Love projects like this, experimenting and bringing new life to existing hardware. Also gives new context to current debates.

what is the point of this "experiment"?

GlassGR:

what is the point of this "experiment"?

Well to reiterate that those mid level skus are unnecessary gimped down to fill the market segments

asturur:

Well to reiterate that those mid level skus are unnecessary gimped down to fill the market segments

Greed is the most played game worldwide. That includes GPUs being castrated...because of reasons.

GlassGR:

what is the point of this "experiment"?

To enjoy doing experiment... A bit like when you open a ford hood and you think you can fit an audi bi turbo engine instead of the 1,6 pinto... Lot of fun indeed.

anticupidon:

Greed is the most played game worldwide. That includes GPUs being castrated...because of reasons.

In it's time it wasn't castrated, texture were in lower res, most gamer were using HD only... it was a good card. Now it's not the same problem, with some game we reach the 8g limit in HD, the tech have evolved... But the raw power of old GPU are still enough and they are now limited by choise that were logical few ago (with some exeption: RX 6500 or 6400... lol).

It seems modders having fun with giving gpus more vram 😀

Why not? We are all tinkers to some degree!

GlassGR:

what is the point of this "experiment"?

Because as humans, we can, its why we arnt still sniffing each others arses in the Rift Valley of Ethiopia.

rl66:

In it's time it wasn't castrated, texture were in lower res, most gamer were using HD only... it was a good card. Now it's not the same problem, with some game we reach the 8g limit in HD, the tech have evolved... But the raw power of old GPU are still enough and they are now limited by choise that were logical few ago (with some exeption: RX 6500 or 6400... lol).

The tech has evolved, memory $ize not $o much. Curios if the Radeon RX 7600 XT will have 8 GB too, per recent leak Gigabyte Radeon RX 7600 8GB Gaming. Not to forget nvidia's cards memory size also.

danny69t:

The tech has evolved, memory $ize not $o much. Curios if the Radeon RX 7600 XT will have 8 GB too, per recent leak Gigabyte Radeon RX 7600 8GB Gaming. Not to forget nvidia's cards memory size also.

They are 1080p cards so 120fps with Med settings on unchallenging shooters using a bit of fake framery. Ideal for the masses who think they are getting a bargain compared to the 4070 series.

Wow, a card that has low VRAM can do better on VRAM constrained scenarios when it gets more VRAM, who knew? Next up is swapping the card chip with a faster one and getting surprised that it puts out more FPS 🙄 There's always a bottleneck somewhere, this is unavoidable. Still, cool experiment for what it is: custom hand-made mod in a very complex hardware.

Ricardo:

Wow, a card that has low VRAM can do better on VRAM constrained scenarios when it gets more VRAM, who knew? Next up is swapping the card chip with a faster one and getting surprised that it puts out more FPS 🙄 There's always a bottleneck somewhere, this is unavoidable. Still, cool experiment for what it is: custom hand-made mod in a very complex hardware.

The point is, adding another 6GB of VRAM isn't that much more expensive, it's a simple enough upgrade that you could do at home if you have the skills, and, there's an upper limit to how much VRAM you actually need. This is something AIB partners could do relatively effortlessly*. While I think 12GB is overkill for a 5600XT, that pretty much guarantees it will not be bottlenecked by memory. Meanwhile, you can upgrade the core but stick with 6GB and you won't see a performance uplift, despite that being a major expense and perhaps something you couldn't do at home. * I assume there must be some sort of agreement between AIB partners, because this would be such an easy way for them to rake in profits. It's a similar thing to mobo partners who found ways to override the way boost clocks work on CPUs/chipsets that don't allow overclocking, so you can squeeze more performance out of lower-end parts. Or, back in the days when you could unlock cores in the BIOS. If AMD and Nvidia want to move more stock, I think an easy way for them to accomplish this would be to have higher VRAM models. Of course, that means future generations would also need more VRAM, which cuts into their all-too-important profit margins.

This shows equipping RTX xx50 and RTX xx60 cards with adequate memory (12GB or 16GB) will greatly affect the sales of RTX xx70 and above cards.

We would all imagine the results, but it's still interesting to see them anyway.

They should make GFX cards modular. Like a PC. Free slots for VRAM, so u can upgrade.

schmidtbag:

The point is, adding another 6GB of VRAM isn't that much more expensive, it's a simple enough upgrade that you could do at home if you have the skills, and, there's an upper limit to how much VRAM you actually need. This is something AIB partners could do relatively effortlessly*. While I think 12GB is overkill for a 5600XT, that pretty much guarantees it will not be bottlenecked by memory. Meanwhile, you can upgrade the core but stick with 6GB and you won't see a performance uplift, despite that being a major expense and perhaps something you couldn't do at home. * I assume there must be some sort of agreement between AIB partners, because this would be such an easy way for them to rake in profits. It's a similar thing to mobo partners who found ways to override the way boost clocks work on CPUs/chipsets that don't allow overclocking, so you can squeeze more performance out of lower-end parts. Or, back in the days when you could unlock cores in the BIOS. If AMD and Nvidia want to move more stock, I think an easy way for them to accomplish this would be to have higher VRAM models. Of course, that means future generations would also need more VRAM, which cuts into their all-too-important profit margins.

