codedivine
Regular
So I found 2 public presentations given by AMD employees which give a little bit of hint (but are very confusing) about the size of cache:
a) http://www.sharcnet.ca/events/ssgc2008/presentations/yang-sharcnet-final.pdf
On slide 11 : It shows that all 4 texture units on rv670 share a single L1 cache and quotes unified 32k L1 texel cache and 32k structure cache unfiltered. It also quotes shared 256kb L2 cache.
From the above, I would understand that the TOTAL L1 texture cache is 32kb and not 32kb/TU as quoted in many articles on the web. Is this correct?
b) In a presentation by Mike Houston he says for Rv770 : "2.5x increase in aggregate L1 compared to ATI Radeon™3800 series"
I take this to mean that the total L1 cache on rv770 is 2.5*32=80kb giving 8kb L1 per SIMD.
Comments? Info? Anyone has any testing results?
a) http://www.sharcnet.ca/events/ssgc2008/presentations/yang-sharcnet-final.pdf
On slide 11 : It shows that all 4 texture units on rv670 share a single L1 cache and quotes unified 32k L1 texel cache and 32k structure cache unfiltered. It also quotes shared 256kb L2 cache.
From the above, I would understand that the TOTAL L1 texture cache is 32kb and not 32kb/TU as quoted in many articles on the web. Is this correct?
b) In a presentation by Mike Houston he says for Rv770 : "2.5x increase in aggregate L1 compared to ATI Radeon™3800 series"
I take this to mean that the total L1 cache on rv770 is 2.5*32=80kb giving 8kb L1 per SIMD.
Comments? Info? Anyone has any testing results?