• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

TPU's GPU Database Portal & Updates

Hi guys,

I think there's something wrong in the charts... Since when is the RTX 3090 Ti about 16% faster than the RTX 3090 ? Lol (If I remember well it's supposed to be around 8-10%)
~2.44% more CUDA Cores, ~9.3% faster Core Clocks and ~7.7% more Memory Bandwidth ! I think you need to re-run some benchmarks


1706313003043.png
 
3090ti pumps in much more power tho it comes pretty much extremely overclocked out of the box thats where your 6% comes from since fe model dont.

10% regular uplift + the hefty oc= +6% 450+ Watts but its like 2ghz out of the box depending on the game of course and model
 
Thank you! I appreciate the feedback and re-run! @W1zzard

So the 3090 was power constrained then... Ampere seems to scale better than Lovelace when overclocked!
I think Lovelace is too Memorg Bandwidth starved (GDDR6X & L2 Cache), and I'm pretty sure a 4090 with just 96MB L2 Cache (instead of 72MB) would see a 10%+ increase in performance

Curious to see what Blackwell will bring though!
W1zzard said:
I just looked into this for you, with my newest benchmarks from nov 2023 the delta is 13%

View attachment 331905
Click to expand...
 
GPUs have been power constrained for a long while now. Very heavily and noticeably at least since Kepler.

3090 power limit is 350W.
3090ti power limit is 450W.
450/350 = 1.28

Yes, diminishing returns at this point but that is still a huge difference.
The other factor with test results is that games seem to have gone more compute heavy and more efficient in using GPU resources since 3090/Ti release.

Edit:
Looked at the TPU 3090Ti launch review and yes, it has 3090Ti 8% faster than 3090. But also I am quite sure 5800X can be a detriment at that level :)
 
Last edited:
londiste said:
GPUs have been power constrained for a long while now. Very heavily and noticeably at least since Kepler.

3090 power limit is 350W.
3090ti power limit is 450W.
450/350 = 1.28

Yes, diminishing returns at this point but that is still a huge difference.
The other factor with test results is that games seem to have gone more compute heavy and more efficient in using GPU resources since 3090/Ti release.

Edit:
Looked at the TPU 3090Ti launch review and yes, it has 3090Ti 8% faster than 3090. But also I am quite sure 5800X can be a detriment at that level :)
Click to expand...
I think GPUs are mostly voltage limited though. You can't get card that goes to 1.2V after all.
Main culprit for this is plain transistor counts we have now (more transistors = more power usage at the same voltage).

Recent RTX 4080 Super is great example of that.
Because more active CCs and TMUs = more heat, turbo boost gains can get into negative (so Super card can get lower frequency than non-Super), at the same voltage levels. End results is the exact same performance for both cards when VRAM bandwidth isn't as important for Super.

I wonder when we get cards sold by voltage they utilize, instead of frequency ranges they go to.
Examples : We buy low voltage card [0.7/0.8V] for lower power bill, cooler requirements (and having downside of 10-20% lower performance), and highest voltage version [~1V] for unlocked OC limits and highest performance.
 
Thank you! I have already uploaded the vbios. Sadly, Inno3d still provides +0% power limit on all their 4000 series boards. I know why they do it, but I don't agree with it. You do get a tangible benefit for increasing the power limit, the most important one being that when you put core voltage to +100%, so it goes to 1100mV, it doesn't fluctuate that much. At 320W, it fluctuates alot from ~1050 to 1100, while at let's say 350w, you get locked inot that 1095-1100mV value.
 
I can't find the nvidia cmp 100-200 6gb and 100-210 12gb cards. These are Titan V based mining cards. I need info for the pcie limitations. Tensor cores are disabled?

So what i find:
CMP 100-200 is a half memory TitanV, with the titanV cooling system. Possible unlock the other half vram. 6+8 pin pcie power
CMP 100-210 is a 12Gb vram Tesla V100, with the tesla stock passive cooling but missing the two bright platic shourd. 8 pin EPS power.

There is a possibility to the pcie lanes are limited, tensor core disabled. I only see these cards on the USA market.
 