You would absolutely see a performance increase by changing the chip with the same amount of VRAM, otherwise a rtx 3060ti would have the same performance of a 3070, which is totally not true. VRAM is mainly used for keeping assets close to the gpu chip, so that those won't need to be streamed through the PCI lanes (which takes forever in computational time). More VRAM means less trips like that, but you can also compress the data and decompress on the fly, which severely improves the memory footprint while avoiding (some of) those trips. Many games these days simply don't apply optimization techniques because there's plenty of VRAM and thus it's less work to be wasteful. Also, they eagerly occupy that space when possible, even if that data isn't gonna be used, which gives the impression of high usage, but it's actually doing nothing. So, considering that the xbonex and the ps5 both have 16GB total memory (~13GB available for the game), it is, for example, absolutely reasonable that mid range cards are coming with 8~12GB of VRAM, because it means that potentially the WHOLE game runtime assets could fit in them using console settings and would need no regular system RAM, ergo: a complete waste of resources. You could argue that more VRAM would equal better graphics, but I think it's more likely that increasing VRAM would equal less optimization, because dev time and money. And video cards would be (even) more expensive.

Ricardo:

You would absolutely see a performance increase by changing the chip with the same amount of VRAM, otherwise a rtx 3060ti would have the same performance of a 3070, which is totally not true.

In cases where you aren't severely bottlenecked by VRAM then yes. But a 3070 has a lot of untapped potential.

VRAM is mainly used for keeping assets close to the gpu chip, so that those won't need to be streamed through the PCI lanes (which takes forever in computational time). More VRAM means less trips like that, but you can also compress the data and decompress on the fly, which severely improves the memory footprint while avoiding (some of) those trips. Many games these days simply don't apply optimization techniques because there's plenty of VRAM and thus it's less work to be wasteful. Also, they eagerly occupy that space when possible, even if that data isn't gonna be used, which gives the impression of high usage, but it's actually doing nothing.

On a different forum I've been in an argument with someone about this (with me saying the same thing as you) and yet I was told I was clueless.

So, considering that the xbonex and the ps5 both have 16GB total memory (~13GB available for the game), it is, for example, absolutely reasonable that mid range cards are coming with 8~12GB of VRAM, because it means that potentially the WHOLE game runtime assets could fit in them using console settings and would need no regular system RAM, ergo: a complete waste of resources.

They're not entirely comparable because not only is that memory shared between the CPU and GPU but those systems load assets in different ways, which greatly decreases the total memory needed. Until DS becomes standardized, we're going to be in a situation where VRAM is fed through the PCIe bus, to the CPU/DRAM, back to the PCIe bus, to storage.

You could argue that more VRAM would equal better graphics, but I think it's more likely that increasing VRAM would equal less optimization, because dev time and money. And video cards would be (even) more expensive.

I strongly agree with this. That's why I'm actually not too opposed to the idea of Nvidia and AMD being slow to release next-gen hardware, because that means game devs are going to be forced to focus on optimization.

schmidtbag:

In cases where you aren't severely bottlenecked by VRAM then yes. But a 3070 has a lot of untapped potential.

I mean, that's exactly my point: there's always some bottleneck. You can just use "medium" quality textures with a 3070 and run the game at ~30% more FPS than a 3060ti, so there's definitively value there. Would you rather buy a 3060 with 12GB and run at ~60% less FPS just because you can use slightly higher resolution textures? I know VRAM isn't only about textures, but that's certainly the majority of what games do with it, and a faster chip is much more important because it can do many other things that are more valuable in a game before hitting the VRAM bottleneck.

schmidtbag:

On a different forum I've been in an argument with someone about this (with me saying the same thing as you) and yet I was told I was clueless.

I'd love to know more.

schmidtbag:

They're not entirely comparable because not only is that memory shared between the CPU and GPU but those systems load assets in different ways, which greatly decreases the total memory needed. Until DS becomes standardized, we're going to be in a situation where VRAM is fed through the PCIe bus, to the CPU/DRAM, back to the PCIe bus, to storage.

Yes, I know. But PCs typically have double (triple?) system RAM to go along with the VRAM, so there's that. You might end up with duplicate data here and there, but overall there's plenty of space to be efficient. Developers expecting VRAM to equal the amount of total memory in a console is just lazy and nuts.

HybOj:

They should make GFX cards modular. Like a PC. Free slots for VRAM, so u can upgrade.

It would solve a lot of problems.