Are the 8 GB cards worth it?

[blinnbanir]

Publishers would never push to rush a product to hit launch windows, cut back on resources and man power to help optimize games. Outsource PC ports to studios who never worked on the original game or move most of the staff onto a new game while leaving a skeleton crew to finalize & patch a title about to launch. You make it sound as if these publishers care more about cost cutting and share holders than us the gamers. Next you will be pushing some crazy conspiracy theory like publishers think games should cost more.

Indie Games are so good. I am currently enjoying the depth of Tempest Rising.

 

[dirtyferret]

Kerbal Space Program 2: Conspiracy Edition

it was a perfect phone call development - love Take Two

 

[Selaya]

8gb is fine for the $200 tier. but unfortunately, in 2025 x60 is $400. and that's not fine anymore.

 

[TSiAhmat]

@lilhasselhoffer this was not in defense of 8GB being enough for the 5060ti in 1080p (as an example) but that more vram doesn't increase your frames. It limits them/it introduces stutter if it needs more than it has. Similar to system RAM, you don't notice it if you have way to much. But the second were the needs exceed the capacity you will.

I am also pretty sure that the 9070 does NOT perform as well as an 9070 xt even at 304W

(Please correct me if I am wrong tho, thanks!)

 

[AusWolf]

AMD is likely to say that then spill the beer, slip in it, and end up in the metaphor hospital given their tendency as of late to botch launches... just sayin'

What botched launches?

 

[R-T-B]

What botched launches?

I was thinking of the poor AGESA firmware alongside the Ryzen 9xxx (and the 7xxx GPU launch issues to a lesser extent) when writing that. Maybe "botched launches" is a bit aggressive terminology.

The products generally end up excellent yes but there is always a bit of teething that should just not be there at all.

 

[Chrispy_]

I voted yes because you didn't ask which 8GB cards are worth it.

• Picking up a new RX6600 is decent performance for $180-220 and it'll let you play most AAA games at low settings.
• If you already are looking at the used market, there are plenty of 8GB GPUs worth buying for up to about $200. Hell, a 3070 is maybe even an okay deal at $250 since you can make use of DLSS

If you are specifically talking about the 5060-series and 9060-series 8GB variants, then no. They're just as bad, if not worse than the 4060-series and RX 7600 in that they have the same VRAM shortage but two full years have passed and the AAA gaming landscape is even more hostile to 8GB cards than it was in 2023. They don't have enough VRAM to support the features they claim to support in several AAA games, and honestly it's only AAA games are mostly where you'll find those features supported and often needed. As time goes on, the number of AAA titles that crash and burn on 8GB GPUs is only going to grow.

IMO, 8GB is a false economy now and it wasn't exactly a great idea two years ago - Just spend the extra $50 on a 12 or 16GB card to avoid the built-in obsolescence.

 

[AusWolf]

I was thinking of the poor AGESA firmware alongside the Ryzen 9xxx (and the 7xxx GPU launch issues to a lesser extent) when writing that. Maybe "botched launches" is a bit aggressive terminology.

The products generally end up excellent yes but there is always a bit of teething that should just not be there at all.

All I can say is...

I, for example, didn't even hear about those AGESA issues with Ryzen 9000.

Sure, the first few BIOSes were a bit funny on my 7700X and then 7800X3D, and I had to wait for 3 weeks for proper Linux support on my 9070 XT, but so what? I've got a decent PC in the end.

Whether your cup is half empty or half full is your choice.

Edit: Or look at all the weekly hotfixes Nvidia pushes out for the 50-series. Or the missing ROPs. If you want to talk about a botched launch...

 

[NSR]

Am I the only guy whose running 1080p on a card above 8GB?

I've noticed games like Cyberpunk and Kingdom Come start at 8GB usage and creep to 12GB+ as I play (RX 6800 & 9070) and these aren't considered to be VRAM-hungry games. I know games optimise differently if you have less to work with but the experience for me is always smooth and I'm not worried about settings or textures. That is a very nice problem to not have.

Granted, the 6800 & 9070 are advertised for higher resolutions but the 5060 matches the 6800 in performance + has newer and better features, so why is the 5060 a 1080p card if the 6800 is not?

