Whats the most difficult to design out of the two ?

Is processor superior or a graphics card in terms of technology ?

Just curious.

They are both quite different from each other. Although they are both processing units, the demands differ a lot, and so does the functionality...

It's relatively easy to design just a GPU or CPU, but it's really hard to design a new, fast, winning model.... at least, I guess so... I never did design one of my own

I think the newer generation GPUs designed these days go hand in hand with the software support. eg: Everytime NVidia designs a new GPU, its got to do with how many DirectX/OpenGL instructions it supports.

Im wondering how the chicken-and-egg story goes. It must be something like a new GPU comes out supporting 'the most amazing dynamic lightening support' ever - then DirectX XY.Z API comes up with support for it. Or it could be vice versa.

GPU and CPU developers usually try to implement what was solved by software before. (because it was cool or even functional).

For instance , in the old days Anti Aliasing was a software trick used in programs , as was quick complex mathematics.

Today there are both implemented in the GPU's and CPU's resulting in the instruction sets known as CISC (complex instruction set computing) , RISC (reduced instruction set computing) or EPIC (Explicitly Parallel Instruction Computing).

Thing is , when using bits and bytes , you have no real scientifical and natural boundaries , but when designing , you do.

Do not take heat , dissipation , energy matter , electronics , magnetic fluxes and electromagnetic fields for granted

At the moment average consumer GPU's are far more advanced than average consumer CPU's.

Im sure GPU are more difficult to make.

I would think a good CPU would be harder to make. I mean, all the GPUs have programmable shader functions, etc. now but compared to a CPU, they're still relatively single purpose chips.

I think it's about the same... they both have their own functionalities. CPU's have all kind of accelerators like MMX and SSE; GPU's however have all kinds of DirectX functions implemented.....

There is no programming language more advanced and faster than the core assembly (or machine language) the CPU or GPU understands.

Trouble is though that you are programming one heck of a stable program on just THAT configuration. Rearrange one single thing and it will malfunction.


DirectX is not an accellerator , DirectX is a compatibility issue , just as OpenGL is. DirectX compatible just means it understands the DirectX functionality. It's a list of translated things.


For instance :

If your program would be assembly code on a NV21 GPU putting a triangle somewhere , that would be the fastest , but perhaps would only work on a NV21 GPU and not even on a NV20 or a NV22.

If you program in DirectX. (Yo DirectX ! Put a triangle there) , it will work (hopefully) the same on every DirectX compatible GPU. Wether it would be NV20 , NV21 , NV22 or Radeon 9700.
The REAL instruction given to the GPU is decided by directx (Uh , Nvidia .. i'll push this... Uh .. Radeon .. i'll push that)


Enhancements in CPU's trigger some abilities inside the CPU itself , letting go of their standard instruction set , but use a far more advanced instruction set.

For instance :

AMD has to be compatible with X86 in order to run a X86 operating system , but if the operating system would understand the 3d!Now technology , the AMD CPU is possible to calculate things a little faster than by using their "normal" X86 instruction set.

Intel does the same with MMX. You can run your X86 operating system on a 486 66Mhz or a Pentium 1 66Mhz. Given they use the same instruction set (Note : actually they don't) at the same speed , their performance would be about the same , but if the operating system would understand the MMX technology , the Intel CPU with MMX would be possible to calculate things a little faster than by using their normal X86 instruction set.


A pain to program ,Fast , but hugely incompatible : Machine code 100% on the unit
Easier to program , Less fast , but more compatible : DirectX , OpenGl , detection and usage of MMX , SSE , whatever
Easiest , Slow , but extremely compatible : Standard instruction sets every CPU and GPU understands.


Quake 2 was actually a very good testing tool for this. Regarding if your GPU understands OpenGL , DirectX or 3DFX tuning , you could check which one is the fastest for your video card