Sludig
Potato del Grande
- 11,201
- 12,090
5070 not a consideration. Though those small leaps apply to every card, 5080 is only like 15% better than the ti.
NVidia GeForce RTX 50x0 cards - 70% performance increase, but AI > you
- Thread starter Izo
- Start date
Prodigal
Shitlord, Offender of the Universe
- 1,501
- 1,969
I picked up a 5080 from Best Buy when they had Cyber Monday “sale” - replaced my 5070Ti since I was moving to a 4K monitor. 5070Ti was solid for my 1440p screen.
Debating whether to sell my 5070ti and also the 3080ti it replaced - probably just sell the 5070ti and have the other for back up.
- 1
Punko
Macho Ma'am
- 8,315
- 13,375
Cyberpunk looks amazing on my 4080.
A while ago I looked into optimizing the settings, which only made things look better and improved performance.
Might be worthwhile for you.
Hateyou
Not Great, Not Terrible
- 18,524
- 50,088
100% keep one for a backup. It would suck to have a card die and have to shell out $1-3k for a quick replacement.
jayrebb
Naxxramas 1.0 Raider
- 15,816
- 15,811
Had GPU cables fail and wouldn't have been able to determine the problem and fix it myself if I didn't have a backup GPU to affirm the problem.
For that reason I'm not big on selling old cards. It can take longer to diagnose something otherwise-- and even with Microcenter when my CPU failed, I had to guide them through their own troubleshooting because "They couldn't take a customers word for it" (which I understand). I put 20+ hours into diagnosing the CPU failure and it paid off as I steered the repair team toward CPU failure throughout their extensive diagnostic process via texting back and forth. I saved them a lot of time steering them away from RAM issues and GPU issues. Only thing I couldn't 100% eliminate as a possible problem was the motherboard which was the most frustrating aspect of it. So I had to say CPU/Mobo. I don't consider myself ultra computer literate, but my tenacious days long googlefu outperformed their whole team and I only brought the rig in when I was positive I had it narrowed down to CPU failure. Initial symptom was memory corruptions in Windows.
Last edited:
Hekotat
FoH nuclear response team
- 12,976
- 13,183
I always keep my old card since prices went insane, I know it'll fail at a bad time. I kept my 7900XTX even though I should sell it.
Get one of those emulator retro console things with 30k+ games or whatever, or start hitting Facebook marketplace for consoles and games that dont require online components to work.
I used to think downloading games would be an exciting future, but with the idea that you rent and never own, games can be delisted and cant be played if not connected online, vid cards being rented and having an hour limit...etc...may very well just go back to the Genesis, SNES up to PS4 era and look for.
$3k+ for cards now, fuck that. Dont care how much I like gaming, Ill never support it.
- 1
- 1
Grabbit Allworth
Confirmed J6 Insurrectionist
- 2,029
- 8,244
Two of the most common, preventable reasons that a card dies are:
1. Not being supported. Please, for the love of everything, use the support bracket that comes with the card and if it doesn't, find something that can prop that front right corner (assuming standard build) up.
2. Running them on old and/or underpowered PSUs. If you do this, your card is going to get a lot hotter than it should and the constant flex of heating up and cooling down pushes it beyond its spec limits and dies early. Easily avoidable by buying a PSU that is rated 10-20% higher than what your system requires. E.g. Need 750, buy an 850.
If you do have a dead card, don't toss it. A lot of them can be repaired and with how expensive they're getting, the $150-$400 repair costs are a lot better than buying a new one.
1. Not being supported. Please, for the love of everything, use the support bracket that comes with the card and if it doesn't, find something that can prop that front right corner (assuming standard build) up.
2. Running them on old and/or underpowered PSUs. If you do this, your card is going to get a lot hotter than it should and the constant flex of heating up and cooling down pushes it beyond its spec limits and dies early. Easily avoidable by buying a PSU that is rated 10-20% higher than what your system requires. E.g. Need 750, buy an 850.
If you do have a dead card, don't toss it. A lot of them can be repaired and with how expensive they're getting, the $150-$400 repair costs are a lot better than buying a new one.
- 1
- 4,295
- 8,061
Had an old out of date PC under my desk. Took the old ass video card out of it (AMD 580X I think) for 'in case'.
I have a 7900XTX now that I love. Not sure I'll buy an Nvidia again , they are just too damn $$$$ for the return. Love love love to play video games but damn the card costing more than the whole rest of the PC. RAM seems to be going back to the bad old days of P3-P4 when RAM Cost was a third of the whole PC. I went AMD then too.
I have a 7900XTX now that I love. Not sure I'll buy an Nvidia again , they are just too damn $$$$ for the return. Love love love to play video games but damn the card costing more than the whole rest of the PC. RAM seems to be going back to the bad old days of P3-P4 when RAM Cost was a third of the whole PC. I went AMD then too.
- 1
- 11,417
- 2,503
Well the way it's going there might not be a next Nvidia card to buy anyway so it's going to be an easier "choice" from now on.
- 1
- 1
- 4,295
- 8,061
Hmm we all should hope not , or those 'affordable' AMD cards will be priced like the Nvidia cards. I don't see Intel staying in the market and they have yet to make anything I would buy anyhow.
Jovec
?
