Originally posted by pal666
View Post
Announcement
Collapse
No announcement yet.
10-Way AMD GPU Comparison For Team Fortress 2 With RadeonSI Mesa 13.1-dev
Collapse
X
-
Originally posted by pal666 View Postwhich cpu bus is faster when videocard is overclocked?
The CPU might be bound in different states, waiting on RAM I/O, waiting on disk I/O, waiting on GPU I/O - A speed up in any of these will see a small bump in performance.
Comment
-
Originally posted by boxie View PostI think you are focussing on the wrong thing here.
Originally posted by boxie View Postif the video card busses are faster, they get data to the overclocked gpu faster, meaning they complete their work faster, getting back to the CPU (which is waiting on I/O from the GPU) faster.Last edited by pal666; 22 December 2016, 06:28 AM.
Comment
-
Originally posted by pal666 View Postsame thing here. you seem to not understand, what 'cpu-bound' means. it means that fps does not rise with faster videocard
Originally posted by pal666 View Postyour theory contradicts reality where several times faster gpu model did not increase fps. and busses which pass data between gpu and cpu are called pcie bus and its speed does not depend on overclockedness of gpu
Disk->RAM
RAM->Cache
Cache->CPU
RAM->PCIE->Video RAM
Video RAM -> GPU
also "Faster" here is a tad ambiguous. The higher end graphics cards are "wider" in their ability to execute, this makes them faster because they can get more things done.
If you have a workload that can fit nicely into a narrower but faster clocked GPU, then things are going to be faster
Same concept as a 22 core xeon is not going to make your single threaded applications run faster!
Comment
-
Originally posted by boxie View Postthere are a shed load of busses at play here, largely they are
Disk->RAM
RAM->Cache
Cache->CPU
RAM->PCIE->Video RAM
Video RAM -> GPU
Originally posted by boxie View Postalso "Faster" here is a tad ambiguous. The higher end graphics cards are "wider" in their ability to execute, this makes them faster because they can get more things done.
If you have a workload that can fit nicely into a narrower but faster clocked GPU, then things are going to be faster
btw, 470 has lower clocks than both 460 and 480. you have to stop drug abuse and start checking your theories against reality
Originally posted by boxie View PostSame concept as a 22 core xeon is not going to make your single threaded applications run faster!
and except if you can make application faster by switching videocard, then this app is not cpu-bound by definition ffsLast edited by pal666; 22 December 2016, 08:20 AM.
Comment
-
Originally posted by pal666 View Postplease stop, otherwise next time you will be citing network latency. no amount of irrelevant information will make you right
if pigs can fly
btw, 470 has lower clocks than both 460 and 480. you have to stop drug abuse and start checking your theories against reality
except gpus are inherently multithreaded !!!!11111
and except if you can make application faster by switching videocard, then this app is not cpu-bound by definition ffs
Now let's address your concerns.
1) Network latency does not come into affect here
I was listing busses as an example of what the game has to interact with, given that most of them are the same between the cards they would have only a small amount of effect on the final results.
The main difference being the speed at which the cards can accept and store and retrieve data from/to the PCIE bus and how fast the onboard videocard RAM can feed the GPU
2) the 470 is overclocked. both in terms of GPU speed and bus speed
3) CPUs are inherently multithreaded too!
4) CPU bound is a tricky thing. what is it doing that is making it not able to do any more? Is the code path for the 470 slightly better than the 480?
as I have said before a CPU can be bound by doing a data transfer, and if that data transfer happens quicker on the overclocked card then it is free to do other things
5) as for the "switching graphics cards makes it faster" The main reasons that the 480 is "faster" than the 460 are: Bus Size (128 vs 256 bit) and GPU Width (more CUs). To a lesser extent the amount if RAM (and therefore fewer costly RAM->VRAM transfers)
programming is hard. contention on a resource could be the reason why we see a lot of the results.
Comment
Comment