Nvidia revives canceled RTX 4080 as 'new' 12GB RTX 4070 TI Nvidia this week quietly, and perhaps unintentionally, revealed the previously canceled RTX 4080 12GB would be reborn as the RTX 4070 TI. The GPU-giant pulled the card shortly after its launch at GTC this fall, after facing backlash for launching two RTX 4080s based on entirely different GPU dies and sporting a hefty $300 …
"Maxed out" is generally misleading, as most applications benefit from having GPU memory assigned to them even if they aren't actively doing anything at that moment.
Every top-level window or browser tab will have several MB of GPU allocated, for double-buffer and compositing, alt-tab "peek" etc, plus whatever GPU acceleration the application uses (smooth scrolling etc). Most "normal" desktop applications don't actively deallocate, they leave it to the OS/driver. That adds up pretty quickly, eg lots of browser tabs etc.
The GPU driver will evict inactive applications when something else actually wants the memory, but until they do, it'll stay "used", even though it isn't really.
There may also be optimisations based on keeping hold of old frames and not bothering to actually re-run processing if the driver can prove it would produce the same result, as that's very likely to be true of most pixels in 2D applications.
Rather like the OS using all your RAM - it's cache, it'll get dropped in a few nanoseconds if anything has an actual use for the memory.
It is annoying as I constantly go between different applications across multiple virtual desktops (joys of ADHD), so things like evictions happen often and there are considerable lags when switching. Especially moving between desktops, I can see everything redrawing sometimes even for a couple of seconds. Couple of seconds can easily make me lose focus and jump onto something else.
I probably should upgrade to 24GB one and have some peace of mind...
And my main card is 4GB without that kind of lag, and I regularly use a desktop and a laptop with "onboard" graphics (sharing main memory) which rarely use more than about 1GB without that kind of lag. I think it depends only partially on the graphics card memory, mostly on the OS (and its memory policy), the window manager, the mix of applications and other things such as the amount of main memory, speed of main storage, even network connections.
And to think that RiscOS on the original Archimedes introduced solid window dragging (up until then, other window managers could only drag window dotted outlines) even on the lowliest A305 which had 512kB of memory, some of which was used as the video framebuffer. The RiscPC with its (maximum) 2MB of dedicated VRAM made that dragging buttery smooth :-)
At work, some of the worst culprits for lag are Office apps, which (particularly Outlook) are network-bound. Watching the resource manager while trying to swap back to an already-loaded-but-minimised Outlook is interesting - there is an initial flurry of activity for "CPU" and "Disk 0", and maybe a bit of shuffling in "Memory" and a few bursts on "Network", but then the lines are fairly flat and low while the Outlook window itself appears, pretty much unresponsive, indeed often has "(Not Responding)" in the title bar.
Then again, my main work machine is pretty underpowered on all fronts. Still has spinning rust to boot from!
M.
The fuck dude...I have 32GB of RAM installed and an RTX 2080...my current RAM usage is less than 3GB and I've got two VM's running...I also have many tabs open in Firefox and a couple of AnyDesk sessions going.
No idea what my GPU RAM usage is right now, but it's all silky smooth*
Something is horrendously wrong with your machine.
*apart from the dual screen bug with NVIDIA that forces all my displays to render at the framerate of my lowest refresh monitor...wish they'd fucking fix that, it's an ancient bug at this point.
I'd suggest looking at your task manager for errant applications, probably crypto-mining malware <cough> I mean application launchers. I run multiple applications on multiple desktops, as well as games, and my PC hasn't been restarted all over christmas. My GPU is using up 2GB of it's 4. If you're using 12 without any 3d apps, something is running you are not aware of.
This post has been deleted by its author
I've seen tests run when this was something that worried people about the 8GB RTX 3070. In testing, they could only find less than a half dozen things that needed more than 8GB, and that was only when running at 4k resolutions. Likely more due to poor programming than actual need. 12GB should be good for several more years. I plan on keeping my QHD monitor for at least a few more years and am only looking at picking up a used 8GB card. Absolutely nothing (bar computational stuff, but that can use every MB you throw at it) needs more than 8GB at QHD.
The AI won't train itself and you need it to stay competitive. Sure you may stay with 3080 or whatever, but business down the road already has 4090 and they can iterate twice as fast. The longer you wait, the higher chance you may never catch up and eventually lose the business.
It is therefore crucial in this time and age to have the best tools available, it's not even a matter of being ahead of competition, but simply being relevant.
Competitive. Hah. With current pricing trends, I doubt it.
Without a reasonable mid range card on sale PC gaming has a rocky few years to go. It's not enough for wealthy 40-somethings to be the only purveyors of such things, especially when many titles are released on console anyway.
Ironically, Intel and it's ARC cards might be able to fulfill the role if neither AMD or Nvidia behave reasonably. The competition is definitely welcome.
""The RTX 4080 12GB is a fantastic graphics card, but it's not named right. Having two GPUs with the 4080 designation is confusing. So, we're pressing the 'unlaunch' button," a news release from October read."
Not named right ?? NV has tried to confuse consumers since the GTX 1650, a shit Turing card, concealed as a Pascal card (16XX) for now a while !
They always do that now, because those shit yields have to go somewhere (ie., be bought) !
You're not going far back enough. They've been pulling the same tricks for way longer.
"Geforce 4 Ti is amazing! Wow! But I'll buy the Geforce 4 MX, it's a lot cheaper - whoops, it's really much closer to a Geforce 2 MX, unrelated architecture, missing many features". That's the first I recall, but there are many more since.
There are 7 configurations of the Geforce 1060.
One of the most egregious things they do is put very slow memory on a card and hide that information a bit - the Geforce 1030 comes in DDR4 and GDDR5 variants, with an enormous (more than double) bandwidth difference between them.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