 

[Chrispy_]

I've noticed games like Cyberpunk and Kingdom Come start at 8GB usage and creep to 12GB+ as I play (RX 6800 & 9070) and these aren't considered to be VRAM-hungry games. I know games optimise differently if you have less to work with but the experience for me is always smooth and I'm not worried about settings or textures. That is a very nice problem to not have.

So the VRAM creep to 12GB+ is because your GPU doesn't have to throw out anything.

The game needs, say, 7.5GB to run and that 7.5GB will grow as new textures are required to the game area in question. On an 8GB GPU, it will evict the oldest textures to make sure there's available VRAM to run the game well, and load them again if needed later.

If you fired up another application that needed 8GB of GPU VRAM whilst running Cyberpunk with 12GB+ allocated, I'm not sure what would happen but my guess is that the driver would attempt to cull the extra 4-5GB of VRAM that CP2077 didn't have locked for its running process to load in the second application you'd just launched.

I deal with people all day who are running multiple CAD models and rendering suites, so rather than being a single application using as much of the GPU memory pool as possible, it's 3+ different applications all sharing the VRAM pool. Task Manager in W10/W11 will show you that there's a paging file for VRAM the same way there's a paging file for regular system RAM. When stuff gets evicted it goes to that paging area:

 

[Shihab]

We are in the era of AI, where VRAM size is the single most essential KPI for gaming technology - not GPU performance measured in metrics like TFLOPS, corecount, clockspeeds or whatnot - those are secondary KPIs.

First day at business school?

 

[ARF]

Beaten by a year older card with half the VRAM.

View attachment 396467

What does VRAM have to do with hitting a certain FPS target? In what world would the same card with 12 GB vs 16 GB have that be the limiter on whether it can hit 144 or not, if it can hit 120?


Well said.

However, "muh VRAM is the problem" posts will still win popular debate, despite evidence showing the contrary, across history.

"Future proof" high VRAM cards still lose to better core/architecture. See 16 GB Radeon VII vs the year older 8 GB 2070.

Vega 20 in the Radeon VII was the worst core / architecture. 16 GB was an overkill for it, also its 1.02 TB/s memory throughput with HBM2.

The point here is that RTX 5060 / RTX 5060 Ti with their 8 GB pools are bottlenecked by the VRAM amount, and they do need at least 12 GB.
This is a mistake by the engineers and management at Nvidia.


Have you seen that the Radeon 890M integrated graphics (in Ryzen AI 9 HX 300 series) is already configured with 8 GB pools?

 

[NSR]

So the VRAM creep to 12GB+ is because your GPU doesn't have to throw out anything.

The game needs, say, 7.5GB to run and that 7.5GB will grow as new textures are required to the game area in question. On an 8GB GPU, it will evict the oldest textures to make sure there's available VRAM to run the game well, and load them again if needed later.

On this in particular, is this why I can drive to the other side of the map and still see a car I left behind when I return like 30 minutes later? I've always assumed things like this always got cleared but have been amused by evidence of my previous endeavours still left behind.

 

[Shihab]

On this in particular, is this why I can drive to the other side of the map and still see a car I left behind when I return like 30 minutes later? I've always assumed things like this always got cleared but have been amused by evidence of my previous endeavours still left behind.

No. Objects (and they are literally called objects in programmer speak) are handled by ye old CPU and stored in the system memory. "Object" here can be as simple as description (car), a state (e.g. position in space, being driven, etc), and a 3d mode plus frills to pass to the GPU to draw.

In all typical cases, an object far from the camera would have its mesh and textures evicted from the VRAM. The instance may be retained in system memory (although not necessarily. Could always be saved to disk and loaded much like a typical game save), as the said state isn't actually that large. The overly simplified example I gave above could fit in less than a hundred bytes (but actual use cases are definitely more than that).

 

[Raffles]

This reminds me of DPI class cars in IMSA. Caddy had a V8 and Mazda had a 4 cylinder. Though it did not win much it showed the performance gap. 8 GB is the 6600 or RX580. Those cards are fine for 1080P but not anything higher.

Yet my 8gb 3070 can play Forza at 4k120

I want to know who keeps employing these aparently incompetent programers... I mean it's a highly technical field and they are hiring village idiots according to *checks notes* gamers.

I totally buy this plot. /s

The issue isn't incompetent devs, its irrational timelines pushed by higher ups and devs thus being forced to quickly produce a product. Even the most skilled devs dont really have time to optimise properly in that context. Back in the day, the same devs likely would have made their own engine. No time for that now: you pick whatever prepackaged engine, load it all with addons to fit your needs, code in some inefficient high level scripting language, and call it a "game."

