Jump to content
HWBOT Community Forums

Discussion of GPU Categories for Quadro/Tesla/Grid GPUs in Database


Recommended Posts

It came up over on Discord and the more I was looking at some of the categories, the more I was curious what direction should be taken with regards to how the Nvidia professional/workstation/server GPUs should be categorized in the HWBOT database. As it is now, there are several categories/families that are sort of dedicated to these cards, including the NVS series, Quadro NVS series, Quadro RTX series, Quadro series, and Tesla series.

Currently, there is a little bit of a mix of how the Nvidia "professional" cards appear to be categorized. Sometimes they are in their name-appropriate category (e.g. a Quadro RTX 5000 being in the Quadro RTX series), but other times, they are in their Geforce counterpart/equivalent category (e.g. the Quadro RTX 4000 is in the Geforce 2000 series), and other times, I think the GPU just has incorrect info about the specs/core and is thus placed in the wrong place entirely (e.g. the RTX A6000 being listed as having a TU102 core and put in the Geforce 2000 series instead of a GA102 core in either the Geforce 3000 or Quadro RTX families).

I guess my question is which direction we'd want to take in regards to what the "right" place to categorize these GPUs. Do they get separated into their own Quadro series families (potentially similar to how TechPowerUp categorizes them with regards to their generations), or do they get moved back over with their Geforce "equivalent" (i.e. by core and/or architecture) families? I'd lean a little towards having more separated Quadro groupings, potentially even by architecture (which then gets into kind of not-as-nice naming separations, like how a potential "Quadro Kepler" family has some cards that are actually Maxwell cores because of cards like the K1200 and K2200, or by naming "conventions" (i.e. the letter before the numbering) the Quadro 410 and 4100 "should" be Fermi, but they're actually Kepler, but don't have a K with the number), but then that kind of just makes a potential ton of new separate GPU families, which then could just get unwieldy overall compared to just having one main "Quadro" family. But that said, it could at least shrink the overall "Quadro" family down if we separated off the Quadro FX family GPUs from the Quadro family GPUs (I think there are ~80 Quadro FX's out of the currently ~160 Quadros in that family the current database), and then I guess it would also be a matter if we wanted to separate off some small categories like the Quadro2 and Quadro4 series and potentially any other stragglers. And, also, I don't know if there are any other thoughts on the Geforce 1000 series grouping together both the Pascal (1000 series) and Turing (1600 series) GPUs and whether people would want to separate that into a 1000 series and a 1600 series, given how Country Cup was basically 16 series GPUs for that one GPUPI stage.

But, that said, things are then treated differently on the AMD side. The AMD GPUs are separated just by their generation (GCN 1st/2nd/3rd/etc), where the Pro/workstation cards are just grouped in with the "consumer" parts. The AMD GPUs categories don't distinguish FirePro vs FireStream vs Radeon vs Radeon Instinct vs Radeon Pro vs Radeon Sky in the same way that the Nvidia ones are separated in the Geforce/NVS/Quadro/Tesla. Granted, it's just a difference in how they're (currently) categorized overall, where Nvidia's side is separated by their series and branding, but AMD's are just purely by architecture/generation (partly because of the difference in naming consistency across the generations (moreso on AMD's side)), and converting them both to having the same type of categorization could just be kind of time consuming for little gain. I do still think that, at least on the Nvidia side, there does need to be a little cleanup and shuffling of things, just given a couple things that I think are errors or are just in the wrong place or separating the Quadro FX off into its own separate category from the "normal" Quadro series (and maybe even potentially merging the "NVS series" in with the "Quadro NVS series"). I'm not sure if there are/were any specific reasons they are separated the way they are (NVS vs Quadro NVS, Quadro FX as part of Quadro, etc), or if it's just how it was and things were added over time and the category just gre large.

