News Brazilian modders double Nvidia RTX 2080 memory but only see a 10% performance boost

Status
Not open for further replies.
I think a 10% performance increase actually sounds pretty impressive.
Look at the (lack of) difference that double VRAM made for the 4060 Ti. I think the 7600 XT was closer to +10% against the 7600 if I'm remembering right, but that also has higher clocks.
TechPowerup does RE4 benchmarks. They show a 2.8% performance increase for the 4060 Ti 16GB over the 8GB at 4K. And around a 5-6% increase for the 7600 XT over the 7600.

edit: my mistake, I missed the bit in the second to last paragraph about average framerates being about the same, and it's the 1% lows that saw a 10% increase. That makes a lot more sense. Apologies! Move along! : P
 
Last edited:
I don't think most benchmarkers are even accounting for the fact that some game engines are silently degrading textures and pulling other tricks when VRAM runs out.

So a 7600 XT looks rather unimpressive in price/perf charts, and it was a little overpriced considering the 6700 XT. But doubling VRAM from 8 GB to 16 GB can be beneficial.
 
No clock speed bump by default, but it did get a slight power increase, so should be able to go a little higher if cooled well enough.

7600 XT isn't super overpriced, so it has that going for it at least. $50 for 8GB of memory, fine. $100 not so good Nvidia, on top of the higher starting price. ($90 right now)
 
Interesting experiment. I am not surprised that they only got 10% but that is still impressive in of itself. I'd be curious how long these chips last. I hope the modders test the longevity of these chips keeping the community updated in the process. Good Job!!
 
  • Like
Reactions: AgentBirdnest
In many ways, newer entry level cards are far more crippled than cards in the past, which makes a small VRAM pool far more detrimental. For example, 8GB of VRAM and a PCIe 4.0 X8 interface, basically means that after around 500-700MB or so of shared memory use, the card will encounter major hitching issues unless the game engine avoids loading bandwidth intensive stuff such as actively used textures and other bulk data into the system RAM. On the other hand, a card with PCIe 4.0 X16, you can often load ~1.5GB or so before major performance issues creep in. If a card has ample VRAM, then the bottleneck of an X8 interface becomes less of an issue, as the PCIe interface is less likely to encounter a situation where it needs to pull double duty.
 
  • Like
Reactions: usertests
The whole point of extra VRAM is smoother frame rates because it doesn't have to load textures from RAM/SSD.
So it's working exactly as expected.

They didn't exactly change the RAM clock speed, memory bus width, or number of active shaders, so of course it's not going to see higher max fps.

Correct. The density of chips is only changed, from 1gb to 2gb, with same specs.

You cant just install much faster - it would require adjustments in bios files which... most cards today don't even accept.
 
Would be interesting for this card to see the stability diffusion image generation times in terms of it/s oh low/regular VRAM settings vs High.

Aside from crappy local AI “art” there are other positive AI/ML implications for this as well especially for students with limited budgets needing more VRAM to train their models.
 
Memory doesn't necessarily improve FPS, sure it adds extra space to store data needed by shaders and what not... but most memory transfers happen asynchronous anyway so if that were to stall it would only result in a frame drop here and there.

TLDR: more memory, more stable frame rate.
 
Status
Not open for further replies.