The problem is way higher up then the devs, most of the time.


That's because a bunch of the transistors aren't working / fused off on the 9070. That doesn't really count.

Well said. I do get sick of the "lazy devs" comment that gets perpetually parroted. If these people had even the faintest clue how difficult it is, and how much hard work it is to make a decent size/looking/playing game from scratch, they'd feel extremely stupid for bandying around that saying.

It's both.

In almost all cases it's not both - it's mainly management, time constraints, ambition of project etc.

It might not be the 90's anymore where people are having to make their own 3D engines and write code in assembly, but by and large software engineers on well known games are still highly skilled and hardworking, the vast majority of the time.

 

[Event Horizon]

Well said. I do get sick of the "lazy devs" comment that gets perpetually parroted. If these people had even the faintest clue how difficult it is, and how much hard work it is to make a decent size/looking/playing game from scratch, they'd feel extremely stupid for bandying around that saying.

The "lazy devs" people have no software development or project management experience. A 10 second glance at a real project's issue tracker would send them into cardiac arrest.

 

[NSR]

No. Objects (and they are literally called objects in programmer speak) are handled by ye old CPU and stored in the system memory. "Object" here can be as simple as description (car), a state (e.g. position in space, being driven, etc), and a 3d mode plus frills to pass to the GPU to draw.

In all typical cases, an object far from the camera would have its mesh and textures evicted from the VRAM. The instance may be retained in system memory (although not necessarily. Could always be saved to disk and loaded much like a typical game save), as the said state isn't actually that large. The overly simplified example I gave above could fit in less than a hundred bytes (but actual use cases are definitely more than that).

Interesting, thanks for the info.

 

[KikeUY]

I am playing on an 8gb gpu and I would pay for an upgrade even if its 8gb but not 400U$S+...
Those prices are insane(for me , of course).

 

[Shihab]

Oh lazy devs are a plenty, they just call themselves "vibe coders" these days.

Fortunately for us tho, non-mobile gamedev has a relatively crappy reward/effort ratio, so you mainly see those people doing webdev and adjacent stuff. </partial flamebait>

 

[lilhasselhoffer]

@lilhasselhoffer this was not in defense of 8GB being enough for the 5060ti in 1080p (as an example) but that more vram doesn't increase your frames. It limits them/it introduces stutter if it needs more than it has. Similar to system RAM, you don't notice it if you have way to much. But the second were the needs exceed the capacity you will.

I am also pretty sure that the 9070 does NOT perform as well as an 9070 xt even at 304W

(Please correct me if I am wrong tho, thanks!)

You seem to not read much on TPU. TPU article on unlocking the 9070

Of note:
"...More importantly, the flashed RX 9070 was found to offer significantly improved overclocking headroom, and PCGH was able to tune up performance by 15-20% over stock RX 9070, bringing its performance to match a stock RX 9070 XT, which was confirmed by a set of 3DMark benchmarks."



Regarding the other bit, yeah. I think we're on the same page about it being a limiter, but not a promoter of FPS. That said, this has been the Nvidia line in the sand for about a decade now. September 2014 was when the 970 launched. Despite this, they've always been proven to be full of crap when their cards eventually become VRAM choked. I'm pretty OK then that this was largely a circular logic loop about VRAM, so thumbs up on comprehension from my side.

The "lazy devs" people have no software development or project management experience. A 10 second glance at a real project's issue tracker would send them into cardiac arrest.

There are no issues. There is only opportunity and the targets of the sprint.


And the bile raises in my throat saying that, with the desire to pimp-slap myself for saying it.

 

[TSiAhmat]

You seem to not read much on TPU. TPU article on unlocking the 9070

Of note:
"...More importantly, the flashed RX 9070 was found to offer significantly improved overclocking headroom, and PCGH was able to tune up performance by 15-20% over stock RX 9070, bringing its performance to match a stock RX 9070 XT, which was confirmed by a set of 3DMark benchmarks."



Regarding the other bit, yeah. I think we're on the same page about it being a limiter, but not a promoter of FPS. That said, this has been the Nvidia line in the sand for about a decade now. September 2014 was when the 970 launched. Despite this, they've always been proven to be full of crap when their cards eventually become VRAM choked. I'm pretty OK then that this was largely a circular logic loop about VRAM, so thumbs up on comprehension from my side.