I'd imagine we don't want to separate out the AMD categories into essentially consumer and workstation/server/compute categories/families given the likely time consuming nature of going through them all, but at least with the Nvidia ones, they're kind of already separated out a bit and would require a lot less shuffling. Somewhat related to the families of GPUs is also how they're separated out for competitions, but I think there's generally the "no workstation/server cards allowed" for most comps (at least ones for the GPU categories, since I don't think most people would care if I'm using a Quadro as my video output when running like superpi, especially if I don't even *have* to list it in the system hardware in the submission), but I feel like a lower end Radeon Pro, like say a WX3200, would go under the radar for a general comp, but people might be quicker to point out a Quadro T600 (even as "bad" as it would likely be for a given comp) as not being allowed because it's a workstation card compared to the Radeon Pro. For comps allowing by architecture, if workstation cards aren't allowed, it may be a little more ideal to have the Quadros off in their own categories (which still potentially "allows" AMD workstation cards to be subbed with, given they're grouped just by architecture, without manually setting competition limitations (based on my limited understanding of the comp engine)), but it would still allow for some of them to be used for those specific comps like Cheapaz where often times it's just picked by the core, which would then allow for random Quadros.

But anyway, yeah, I figured I'd just see what other people's thoughts were, but also list a couple GPUs that I noticed/think are out of place (below).

- Quadro FX 350M is listed in the Quadro NVS series, but should be in the Quadro series (or potential new Quadro FX series)
- Quadro RTX 5000/6000/8000 are listed as RT102/104GL cores, but I think should be TU104/102 cores
- Quadro RTX A6000 is listed as having a TU102 core, but I think it should be a GA102 core
- Grid K1 is under Quadro series, but Grid K340 is under Geforce 600 series, but perhaps they should both just be under a new Grid family
- NVS 5100M is under Quadro series, but should be under NVS Series
- NVS 810 is under the Tesla series, but should be under NVS Series
- Not sure how the Pascal/Turing/Ampere Quadros (both Quadro and non-Quadro branded) should be grouped (i.e. stay with Geforce 1000/2000/3000 or move over to Quadro/Quadro RTX)
- Tesla T4 is under Geforce 2000 series, but perhaps should be moved to Tesla Series
- Quadro FX Go1000 is under Geforce 5 series, but maybe should be moved to Quadro series (or a new Quadro FX series)
- Geforce GTX 780 (2880 Shaders) is under Geforce 600 series, should this be under Geforce 700 series?
- Geforce 830A is under Geforce 900 series, should this be under Geforce 800 series?

Link to comment
Share on other sites

  • Crew

Thx for this massive write-up. What you see here is a big mess because of some limitations the bot has since the beginning. If a GPU model is used in multiple product families, we have to add a second GPU Core like GA102' or GA102GL. I have only with a new core the possibility to link it to a new Subfamily. So most of the problems you see here is our laziness. And I honestly dont want to inflate our db with gazillions of the same cores just with an apostrophe on their end. 

I proposed years ago to have a similar organisation like the CPUs in the db. Means: Manufacturer >> Family (architecture) >> Subfamily (Product range) >> Specific Model. I dont know if this scheme is accepted by our owner and will be programmed. Until I dont have an answer, I wont do much work on this cause I dont want to do things twice and the effort for Nvidia would be huge. In the end we would need to a massive cleanup anyway, touching every card, but thats something I dont have time for at the moment. Mistakes like wrong GPU cores and specs will be fixed though, its best to write me a PM if you spot something. 

The question why AMD is organised another way is simple: I had the time for it back then and it felt necessary. The reason was AMD spawned multiple recycled cards with same specs but different names even under other under product families but GPU-Z cant distinguish them. To combine the cards I also had to remove the subfamilies and group every model under their specific architecture. I could ofc do the same for nvidia but again, it would be better to fix this the proper way. 


Quadro RTX 5000/6000/8000 are listed as RT102/104GL cores, but I think should be TU104/102 cores
Quadro RTX A6000 is listed as having a TU102 core, but I think it should be a GA102 core
Geforce GTX 780 (2880 Shaders) is under Geforce 600 series, should this be under Geforce 700 series?
Geforce 830A is under Geforce 900 series, should this be under Geforce 800 series?


Edited by Strunkenbold
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Create New...