Are PS3 devs using the two mem pools for textures?
- Thread starter Love_In_Rio
- Start date
nAo meant exactly what he said. There are lots of issues in a console, all of which matter. Still, the answer you're after is 'no' as the devs would work around it. And I suppose 'yes' when they don't!
Perhaps nAo's reference to the GPU being 'designed' to offset that relates to changes reportedly made in the cache system. I'm not sure what the hit-rate is like with the cache before, but perhaps for data access going over flexio, they improved it further (via size or otherwise) so that more subsequent requests could be handled by cache rather than having to go back over flexio again.
That asides , as for what programmers can separately do to mitigate differences in latency between different data accesses, I guess whether it matters at all, or how much it matters, depends on the workload.
That asides , as for what programmers can separately do to mitigate differences in latency between different data accesses, I guess whether it matters at all, or how much it matters, depends on the workload.
Yep, it's very clear that you didn't sign any NDA
Texture caches can only help you when the chip is re-using texels, but they don't address double (or ten times more) latency per se, every texture cache miss is going to stall/starve your chip if you did not design it to cover your average latency.
Does that figure also include grabbing geometry
How might one deal with the extra latency in a game situation? Perhaps streaming objects not in a player's field of view?
GPUs have been able to fetch vertices from system memory since ages, nothing really new here.Does that figure also include grabbing geometry
You shouldn't see it as "extra latency", it's just latency, it's not like GPUs started to handle hundred cycles latency yesterday
I get the feeling that there should have been a smilie at the end of that. I certainly didn't see it as rude.
*nudge* I didn't say it but I'm just going to point out the person who said something *nudge* but I'm not going to confirm anything *wink*.
Leave it to AlStrong to ask the really newbie question.
I suppose I'm having a hard time seeing the implications of the higher latency of accessing the XDR versus the GDDR3. It is a developer thing, but I'm too curious to know a specific example (one out of eleventy billion that might be used).You shouldn't see it as "extra latency", it's just latency, it's not like GPUs started to handle hundred cycles latency yesterday
Unlike press NDA's which are used to control the timing of information release, developer NDA's never expire.
I suppose I'm having a hard time seeing the implications of the higher latency of accessing the XDR versus the GDDR3. It is a developer thing, but I'm too curious to know a specific example (one out of eleventy billion that might be used).
IF the GPU can't hide the latency then it would cause the GPU stall fairly continuously resulting in slower rendering performance.
I was not rude, I was simply stating the obvious given that he was giving a blunt statement about stuff he can't possibly know.
As I said you should see the implications of latency not higher latency, GPUs hide latency all the time, it does not seem a miracle to me if a GPU can hide latency.I suppose I'm having a hard time seeing the implications of the higher latency of accessing the XDR versus the GDDR3. It is a developer thing, but I'm too curious to know a specific example (one out of eleventy billion that might be used).
If you are a good GPUs architect you will design a GPU that will cope with your whole system latency, if you don't..well..they should fire you!
Similar threads
- Locked
