Why are GPUs so big, while CPUs are so small?
In: Engineering
You gotta make the analogy as motherboards vs GPU because that gpu box is the motherboard that the actually gpu chip runs on.
Honestly it comes down to marketing. Consumer gpus have massively overbuilt coolers that can be used to differentiate brands when the core of the product, the silicon die, is nearly identical. Brands will also try to overbuild the cooler such that they can use it up and down the product stack. It is cheaper, in most cases, to reuse the same cooler on multiple cards rather than have multiple assembly lines. Professional gpus use the same dies at similar TDPs, something like the 4090 vs the ada RTX 5000, use dual slot coolers while being identical in performance.
EDIT: Thought you were more talking about the coolers than the card itself. As for the form factor, graphics cards also have memory and power management circuitry that is usually built onto the the motherboard. That is why you see them more as add in cards. There are form factors in more specialized systems like SXM form factors. They are still bigger than say a server cpu but not by that much.
The actual processing chips of GPUs are fairly similar to CPUs. Graphics cards are essentially self contained specialised computers meant to be good at some very specific tasks. As video games, rendering software, and other high intensity processing needs grow bigger, so do GPUs. They’re kind of like video game consoles, again purpose built computers meant to serve a very specific function, though consoles are proprietary and have more unique architecture.
Wish I could answer this… to think I’d be this computer illiterate at 33 when I was messing with motherboards at age 10 and setting up one of the first WiFi networks in my neighborhood with wireless printing. It was fancy back then and I didn’t even have to use a software installer. Of course those were the days of tricking my parents into getting me high speed so I can watch unlimited porn. 😂
When comparing them by mass and/or volume, remember that when you picture a typical GPU that you buy as a consumer (so not just the graphics chip in isolation), that includes huge chucks of heatsinks and fans. Depending on which manufacturer GPU you have, you might be able to directly see many rows of metal which are for dissipating heat.
So when you are comparing GPU to CPU size, you should also envision the CPU as including something [like this](https://m.media-amazon.com/images/I/81oNtbH-3lL.jpg) or [like this](https://m.media-amazon.com/images/I/81o2dHjNItL.jpg) for a fair comparison, instead of just the chip.
> Why are GPUs so big, while CPUs are so small?
Genetics.
A CPU and GPU are just chips, and they’re about the same size.
A GPU gets soldered onto a circuit board along with video RAM chips, and a big cooler is mounted to it. This entire assembly is a video card, and it’s quite big.
A CPU gets mounted onto a motherboard, gets a big cooler attached on top of it, and RAM sticks are also installed into the motherboard. This entire assembly is comparable to a video card, and is in fact typically quite a bit larger.
This depends a lot on what you’re actually asking here.
**If your question “why is the chip that is a CPU smaller than the chip that is a GPU”?**
CPU dies are smaller than GPU dies because the work that GPUs do can be better facilitated by shoving more transistors into the space it occupies. CPUs have to handle more specialized work that requires more variation which means they don’t have the opportunity to have a ton of transistors shoved into them and take up a bunch of space.
The CPU is an office building full of engineers and technicians building and designing things.
The GPU is a factory full of assembly lines and operators.
The GPU benefits from shoving more assembly lines and operators into it so that you can feed it more inputs to get more outputs. It doesn’t need engineers and technicians, it needs operators that can make more “things”. The CPU doesn’t strictly benefit from shoving more technicians and engineers into trying to solve the same problem. The CPU would rather have *better* engineers and technicians instead of more of them. (Amusingly, this is what caused the Pentium 4 to have such problems. The design theory of the Pentium processors had hit a physical wall on how fast they could make a processor do processor things. But they NEEDED to make a faster processor…so just shove more “technicians and engineers” into it! Which resulted in a processor that ran insanely hot for little gain.)
**If your question is “why is my graphics card so big while my processor is so small”?**
The graphics card, the whole thing you plug into your motherboard, is better compared against the motherboard+CPU as an entity. Because that’s basically what a graphics card is. It’s a smaller motherboard with its own processor, memory, etc., that plugs into the main motherboard that also has its own processor, memory, etc. Your processor is like the *actual GPU* that is on your graphics card.
Because the dies used in GPUs are actually much closer to the kind of top end dies used in datacenter. The 3090 uses a 628 sqmm die while the corresponding Tesla A100 only uses a 628 sqmm die.
For comparison a 14900k uses a 257 sqmm die while a Xeon 8570 uses two 763 sqmm dies.
The reason that is is because gpu workloads get much more parallel than cpu workloads where there is often only so much you can do in parallel.
That said: The gap is really not that large. Ultimately it is the same silicon base.
Unless you mean the full product: Then the answer is simply that the whole graphics card has many more components including the gpu die than a cpu.
It’s probably already been said, but the entire GPU assembly has its own “motherboard” (PCB)RAM and cooling system. They’re both chips that are attached to a heatsink/cooling apparatus, but using this logic, the CPU system is much larger than a GPU system if you consider the case, RAM, PSU etc.
Latest Answers