There are no issues. There is only opportunity and the targets of the sprint.


And the bile raises in my throat saying that, with the desire to pimp-slap myself for saying it.

no I've read it, but I didn't remember it giving such a huge boost in performance. Thanks for the info. I assume undervolting a 9070xt & overclocking a 9070xt still provides better Performance (in either higher fps or in better W/fps [not stock]) but still pretty cool what a little extra power can do!

 

[blinnbanir]

Yet my 8gb 3070 can play Forza at 4k120

Racing Games are some of the easy to run.

This is at 4K

 

[yfn_ratchet]

Well said. I do get sick of the "lazy devs" comment that gets perpetually parroted. If these people had even the faintest clue how difficult it is, and how much hard work it is to make a decent size/looking/playing game from scratch, they'd feel extremely stupid for bandying around that saying.

In almost all cases it's not both - it's mainly management, time constraints, ambition of project etc.

It might not be the 90's anymore where people are having to make their own 3D engines and write code in assembly, but by and large software engineers on well known games are still highly skilled and hardworking, the vast majority of the time.

I concur. Mind you, the history of gaming has always been a trendline where rudimentary techniques are refined, optimized, and then packaged into a generalized form that is then tucked into the trunk of a newfangled dev tool that makes the whole process faster and more convenient, though at the cost of computational efficiency. That has been true since computers came with programmable memory. People used to raise a stink over writing in C and compiling it as opposed to writing in ASM, said it was slow and bad. When was the last time you wrote in ASM?

Computers are more powerful, and that is partially meant to compensate for the compounding inefficiencies we have accepted in order to make the development process easier on the fleshy humans that have to do the work. Unfortunately, this idea has been leveraged to crunch devs further and further, not necessarily by shortening dev times, lord no, games used to go from pre-production to gold in 24-36 months, but by hyperbolically increasing the scope and detail of the project. Businesspeople saw that we filled pools with water trucks and big hoses instead of straight off the municipal tap from a garden hose, walked up to the man in the truck, and demanded he fill in Lake Michigan.

 

[R-T-B]

I, for example, didn't even hear about those AGESA issues with Ryzen 9000.

I owned one during that period. You read the reviews didn't you? Latency issues had it having performance regressions in some cases vs the 7xxx. Nothing major though in the long or even medium term (AGESA fixes were out in literally just a week or two), was more just poking fun then having any serious complaints (my 9950x has actually been a delight). You are right that attitude determines your experience!

Edit: Or look at all the weekly hotfixes Nvidia pushes out for the 50-series. Or the missing ROPs. If you want to talk about a botched launch...

lol that honestly is an excellent point.

 

[Raffles]

I concur. Mind you, the history of gaming has always been a trendline where rudimentary techniques are refined, optimized, and then packaged into a generalized form that is then tucked into the trunk of a newfangled dev tool that makes the whole process faster and more convenient, though at the cost of computational efficiency. That has been true since computers came with programmable memory. People used to raise a stink over writing in C and compiling it as opposed to writing in ASM, said it was slow and bad. When was the last time you wrote in ASM?

Computers are more powerful, and that is partially meant to compensate for the compounding inefficiencies we have accepted in order to make the development process easier on the fleshy humans that have to do the work. Unfortunately, this idea has been leveraged to crunch devs further and further, not necessarily by shortening dev times, lord no, games used to go from pre-production to gold in 24-36 months, but by hyperbolically increasing the scope and detail of the project. Businesspeople saw that we filled pools with water trucks and big hoses instead of straight off the municipal tap from a garden hose, walked up to the man in the truck, and demanded he fill in Lake Michigan.

Indeed, pretty much everything you said is true.

And I do think the techniques being repurposed into higher level code and engines, designed for powerful modern machines, has somewhat dilluted the skillset of yesteryear's software engineers - whereby optimisation was a necessary artform, in order to get something that looks like GT1 running on a machine with a 33mhz processor and 3mb total RAM, for example. It still doesn't make the "lazy devs" comment any less stupid, ignorant, and insulting though.

To answer your question, a good few years ago I did actually have to write shaders in assembly. It wasn't fun, but somehow I preferred it to writing shaders in modern high level shader code

The worst are those horrible nodes you drag around, "visual scripting". I think I'd rather use assembly