Why is it that GPU's have progressed so dramatically over the years while CPU's while progressing haven't really improved very dramatically?
Why is it that GPU's have progressed so dramatically over the years while CPU's while progressing haven't really...
we cant make things faster anymore, we can only give them more cores
First learn to English
Second, are you retarded?
yes we can
7nm is being rolled out
what kind of improvement is that? what ghz?
Improvement is measured in TFLOPS not Ghz
Graphics and therefore GPUs are trivial to parallelize. You can take the same design, throw more transistors at it (make moar coars!) and wind up with something faster in real-world usage, roughly proportional to however much you added. The only limits are the amount of power you can draw and the amount of heat you can dissipate, and that's what shrinking the process improves.
CPUs, on the other hand, are expected to do things that aren't necessarily very parallel at all.
tflops doesn't measure the speed of a single processor, so tell me what ghz we can expect to see from this new technology
Why are GPUs giant plugins for computers, but CPUs are still tiny squares? Wouldn't we get more performance out of CPUs if we just increased the size? Is there any real reason we keep it in such a relatively small size when we actually care about it?
dra åt helvete vitun hurri
Same reason Japs prefer lolis. Less resistance in a smaller package
He's asking why GPUs are lagging behind. GPU performance is measured in TFLOPS
Performance on CPUs aren't determined solely on clock speeds. If you want to know how a CPU performs you have to test it on different scenarios. There isn't a single value to determine a processor performance.
Thank Intel and their 1℅ IPC gains
du kan gå först
You do realize that GPU dies are same size as CPU ones, right ?
this.
But to elaborate:
The traditional way of making transistors smaller has hit a brick wall.
To make transistors even smaller they needed a radically new way of fabricating called euv, but euv is an engineering nightmare and has been delayed again and again for over 10 years now.
In the end it all comes down to power consumption and thermal dynamics
We could have massive GPUs with gains equivalent to that of CPUs from one gen to the next, but if companies did that we would need a small reactor in our houses to power those cards and professional grade cooling to keep them from literally melting down
Take pic related for example, it easily reached 95°C and 300W at load
X86 tax basically. Up until recently GPUs had to rely on people buying them to play video games, there's only so much people will pay for that. With CPUs they're driven by server markets, companies buying server CPUs will pay almost anything so they drip feed them models with slightly higher core counts far below technical limitations while including artificial limitations you have to pay 50% to 100% more to access like being allowed to use 2 of them in one computer.
If x86 CPUs were a commodity product not controlled by a duopoly the top 28 core intel would be maybe $1200, there would be no separate 1p/2p versions or different prices based on clock speeds.
this man is still living in the 2000s
this man is living in the world of marketing lies and technical illiteracy
floating point operations are not a meaningful measure of single core performance
Absolutely based Japs
Going wide is easier than going up.
those were the days