Recommended reading

Nvidia GTX 970 memory mod boosts performance with 8GB VRAM upgrade

News
By published

10-year-old GPU nearly doubles its benchmark score after VRAM upgrade

A modded GTX 970 graphics card with 8GB VRAM
(Image credit: Peperaio Hardware)

The Nvidia GeForce GTX 970 was a promising graphics card at launch offering performance similar to AMD’s Radeon R9 290X at a more affordable price range. Ten years later, the GPU has been given a new life thanks to Brazilian modder Paulo Gomes and his team who managed to add more VRAM, doubling the original from 4GB to 8GB.

To increase the memory, the mod involved replacing the original 512MB, 7 Gbps GDDR5 memory modules with 1GB, 8 Gbps chips along with a resistor so that the GPU can recognize the higher density VRAM. By having access to 8GB of upgraded VRAM, the modder noted an improvement in Unigine Superposition benchmark showing nearly double the score.

This one-of-a-kind Asus Strix-branded RX 970 was then handed to Youtuber Peperaio Hardware to evaluate whether increasing VRAM on an older GPU translates to any improvement in gaming. Testing was done using a mix of old and new games where titles like Red Dead Redemption 2, GTA V Enhanced, and Plague Tale Requiem didn’t show any signs of improvement. The results were compared to a Gigabyte Windforce GTX 970 with 4GB of memory.

In Cyberpunk 2077, the results varied depending on the graphics settings with an uplift of 5-15% in performance on the 8GB modded GPU. The two games that gained noticeable benefit from the increased VRAM were The Last of Us Part II Remastered with a 24% increase in frame rate and Horizon Forbidden West offering as much as 40% improvement.

Back in February last year, the same team of modders managed to repair and upgrade a defective RTX 3070 by replacing its original 8GB of GDDR6 memory with 12GB. The team soldered in new 2GB memory chips and modified the GPU’s BIOS to correctly detect and utilize the additional VRAM. After the upgrade, the graphics card operated stably and passed all stress tests, showing that the mod was technically viable and functionally solid.

The upgrade had led to a performance gain in games including Resident Evil 4 Remake, where the modified RTX 3070 delivered up to a 66% performance increase compared to the standard 8GB version. In The Last of Us Part I, performance saw a boost of around 25%, while Hogwarts Legacy ran 20% better with the extra VRAM.

Both Nvidia and AMD have continued to claim that 8GB of VRAM is sufficient for most gamers. However, real-world mods like the ones mentioned above, suggest otherwise. Even as modern titles become increasingly demanding, Nvidia went on to launch the RTX 5060 and 5060 Ti in 8GB variants, while AMD has announced that its new Radeon RX 9060 XT will also be coming with 8GB of VRAM.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

See more GPUs News
Kunal Khullar
Kunal Khullar
News Contributor

Kunal Khullar is a contributing writer at Tom’s Hardware.  He is a long time technology journalist and reviewer specializing in PC components and peripherals, and welcomes any and every question around building a PC.

10 Comments Comment from the forums
  • usertests
    Finally. This is the amount of VRAM everybody needs. ;)
    Reply
  • Alvar "Miles" Udell
    "GPU manufacturers have been gimping performance for years by limiting VRAM"

    Should be news to absolutely nobody.
    Reply
  • jlake3
    Hm. I'd be curious to know the settings they used in Unigine Superposition, as I have both a GTX 970 and a Quadro M4000 in my collection, which are both GM204 chips with 1664 CUDA cores. The M4000 has more VRAM and ROPs while the 970 has higher clocks and a higher TDP (with a stronger cooler), and in my experience, the GTX 970 performs better by just a bit less than what the clock difference would imply in not just Unigine but in everything else I tested... except in Superposition's "4k Optimized" preset, where the M4000 turned in a score but the 970 warned at launch that it did not have enough VRAM, so I didn't bother attempting a run.

    The M4000's Superposition 4k Optimized score scaled up to the 970's clockspeed would be about on par with the RX 580 8gb, for reference.
    Reply
  • mac_angel
    jlake3 said:
    Hm. I'd be curious to know the settings they used in Unigine Superposition, as I have both a GTX 970 and a Quadro M4000 in my collection, which are both GM204 chips with 1664 CUDA cores. The M4000 has more VRAM and ROPs while the 970 has higher clocks and a higher TDP (with a stronger cooler), and in my experience, the GTX 970 performs better by just a bit less than what the clock difference would imply in not just Unigine but in everything else I tested... except in Superposition's "4k Optimized" preset, where the M4000 turned in a score but the 970 warned at launch that it did not have enough VRAM, so I didn't bother attempting a run.

    The M4000's Superposition 4k Optimized score scaled up to the 970's clockspeed would be about on par with the RX 580 8gb, for reference.
    GTX 970 has faster memory, 224GB/s vs 192GB/s
    GTX 970 has a lot faster clock speed, 1178MHz boost vs 773MHz. And remember, that's per CUDA core, so that difference is applied to 1664 CUDA cores.
    Drivers probably have some differences, too.

    As for my own comment. I think this is just more proof that NVidia purposely gimps their GPUs - planned obsolescence.
    Reply
  • FunSurfer
    But... was it really 8GB of VRAM or 7GB + 1GB slow VRAM? Why nobody bothered to ask THE question?
    Reply
  • jlake3
    mac_angel said:
    GTX 970 has faster memory, 224GB/s vs 192GB/s
    GTX 970 has a lot faster clock speed, 1178MHz boost vs 773MHz. And remember, that's per CUDA core, so that difference is applied to 1664 CUDA cores.
    Drivers probably have some differences, too.