- 912
- 512
I think there will be future NV cards (and AMD), but NV will hard-code DLSS and eventually FG into every card except the X090 series. As an example with current gen, imagine a 5070 that was really a 5060 die but forced DLSS quality at the hardware/hidden driver level. The user could select the balanced or performacne level manually, but never native rendering. Same price, but smaller die.
I think most people are okay with using modern upscaling, so there will be some online bitching about lack of native rendering, but that will die down quickly until it just becomes the norm. Then follow that up with 1x FG hard-coded in the next gen, with the user able to select multi-FG options, but never disable it.
AMD might try to keep native rendering as a feature, but maybe only for a generation before they follow suit. NV has already laid the groundwork for this with their failed 4080 12GB unluanch (trying to sell the 4070 TI as a 4080) and their 5070 marketing (4090 performance for $550).
Jovec
?
- 912
- 512
GPU ranks are roughly 9070XT = 5070Ti < 5080 < 4090 < 5090. In fairness, you are running an extremely high-end card and it should perform well for years and years to come. Eventually though peformance and (lack of) features will start to show. HUB did some quick benchmarks of Nvidia's DLSS 4.5 (beta?) and in some games, older cards like the 2080 Ti and 3090 saw worse than native performance when using DLSS 4.5 since they don't have good hardware support for the instructions it uses. Eventually there will something that the 5070Ti doesn't do well enough at and that will prompt an upgrade.
Last edited:
- 3,035
- 4,911
This may never work, at least not for now. Modern APIs like Vulkan and DirectX 12 are more direct, bare-metal GPU interfaces compared to OpenGL or DirectX 11. Over the years we basically went all the way from the GPU and driver managing the rendering stage themselves (e.g. with OpenGL display lists, that were deprecated in OpenGL 3.1 or so), to more low level control of the GPU (what OpenGL called "immediate mode" back in the day).
Basically how Vulkan works, is that the application/game has to manage all the direct low level GPU details instead of the API abstracting away most of them, then putting GPU rendering calls in CommandBuffers, and then queueing them to the GPU. And it also allows pretty low level access to GPU memory management. It's the application's job to queue in the right order and to make sure it doesn't trip over itself in a multi-threaded environment.
How upscaling (and frame generation) usually works, is that only the rendered field (basically the 3D world) is upscaled and the user interface is overlayed on top of it. Sometimes there are multiple overlays over each other, and the graphics API wouldn't even know what to upscane and what not without the game specifically telling it.
In addition to that, what upscaling really is: It is basically a small AI model that is running on the GPU, and takes up a bit of memory. DLSS 4 has been optimized for RAM usage, but it's still 100 MB or so for the 1080p model, and 300 MB for the 4K model. You could theoretically cast the model into static hardware, but Nvidia is steadily optimizing and re-training the model, with driver updates delivering updates. Having the driver load the upscaling model into memory would make no difference to how it currently works.
My personal opinion: Frame generation is basically bullshit, because the esports people that really care about really high FPS (>200) are the same ones that don't use it because of the render lag, which is the reason they want high FPS in the first place. It's also not linear, e.g. a 30 FPS native game doesn't magically get 60 FPS with 2x generation, more like around 48 FPS.
- 2
Jovec
?
- 912
- 512
It's easily enabled/forced by hiding driver options; it doesn't have to be at the hardware level. Right now, you can run DLSS off (native), or Quality, Balanced, or Performance. NV simply forces DLSS Quality to always be on, then creates three new tiers of upscaling they call Quality, Balanced, and Performance that the user can pick from.
Nvidia already tried to compare the 5070 to the 4090 at launch because they used DLSS and 4x Frame Gen on the 5070 against native on the 4090. They did get laughed at in certain circles for it, but sentiment is changing. Nvidia has also been down-sizing (by percentage) Cuda cores of the 80-and-below series against the flagship. IOW, the 3070 was closer to the 3080 than the 5070 is to the 5080. All the groundwork is there, and in a year or three when the 6060 or 7070 costs $1000, gamers either won't care or will just begrudginly accept that software tricks are replacing hardware horsepower.
Frame Gen will likely follow a generation later with 1x being forced as always on in the driver. Have an anti-lag setting that automatically disables 1x FG and lowers DLSS quailty one level (all behind the scenes) when base FPS is too low for FG to be effectively used.
- 1
Folanlron
Trakanon Raider
- 2,487
- 783
You have current generations engine that are a fucking mess, Unity/UE5.
Any change like this would be a massive over haul, not only for engines but for EVERYTHING, dx12(a large section of dx would need a ground up rework)/OpenGL/Vulkan would all have to be "updated".
If this does happen it will be a long way out, although AMD kind of has a head start on that with Navi for the Ps4/5/XboxX3x6x0 whatever fucking number they use.
Jovec
?
- 912
- 512
Nothing needs to be modified for forced upscaling and FG besides the drivers. Not engine nor APIs. Engine integration makes it better, but is not required. See Lossless Scaling, for example. On one hand, upscaling can be viewed as a software optimization assuming the image quality is indistinguishabe from native, but FG has a long way to go. FWIW, Nvidia just annouced MFG 6x for the 5000 series along with variable FG, so you set a desried FPS and the driver will increase/decrease MFG to keep you at that FPS. Presumable DLSS can easily be added into this logic, since it might be better to dynamically adjust the render resolution down if that means doing less FG.