• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

The Nvidia Hopper thread

4K8KW10 said:
Except that Tesla, Fermi, Kepler, Maxwell, Pascal, Turing and Ampere have nothing to do with computers. Actually, the computer was invented much later after them.
Click to expand...

Sorry to nitpick, but Maxwell is literally THE guy for electrodynamics, and his equations are the reason we have high speed electronics. I can't overstate how big an influence he has had on the world, think Einstein level.
Maxwell's work was based on Amperes, and again, Ampere had an ASTONISHING contribution to electronics as we know them today.

Turing is obvious, being heavily involved in computing and early architectures, algorithms, etc.

Pascal was into mechanical computing, which inspired early electronic computers.

So yeah, apart from Tesla, Fermi and Kepler, they're astonishingly relevant to modern computers.
 
0h_no_not_again said:
So yeah, apart from Tesla, Fermi and Kepler, they're astonishingly relevant to modern computers.
Click to expand...

Fermi also designed/advanced computer systems with some of his work to create hardware for his experiments. I don't know much about Kepler. Tesla gets a pass anyhow.
 
Grim5 said:
Multiple dies on a single card. Let's hope just like for CPU's, we get an explosion in core counts. Let's see 20,000 Cuda cores on RTX4000
Click to expand...

Yeah because of this I actually expect Hopper may bring about massive performance gains for the exact same reason chiplets allow AMD to produce high core count CPUs more efficiently. And extra core count is way more important in a GPU than a CPU so could make a significantly bugger impact than it's done for Zen. but presumably there are other technical challenges involved.
 
opethdisciple said:
This thread could be 2 years old before we get any proper credible info.
Click to expand...

This is definitely a lesson I have learnt with Ampere launch. The kind of news I was hoping for only started to come out about a week before 01/09 and to be honest it was becoming a bit unhealthy. Will try to avoid tech news/leaks until Autumn 2022. I wish I was at that stage where I was finding out about new hardware AFTER it launched.
 
Seeing the 10k plus cudacores on these ampere card is fun after seeing the initial statements about anything over 4000 is diminishing returns at the start of this thread!
 
Interesting take on this that one approach might be (I think this is just experimental and might not be a thing in a working product) the opposite to what some people think instead of multiple chiplets appearing as one logical GPU* software wise it might actually create multiple virtual GPUs with the hardware able to dynamically allocate resources as needed from the pool and then the rendering farmed out to say a dozen virtual GPUs software (OS/driver) wise (transparent to the application).

EDIT: End of the day though the easiest and most efficient way for it to work is to get application developers to program for multi-adapter.


* Which without some significant (and unlikely) breakthroughs is unlikely to ever be employed for a gaming solution though can work for general purpose compute.
 
Rroff said:
Interesting take on this that one approach might be (I think this is just experimental and might not be a thing in a working product) the opposite to what some people think instead of multiple chiplets appearing as one logical GPU* software wise it might actually create multiple virtual GPUs with the hardware able to dynamically allocate resources as needed from the pool and then the rendering farmed out to say a dozen virtual GPUs software (OS/driver) wise (transparent to the application).

EDIT: End of the day though the easiest and most efficient way for it to work is to get application developers to program for multi-adapter.


* Which without some significant (and unlikely) breakthroughs is unlikely to ever be employed for a gaming solution though can work for general purpose compute.
Click to expand...
They need to make something that is no different to program for. Something that is still seen as one GPU for it to workout no?

I am looking forward to a multiple chiplet design GPU, hopefully both AMD and Nvidia come to market with it at a similar time as it will hopefully improve price for performance a lot! :D
 
You must log in or register to reply here.
Back
Top Bottom