No - the average number of ALUs used per instruction is a bit lower for VLIW4 than for VLIW5, since there are a number of cases where the shader compiler could pack a single component operation into the same instruction as a 4-vector operation. [...]
In general VLIW5 was better for pure graphics workloads, but as compute became a larger part of GPU workload (there's a lot of compute hidden in modern graphical apps as well) then VLIW4 became a better fit.
what kind of graphics workload? raster or ray-tracing ? because ray-tracing graphic load is pure compute workload....
and right: "(there's a lot of compute hidden in modern graphical apps as well) then VLIW4 became a better fit."
why he should buy hardware for obsolete stuff? the 4D VLIW architecture is better for modern games.
Originally Posted by bridgman
The point was that the utilization as a percentage was slightly better with VLIW4.
that was my argument but in the past i read something about a average usage of 3,5
Yep, I agree that a VLIW4 GPU is a bit more future-proof than a VLIW5.
FWIW, I didn't get the impression that crazycheese was planning to buy an HD58xx, just wondering what performance was like these days.
Correct, I was just planning to do it in near future(1-2 months from now). But only if the results are acceptable. Unfortunately, I was unable to find any benchmarks of this GPU especially with open driver up at openbenchmarking. The VLIW GPU is also very solid-looking for non-graphical stuff. Besides, there are Marek&co who actively were seeking the ways to improve the driver.
Now, I forgot that pre-SI GPUs have 5 units/block, yet they are in 1/4 config. So, its either 6950/70 or 7xxx area for me left. 7xxx area is unstable and inefficient, so it seems I have plenty of time.
Have been reading on Itanium VLIW implementation today, and besides mentioning (absent in opensource) VLIW compiler as being the most important and required for VLIW hardware to work efficiently, there was an indication of "special hardware feature" allowing to profile the current execution within VLIW for the goal of improving the compiler itself. What makes me wonder is if AMD is willing to provide the documentation for that mode (in case it was implemented, and not hardware simulations were methods to optimize the execution)?.. Would be real help for pre-NI driver.
Originally Posted by necro-lover
my argument was also a "Future" argument *in the Future the 4D VLIW is much better than the old 5D VLIW cards*
I just don't want him to buy a hd5000 card because its technically bullshit in a modern world of more and more complex shaders.
If he buy a hd7970 he buy 4 shaders and use 3-3.5 then he lost only 1-0,5 instead of you buy 5 shaders and only use 2-3 and lost 2-3
I've got and understood your warning in your first post, n-l! Thanks for it!
Last edited by crazycheese; 09-15-2012 at 06:27 PM.
I think bridgman was referring to the technique here, meaning that APU (trinity, not previous ones) benchmarkings of opensource vs closed source will deliver same difference levels as if when using 6950/70 for this.
Trinity is not yet available for desktop platform.
Last edited by crazycheese; 09-16-2012 at 10:17 AM.
- VLIW5 is less efficient and more complex than VLIW4. Pre 6xxx are better recycled.
- The key why all Radeons (pre-Northern Island GPUs) are so slow with opensource driver - is absence of efficient compiler or whitepaper how to write it; which AMD is not releasing.