HybOj:

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

Chrysalis:

It would solve a lot of problems.

I would love to see this, where you could also swap VRMs and heatsinks. Normally I'd argue the big problem about this is how modular designs like that take up a lot of space, but... now that triple slot designs are starting to become the norm, I don't think that's a valid reason anymore.

^ They would sell less GPUs. Not going to happen, but would be nice.

Embra:

^ They would sell less GPUs. Not going to happen, but would be nice.

They would sell less overall but they would sell more product more often, they could upcharge for individual parts, and they would probably pay a lot less in total engineering costs. While I'm a strong proponent of open standards, nothing is preventing AIB partners from making their own proprietary modular system that only they get to use, keeping people within their ecosystem. Think of it like this: Let's say hypothetically you bought your 6950XT for $1100. If you did a custom-built version configured exactly as you have it now, that might have cost $1320. Now, let's say you wanted to upgrade to the 8900 XT whenever that gets released. If you're on a budget, it would be hard enough to justify spending $1100 again, but realistically, the price is probably just going to get higher. However, if you went with a modular GPU where you could save your VRM, VRAM, and HSF, it might only cost you half as much, which perhaps you could afford/justify. In this situation, not only would Powercolor not see you buy the 8900 XT (at least not at MSRP) but with the modular approach, they might actually see close to the same amount of profit simply due to the markup of individual parts and reduced engineering costs. "What if Powercolor doesn't use a proprietary design and I went with a competitor for the 8900 XT?" you might ask: well, with a pre-built model, you weren't going to be a returning [MSRP] customer anyway, so ultimately I don't think that changes anything. As far as I understand, the main reason we don't have modular GPUs has largely been due to tradition. Up until a little over a decade ago, GPUs rarely exceeded 150W, most people used the same handful of screen resolutions, and most software had roughly the same overall memory requirements. GPU compute tasks were still in their infancy. So, you could have 1GB and that would probably be enough in most cases. However, we're now living in a time when we range from low-VRAM office workloads to unoptimized games trying be played in 8K while doing GPU-encoded streaming in the background. We have GPUs ranging from 15W to 450W. We have GPUs that may lean on PCIe to keep their buffer filled. GPUs are slowly replacing a lot of tasks that used to be limited to CPUs. We have to factor in the load from AI and RT too. There's just too much variation in how we use GPUs these days, and sometimes we're forced to buy a product that is a next step up from the one we need simply so we get more VRAM, but perhaps that step up is too physically large, too power hungry, or too expensive.

Hmm looking at that. Well, all's I can say is its nice to have a bit of luck. 🙂

HybOj:

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

This is one of those ideas that look amazing on theory but that are awful in reality. This would add more complexity to an already extremely complex hardware part and would make cards even more expensive because AIBs would have to design a card to support different VRAM modules and ensure that they would all work. Then most people wouldn`t change VRAM anyway because taking off a modern cooler and reassemble it correctly is not that easy nowadays. And i don`t want to think about all the warranty issues that this would create for companies... More importantly, adding more VRAM is useless for 90% of the cases because they would run out of performance anyway. Conclusion, not everything has to be modular.

schmidtbag:

They would sell less overall but they would sell more product more often, they could upcharge for individual parts, and they would probably pay a lot less in total engineering costs. While I'm a strong proponent of open standards, nothing is preventing AIB partners from making their own proprietary modular system that only they get to use, keeping people within their ecosystem. Think of it like this: Let's say hypothetically you bought your 6950XT for $1100. If you did a custom-built version configured exactly as you have it now, that might have cost $1320. Now, let's say you wanted to upgrade to the 8900 XT whenever that gets released. If you're on a budget, it would be hard enough to justify spending $1100 again, but realistically, the price is probably just going to get higher. However, if you went with a modular GPU where you could save your VRM, VRAM, and HSF, it might only cost you half as much, which perhaps you could afford/justify. In this situation, not only would Powercolor not see you buy the 8900 XT (at least not at MSRP) but with the modular approach, they might actually see close to the same amount of profit simply due to the markup of individual parts and reduced engineering costs. "What if Powercolor doesn't use a proprietary design and I went with a competitor for the 8900 XT?" you might ask: well, with a pre-built model, you weren't going to be a returning [MSRP] customer anyway, so ultimately I don't think that changes anything. As far as I understand, the main reason we don't have modular GPUs has largely been due to tradition. Up until a little over a decade ago, GPUs rarely exceeded 150W, most people used the same handful of screen resolutions, and most software had roughly the same overall memory requirements. GPU compute tasks were still in their infancy. So, you could have 1GB and that would probably be enough in most cases. However, we're now living in a time when we range from low-VRAM office workloads to unoptimized games trying be played in 8K while doing GPU-encoded streaming in the background. We have GPUs ranging from 15W to 450W. We have GPUs that may lean on PCIe to keep their buffer filled. GPUs are slowly replacing a lot of tasks that used to be limited to CPUs. We have to factor in the load from AI and RT too. There's just too much variation in how we use GPUs these days, and sometimes we're forced to buy a product that is a next step up from the one we need simply so we get more VRAM, but perhaps that step up is too physically large, too power hungry, or too expensive.

Turn's a gpu into a micro pc, which probably isn't to most people's interest (despite the good forward going argument) if initial cost increases. Modular phones had a similar idea, and tbh I love the thought of it, but have yet to see more than 1 be available domestically (LG G5), and when it was, none of the modular components were available to purchase.

HybOj:

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

There was a time were it was possible on Hercule GPU...

rl66:

There was a time were it was possible on Hercule GPU...

In 80'-90' we had a lot of VGA cards with upgradable memory options.

HybOj:

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

I would totally be in for this ! Unfortunately not going to happen things move towards more integration for a) maximizing profits and b) minimizing latency . But I would totally be down for a GPU that you buy the memory and GPU you prefer to stick on it !!!