Nvidia Graphics Card Comparison

[Slaughtahouse]

Hi all,

 

I've drafted a Google Sheet to document various desktop, gaming Nvidia graphics cards. Items italicized are rumoured or unconfirmed specs, products. 

 

The reason for doing this was to help provide a quick overview of a comparison between graphics cards and their GPUs (size, features etc.) for discussion. This is a live file that will be continuously updated over time.

 

Subject to change: As of today, you'll see two (2) tabs.

• One (1) providing an overview of all the detailed specifications for the RTX series.
• Another tab quickly pulling that data to outline the difference (%) in specifications between the flagship graphics card of a specific series (20, 30, 40 series etc.) and all the other cards of that series. I've opted to draw the comparison against the flagship graphics card vs. the GPU since some GPUs (full dies) never make it to products.
  • Example: There is no RTX series graphics card that features a full Ada Lovelace GPU (AD102).
  • The RTX 4090 is used as the baseline (100%) and all other GPUs in the 40 series are compared against.
  • A column has been added to average the difference between all the specs; No weight has been provided to each spec. See the RTX Series sheet for detailed comparisons and differences between specifications like core count.

 

All specifications have been manually sourced from websites including but not limited to: Videocardz and Techpowerup.

Edited October 24 by Slaughtahouse

[Slaughtahouse]

 

[Snakecharmed]

Great work. I think we all intuitively knew that -80 cards have been getting pushed further downmarket (performance-wise, because they're still insane price-wise) compared to the generation's flagship, but to see it quantified like this really hammers it home. The upsetting part is that there's nothing to fill the widening gap between the -90 and -80, so the only option is to get a used -90 from the previous generation.

[Slaughtahouse]

2 hours ago, Snakecharmed said:

Great work. I think we all intuitively knew that -80 cards have been getting pushed further downmarket (performance-wise, because they're still insane price-wise) compared to the generation's flagship, but to see it quantified like this really hammers it home. The upsetting part is that there's nothing to fill the widening gap between the -90 and -80, so the only option is to get a used -90 from the previous generation.

 

That reminds me, I’ll add MSRP!

[Slaughtahouse]

I’ve made some small tweaks to the language. It’s not as short as Id like it but I feel it’s clearer. Looking for feedback here if there is better verbiage.

 

- For all columns highlighting the % of the specification (when compared to the flagship of that generation), I removed “difference” and revised to “Available (vs. flagship)”.

 

Since “difference” wasn’t objectively accurate and I want to highlight how much of a given specification is available or remaining after a graphics card is cutdown or modified by Nvidia vs the flagship of that gen. Example: A 4080 Super’s CUDA core count is effectively 63% of a 4090’s CUDA core count. 10240 are available out of the possible 16348. 

 

I also included a new column at the very end which highlights the average, absolute difference. In other words, it tells you much a card is cut down.

 

So if the average available specs of RTX 4080 Super is 62% (CUDA cores , bus width, RT cores, Tensor cores, SMUs, ROPs etc.) vs. a 4090, the average absolute difference is 38%. 

 

Just a different way to express the same information. Really comes down to if you’re a glass half full or half empty kinda person XD

Edited October 26 by Slaughtahouse

[bridgypoo]

Great job! As a fellow data nerd and spreadsheet person I appreciate the way you've laid this out.

[Snakecharmed]

While the average absolute difference column is technically correct, I wonder if removing the word "absolute" in the column header and just putting minus signs in front of the percentage values would be more intuitive to the viewer.

 

Whether the flagship baseline is 100% or 0%, a worse performer should not show a numerically higher value.

[Slaughtahouse]

52 minutes ago, Snakecharmed said:

While the average absolute difference column is technically correct, I wonder if removing the word "absolute" in the column header and just putting minus signs in front of the percentage values would be more intuitive to the viewer.

 

Whether the flagship baseline is 100% or 0%, a worse performer should not show a numerically higher value.

 

Done and done! Nice catch, I was subtracting the values in the wrong order. It now shows correctly

[Slaughtahouse]

I’ve made some small updates to the sheet for Blackwell GPUs. This is a result of various, new leaks for the 5090, 5070 Ti.

 

https://videocardz.com/newz/nvidia-geforce-rtx-5090-gb202-gpu-die-reportedly-measures-744-mm2-20-larger-than-ad102

 

https://www.techpowerup.com/329123/nvidia-geforce-rtx-5070-ti-specs-leak-same-die-as-rtx-5080-300-w-tdp

[SamsTechStuff]

Very cool! Would it make sense to rearrange the newest series towards the top of the sheets vs scrolling down? Also, I like the little bit of color on the first page, I think adding some color maybe for the flagships to help make it a little easier to read quickly / at a glance . Maybe a light green fill or one of the cells or something.

[Slaughtahouse]

1 hour ago, SamsTechStuff said:

Very cool! Would it make sense to rearrange the newest series towards the top of the sheets vs scrolling down? Also, I like the little bit of color on the first page, I think adding some color maybe for the flagships to help make it a little easier to read quickly / at a glance . Maybe a light green fill or one of the cells or something.

 

I can adjust the pivot table, reorder to descending order for “GPU Series” and apply a condition format the cell to any GPU that is a flagship.

 

For the data page, where the pivot pulls from, Id personally rather leave it as is and let it be a data dump in its current format. As that’s why I did the pivot, to quickly visualize the comparison for discussion

Edited November 23 by Slaughtahouse

[Sir Beregond]

One correction: RTX 4080 had the $1199 MSRP. MSRP for the RTX 4080 Super was $999.

[Slaughtahouse]

7 hours ago, Sir Beregond said:

One correction: RTX 4080 had the $1199 MSRP. MSRP for the RTX 4080 Super was $999.

Thanks. Adjusted

 

Other: In the near future, I’ll begin to plug in GTX Geforce cards, starting with Pascal and going back to Fermi.

 

Ultimate goal will be to go back to the start (256). Baby steps…