess is they can be used for design/CAD/CAM workstations where you don't need a high framerate, but you do need your model to look 100% correct and have a stable reliable system to work on. Then they can do the proper final rendering of whatever is being worked on powerful off-location servers. But that's just a guess, I'm not an engineer...

@fantaskarsef it's also efficiency that's into play

I mean, thinking about it for half a day, I could imagine this being an entry card to CAD / CAM / architectural rendering, for instance. Only that proper plans on AutoCAD for instance already start being a pita with 8GB VRAM, and even worse if you want to push those textures through some proper lighting rendering. Hence why I asked, since it doesn't make sense in those use cases I could see it. Even with calculating stuff like sound wave travelling and proofing or protection, e.g. along railroad tracks and highways, I don't see this helping much, or calculating things like construction (static and structural engineering or construction physics when it comes to properly calculating the need for insulation or better matierials for weight distribution) needs more processing power than offered by this, and more often than not, takes days to complete even on a potent workstation PC. At least in projects and fields I know, and from people's first hand experience in my field of traini, this thing is underpowered... and certainly overpowered if it comes down to basically having a GPU in there to deal with lots of visual data to display / more than two displays to feed. Maybe other fields of expertise can make better use of this?

It's built on the 3000 series process by the look of it, so I'm guessing this is a relatively entry level piece of kit?

fantaskarsef:

Can somebody help me understand what you do with a 50W TDP "professional" card? Either it sounds underpowered to me, or overpowered if you buy something like that with RT cores. Or does one use this to test if something even runs properly, no matter the performance (e.g. proof of concept)

We use the small Quadro GPUs as display adapters next to the more powerful ones, so that we can put the more powerful ones into TCC mode, where they are dedicated to compute work. This prevents Windows from interrupting compute work with display updates etc, so these powerful cards can be fully used for compute. Up to now we used TXX00's (usually T1000's) for this, but since those are based on the same GPU type as the GTX 16XX series, which has been declared End of Life, it seems smart to introduce equivalent Ampere based cards for this purpose. Since it only drives the display, it doesn't need to be that powerful, so a tiny GP like these is more than sufficient for that task.

Crazy Joe:

We use the small Quadro GPUs as display adapters next to the more powerful ones, so that we can put the more powerful ones into TCC mode, where they are dedicated to compute work. This prevents Windows from interrupting compute work with display updates etc, so these powerful cards can be fully used for compute. Up to now we used TXX00's (usually T1000's) for this, but since those are based on the same GPU type as the GTX 16XX series, which has been declared End of Life, it seems smart to introduce equivalent Ampere based cards for this purpose. Since it only drives the display, it doesn't need to be that powerful, so a tiny GP like these is more than sufficient for that task.

Thank you for the insight! I wouldn't have expected people to buy entry level pro cards to run displays only. Or that windows updates couldn't be configured well enough to not need that workaround you described.

Video encoding as well, I'd think. AV1 encoding with NVENC is pretty great.

fantaskarsef:

Thank you for the insight! I wouldn't have expected people to buy entry level pro cards to run displays only. Or that windows updates couldn't be configured well enough to not need that workaround you described.

It's not really a workaround. The GPU that is used for display will get so called WDDM events inserted into its workflow every so often in order to update the display. While these events occur, the GPU can't execute any compute work, so if you want to get the most out of a GPU, not having it have to deal with those WDDM events is the best way to achieve this.

fantaskarsef:

I wouldn't have expected people to buy entry level pro cards to run displays only.

I see I forgot to address that part of your reply. The reason for using the pro cards is that they are available for longer periods of time, since NVIDIA has a Long Term Supply program for a range of their pro cards. This allows product designers to keep supply of a product that uses these GPUs going for longer before they need to change the product specs. In the business I'm in it is not unusual to have to maintain a product for 10 years or more, so having GPUs that NVIDIA guarantees will be available for a long period makes supporting those products simpler and avoids having to stockpile a huge amount of GPUs in our own warehouses for that period just in case they are needed.

Crazy Joe:

I see I forgot to address that part of your reply. The reason for using the pro cards is that they are available for longer periods of time, since NVIDIA has a Long Term Supply program for a range of their pro cards. This allows product designers to keep supply of a product that uses these GPUs going for longer before they need to change the product specs. In the business I'm in it is not unusual to have to maintain a product for 10 years or more, so having GPUs that NVIDIA guarantees will be available for a long period makes supporting those products simpler and avoids having to stockpile a huge amount of GPUs in our own warehouses for that period just in case they are needed.

Well, I probably see an issue where it is not, but the T1000 you mentioned above was released in 2021, and now you're replacing them even though that prolongued time of maintenance would be the reason to go for it... it sounds like it contradicts itself, but I am firm in my believe I just don't know the whole picture. Again, thanks for the insight. It still doesn't make too much sense for me, but I guess being in the position to have to take care of such systems, I'd see it differently, too.