Last edited:
masterdeejay said:
I can't find the nvidia cmp 100-200 6gb and 100-210 12gb cards. These are Titan V based mining cards. I need info for the pcie limitations. Tensor cores are disabled?

So what i find:
CMP 100-200 is a half memory TitanV, with the titanV cooling system. Possible unlock the other half vram. 6+8 pin pcie power
CMP 100-210 is a 12Gb vram Tesla V100, with the tesla stock passive cooling but missing the two bright platic shourd. 8 pin EPS power.

There is a possibility to the pcie lanes are limited, tensor core disabled. I only see these cards on the USA market.
Click to expand...
Thanks for the update!!!

https://www.techpowerup.com/gpu-specs/cmp-100hx-210.c4202

a few things:
The 100-210 uses the V100 PCB, it has one 8 Pin EPS power connector
pcie is possible works 1x mode only like on the most of the CMP cards
 
Last edited:
Hey, when cataloging cards from around 2000 and such, who were you guys using when you would look for first-party reviewers/information about specs?
 
Hi experts! I have quite a technical question:

To my knowledge, TPU is the only database which offers a user friendly access to identify, check and compare subsystem IDs for graphics cards (e.g., "Subsystem Id: 148C 2422"). The second block of this ID normally indicates the exact card model, e.g. "Subsystem Id: 1849 5301" indicates the ASRock Aqua (7900 XTX).
But there are manufacturers who use the same model ID for all of their different models. Powercolor uses the same ID (Subsystem Id: 148C 2422) for all 7900 XTX models, Hellhound black and white, Red Devil, Liquid Devil.

Is there any alternative way to identify the exact model if the second subsystem ID if the same code is used for all different models?
@T4C Fantasy
 
Thank you all, I can see comments spanning twelve years in one post, and these databases are very helpful to me.
 
No idea if that was already suggested (and I apologise if it was) but I really wanna have the dimensions columns with sorting being available there. Length and width are important for many users.
1728396450265.png
 
ApolloX said:
Hi experts! I have quite a technical question:

To my knowledge, TPU is the only database which offers a user friendly access to identify, check and compare subsystem IDs for graphics cards (e.g., "Subsystem Id: 148C 2422"). The second block of this ID normally indicates the exact card model, e.g. "Subsystem Id: 1849 5301" indicates the ASRock Aqua (7900 XTX).
But there are manufacturers who use the same model ID for all of their different models. Powercolor uses the same ID (Subsystem Id: 148C 2422) for all 7900 XTX models, Hellhound black and white, Red Devil, Liquid Devil.

Is there any alternative way to identify the exact model if the second subsystem ID if the same code is used for all different models?
@T4C Fantasy
Click to expand...
Just clock speed, if those are the same then no, just lazy gpu companies

Beginner Macro Device said:
No idea if that was already suggested (and I apologise if it was) but I really wanna have the dimensions columns with sorting being available there. Length and width are important for many users.
View attachment 366613
Click to expand...
The issue is making it fit nicely, that has always been the challenge, so we figure if you see a change you can click on the card for full dimensions.

Edit: ill look into sorting.
 
T4C Fantasy said:
that has always been the challenge
Click to expand...
Why? Size is the same type of data as the frequencies. The only issue here is how to fill all these entries. Will take at least a dozen eternities manually and I got no clue how to automate it.

At least this is how I see it.
 
Old stuff
NVIDIA GeForce RTX 3050 6 GB single slot low profile cards:
https://yestonstore.com/products/yeston-rtx-3050-6gb-gddr6-gaming-graphics-card - Yeston RTX 3050-6G D6
https://www.aliexpress.com/item/1005007788674782.html - ASL RTX 3050 Sabre Slim 6GD6

AMD Radeon RX 6500 single slot low profile card:
https://www.dell.com/en-au/shop/del...raphics-card/apd/490-bjhy/graphic-video-cards - Dell F6KP8