    As for my own comment. I think this is just more proof that NVidia purposely gimps their GPUs - planned obsolescence.
    I own both cards, and I've run both of them in Superposition on exactly the same hardware. I know experimentally what the performance delta is between the two of them when using less than the 970's 4gb VRAM limit. Theoretically; 1178MHz/773MHz = 1.53, so the GTX 970 should be a minimum of 53% faster, plus some additional benefit from faster memory... but in practice, it does worse than 53%:

    (Superposition DirectX results)M4000GTX 970Difference720p low135781896637.85%1080p med5115753247.25%1080p extreme1415199140.71%4k optimized2278Over VRAMN/A
    Using the M4000 as a proxy because I don't have an 8gb GTX 970, it's pretty hard to find situations that would occur naturally where the 970 is VRAM limited but (M4000's FPS x 1.53) would still be a good experience. That had me suspecting they were running some unreasonable settings that you would not normally encounter to get that doubling of benchmark scores... which digging into the sources, is exactly what they did. They cranked the settings to "8K Optimized", and scored 6.5FPS without the mod, and a still-unplayable 11.8FPS with.

    It's interesting to see it built and tested and there's some edge cases where you see a benefit, but it's not a massive swing, and the linked video only seems to have seen a benefit in games that came out 8+ years after the card released and much less than a doubling. Nvidia has definitely made some really bad calls on their newer cards, but the case they intentionally under-VRAM'ed the 970 as part of a nefarious plot is weak.

    FunSurfer said:
    But... was it really 8GB of VRAM or 7GB + 1GB slow VRAM? Why nobody bothered to ask THE question?
    That's due to the memory bus design, not the chips, so this would have the same issue where the last 1/8th of the VRAM is slow. Doesn't look like they push the VRAM hard enough to encounter it in the videos the article links to.
    Reply
  • usertests
    FunSurfer said:
    But... was it really 8GB of VRAM or 7GB + 1GB slow VRAM? Why nobody bothered to ask THE question?
    I think a previous article found it was 7GB + 1GB as @jlake3 stated. Obviously better to have than 3.5GB + 0.5GB though.

    This mod was talked about in March, and yes, it's 7 GB + 1 GB:
    https://videocardz.com/newz/geforce-gtx-970-gets-new-life-brazilian-modders-upgrade-memory-to-8gbhttps://hackaday.com/2025/03/25/brazilian-modders-upgrade-nvidia-geforce-gtx-970-to-8-gb-of-vram/https://www.techpowerup.com/334585/brazilian-modders-give-8gb-memory-to-geforce-gtx-970
    Reply
  • das_stig
    1981 Bill Gates - 640KB is enough for anybody
    2025 NV/AMD - 8GB is enough for anybody

    Disclaimer,

    Myth, BG never said this.
    NV/AMD will also sell you the identical card, with a different name, with more ram, at an extortionate price, claiming massive performance improvements, while denying they are ripping off the customer.
    Reply
  • mac_angel
    jlake3 said:
    I own both cards, and I've run both of them in Superposition on exactly the same hardware. I know experimentally what the performance delta is between the two of them when using less than the 970's 4gb VRAM limit. Theoretically; 1178MHz/773MHz = 1.53, so the GTX 970 should be a minimum of 53% faster, plus some additional benefit from faster memory... but in practice, it does worse than 53%:

    (Superposition DirectX results)M4000GTX 970Difference720p low135781896637.85%1080p med5115753247.25%1080p extreme1415199140.71%4k optimized2278Over VRAMN/A
    Using the M4000 as a proxy because I don't have an 8gb GTX 970, it's pretty hard to find situations that would occur naturally where the 970 is VRAM limited but (M4000's FPS x 1.53) would still be a good experience. That had me suspecting they were running some unreasonable settings that you would not normally encounter to get that doubling of benchmark scores... which digging into the sources, is exactly what they did. They cranked the settings to "8K Optimized", and scored 6.5FPS without the mod, and a still-unplayable 11.8FPS with.

    It's interesting to see it built and tested and there's some edge cases where you see a benefit, but it's not a massive swing, and the linked video only seems to have seen a benefit in games that came out 8+ years after the card released and much less than a doubling. Nvidia has definitely made some really bad calls on their newer cards, but the case they intentionally under-VRAM'ed the 970 as part of a nefarious plot is weak.


    That's due to the memory bus design, not the chips, so this would have the same issue where the last 1/8th of the VRAM is slow. Doesn't look like they push the VRAM hard enough to encounter it in the videos the article links to.
    NVidia didn't make "bad calls", they do it on purpose. It's called "planned obsolescence". The 'dumb down' their cards for retail use, and put the bare minimum of VRAM in the cards that they can get away with. Not to save money, but so that the cards have no 'future proof' aspect. They can easily put 12 and 16GB in their entry cards and make them last longer for up coming games, but that's not what NVidia wants. They want consumers to have to keep buying another GPU every generation, or two at max. The only way to come close to 'future proofing' your GPU is buying their x090 series, and even that's limited with NVidia changing, editing, or adding new tech and features into their newer cards.
    Reply
  • The Historical Fidelity
    Can anyone confirm if the memory bus is running at 8 Gbps? Because adding 8 Gbps capable memory chips is very different from the memory controller being able to run the chips at that speed. Most likely the memory controller is still running at 7 Gbps
    Reply