I just built my new PC but I'm currently using my older 3060 TI while waiting for the new Nvidia cards to drop and hopefully be more affordable. This got me thinking – why can't we configure VRAM like we configure system RAM? I'd love to hear from someone who knows the ins and outs of GPU architecture to clarify this for me.
5 Answers
It used to be possible with some very old graphics cards that had socketed memory. However, modern GPUs need extremely high speeds, and adding a socket just wouldn’t work well with today’s requirements for VRAM.
A few Intel GPUs do allow some configuration by board partners, but it’s not user-configurable. It’s about as close as we can get these days!
Performance is critical. To maximize speed, VRAM has to be soldered directly to the GPU. Any removable solution would hurt performance drastically. We could have upgradeable VRAM, but it would likely lead to massive slowdowns.
Exactly! And then there’s the added complexity of keeping things cool if we had removable modules. It just doesn’t seem worth the hassle.
That makes sense! Is there really any benefit in having upgradable VRAM then, if it won’t work any better?
Latency is a big deal here. VRAM has much higher performance needs compared to regular RAM because it processes textures and frames at lightning speed, which isn't as critical for the CPU.
That's true! In fact, VRAM typically has a latency of 200ns, while regular RAM is around 50-80ns. The distinction in their speed needs is huge.
I think we’re getting hung up on terminology here. It’s more about bandwidth for VRAM than just speed, and minimizing distance plays a huge part.
At the end of the day, it’s all about business. Companies might prefer selling more expensive models instead of allowing upgrades; it’s way more profitable for them.
Right? That's capitalism for you! They want you to keep shelling out for the latest models instead of upgrading what you have.
I agree! If they allowed us to upgrade, that would hinder their sales strategy around newer models.
The main issue is that VRAM needs to be as close to the GPU as possible for the best performance. If VRAM was on a separate board, the longer connections would degrade signal quality and add latency, which would slow things down.
Absolutely! The precision in VRAM communication must remain tight, and any extra distance can really mess with performance.
I like where you’re going with this, but let’s not act like it’s impossible – PCB trace length and signal quality are issues that can be tackled.
Exactly! Plus, making VRAM upgradable doesn't seem to align with Nvidia's business model, which probably prefers us to keep buying new cards.