AMD Radeon RX 6600 single slot low profile cards:
https://www.fsi-embedded.jp/product_detail/12282/ - FSI EP23GL-LP4AU
https://www.advantech.com/en-us/pro...8-6t/mod_9ce2adad-4476-453c-8571-b925f18866f5 - Advantech 96VG-A6600FX8-6T
https://www.advantech.com/en-us/pro...8-6t/mod_f10583f8-bb9c-41f3-b923-69d8a76303aa - Advantech 96VG-A6600LX8-6T

New stuff
NVIDIA RTX 2000E Ada 900-5G192-2220-E00 cards:
https://www.pny.com/en-eu/rtx-2000e-ada-generation - PNY VCNRTX2000EADA-SB
https://www.advantech.com/en-us/pro...a-16/mod_b7f15a87-0d17-4661-a00b-9ef4b75bbd95 - Advantech
https://www.adlinktech.com/Products/Embedded_Graphics/PCIe_Graphics_Card/NVIDIA_RTX_2000E_Ada - ADLINK
 
NVIDIA GeForce4 MX 440-8x VER-B
it is my photos, for free, if need
 

Attachments

  • 20241121_105733.jpg
    20241121_105733.jpg
    2.4 MB · Views: 44
  • 20241121_105813.jpg
    20241121_105813.jpg
    1.1 MB · Views: 37
  • 20241121_105839.jpg
    20241121_105839.jpg
    1.7 MB · Views: 42
  • 20241121_110424.jpg
    20241121_110424.jpg
    2.8 MB · Views: 50
  • NVIDIA GeForce4 MX 440-8x VER-B.jpg
    NVIDIA GeForce4 MX 440-8x VER-B.jpg
    2.7 MB · Views: 52
sup0s said:
Hey, I have this card ! Here is the specs.
Click to expand...
Upload the bios to the vga bios collection, be detailed about it please
 
Hello @eidairaman1
I would love to upload the vBIOS my new office graphics card.
It's a Nvidia GeForce GT 1030 4GHD4 LP OC with 4 GiB DDR4 as vRAM.

It's already in the GPU database, but vBIOS is not in the Video BIOS Collection.

MSI GT 1030 Low Profile Passive OC DDR4 Specs

NVIDIA GP108, 1430 MHz, 384 Cores, 24 TMUs, 16 ROPs, 2048 MB DDR4, 1050 MHz, 64 bit
www.techpowerup.com

Thanks
 

Attachments

  • GPU-Z_GT1030.gif
    GPU-Z_GT1030.gif
    39.2 KB · Views: 28
  • GP108.rom
    GP108.rom
    257 KB · Views: 19
www.techpowerup.com

MSI GT 1030 Low Profile Passive OC 4 GB DDR4 Specs

NVIDIA GP108, 1430 MHz, 384 Cores, 24 TMUs, 16 ROPs, 4096 MB DDR4, 1050 MHz, 64 bit
www.techpowerup.com www.techpowerup.com
Tanzmusikus said:
Hello @eidairaman1
I would love to upload the vBIOS my new office graphics card.
It's a Nvidia GeForce GT 1030 4GHD4 LP OC with 4 GiB DDR4 as vRAM.

It's already in the GPU database, but vBIOS is not in the Video BIOS Collection.

MSI GT 1030 Low Profile Passive OC DDR4 Specs

NVIDIA GP108, 1430 MHz, 384 Cores, 24 TMUs, 16 ROPs, 2048 MB DDR4, 1050 MHz, 64 bit
www.techpowerup.com

Thanks
Click to expand...

i cant upload the bios but the lookup button should work now
 
Tanzmusikus said:
Hello @eidairaman1
I would love to upload the vBIOS my new office graphics card.
It's a Nvidia GeForce GT 1030 4GHD4 LP OC with 4 GiB DDR4 as vRAM.

It's already in the GPU database, but vBIOS is not in the Video BIOS Collection.

MSI GT 1030 Low Profile Passive OC DDR4 Specs

NVIDIA GP108, 1430 MHz, 384 Cores, 24 TMUs, 16 ROPs, 2048 MB DDR4, 1050 MHz, 64 bit
www.techpowerup.com

Thanks
Click to expand...
I do not do anything related to the bios collection as of the database level,that is @W1zzard or @T4C Fantasy Fantasy responsibility.
 
You must log in or register to reply here.
Back
Top