PS2 and Gamcube question

[Wildstyle]

I was just wondering what is the efficentcy of the texture cache's in both of these systems. I have seen on various sites that the 1T-SRAM memory is only about 90% efficent so wouldn't that make it's texture cache 90% efficent but I have no idea about the PS2 and it's eDRAM so maybe one of you guys could please fill me in thanks in advance for all you time .

 

[Teasy]

I have seen on various sites that the 1T-SRAM memory is only about 90% efficent

90% efficiency for ram is actually extremely high.

As for the overall question. This was asked a while ago in another thread. Nobody seems to be able to answer it.

It may be impossible to compare the efficiency of GC and PS2's texture caches. Because for PS2 I think the skill of the developer has a large effect on the texture cache efficiency. Where as on GC the texture cache is totally controlled by the system (by defualt anyway). So its efficiency is less dependant on the ability of the dev using the system.

I suppose all you could do is compare the efficiency of the ram used for both caches. But that wouldn't really tell you whole story on the efficiency of the caches in GC and PS2.

I don't know what sort of memory PS2 uses for its texture cache, or what sort of latency it has. But I do know that the 1T-Sram used for GC's cache is low latency. Do any devs out there know what sort of latency PS2's texture cache/scratch pad memory has?

 

[Wildstyle]

Oh I see so basically if you program the PS2 good enough the texture cache could be 100% efficient. But I also think it's kind of odd that know one really knows it's true efficientcy especially since the console has been out for so long but isn't there some PS2 programers in this forum that could maybe fill us in guess i'll will have to wait and see for there input.

 

[Grall]

PS2 doesn't have a texture cache per se. The eDRAM is not a cache, it's just a memory area.

It won't be 100% effective, nothing ever is.

*G*

 

[Wildstyle]

PS2 doesn't have a texture cache per se. The eDRAM is not a cache, it's just a memory area.

It won't be 100% effective, nothing ever is.

*G*

Yeah your right but the PS2's eDRAM should be pretty efficient since it's DRAM instead of SRAM I am guessing it would have DRAM efficiency like around 75 to 85% but then again I have no idea and I am hopeing one of the PS2 programers in this forum could fill us in.

 

[Teasy]

Oh I see so basically if you program the PS2 good enough the texture cache could be 100% efficient.

No it couldn't be 100% efficient. I just meant that its hard to pin down its efficiency since its very dependant on the ability of the programmer. Probably the best you could do is find out the latency of the memory used.

Yeah your right but the PS2's eDRAM should be pretty efficient since it's DRAM instead of SRAM

I'm pretty sure DRAM isn't particularly efficient compared to SRAM. Infact AFAIK SRAM is more efficient at the expensive of extra size. AFAIR SRAM doesn't need to be refreshed like DRAM does.

 

[Fafalada]

Mosys mem is still DRam cells, refresh hidden by some nice caching tech.
And further on the note of GS memory, I believe Archie did address this issue to some extent.
Namely eDram on GS uses something of a similar principle. There are page buffers sitting between physical memory and rest of the chip, access to which is more or less latency free.
And loads from eDram to page buffers are performed at obscenely high bandwith (something over 150MB/s), keeping the effect of page breaks to minimum too.
Outside pathological cases I doubt you'd see large deviations in performance due to efficiency in case of either memory - which is pretty much the idea behind using them

Debating texture caching efficiency on system scale isn't particularly meaningfull though.
In general case GC wins by default - PS2 not really having a cache to speak of. (The one exception being render-to-texture where situation is reversed and PS2 wins by default).
Programmed schemes could be debated for awhile I guess, but personally I wouldn't mind trading a notable amount of CPU overhead and most likely less efficient bandwith use for arguably less efficient scheduling.

 

[Wildstyle]

Mosys mem is still DRam cells, refresh hidden by some nice caching tech.
And further on the note of GS memory, I believe Archie did address this issue to some extent.
Namely eDram on GS uses something of a similar principle. There are page buffers sitting between physical memory and rest of the chip, access to which is more or less latency free.
And loads from eDram to page buffers are performed at obscenely high bandwith (something over 150MB/s), keeping the effect of page breaks to minimum too.
Outside pathological cases I doubt you'd see large deviations in performance due to efficiency in case of either memory - which is pretty much the idea behind using them

Debating texture caching efficiency on system scale isn't particularly meaningfull though.
In general case GC wins by default - PS2 not really having a cache to speak of. (The one exception being render-to-texture where situation is reversed and PS2 wins by default).
Programmed schemes could be debated for awhile I guess, but personally I wouldn't mind trading a notable amount of CPU overhead and most likely less efficient bandwith use for arguably less efficient scheduling.

Thanks Faf well in that case the PS2's VRAM is pretty much 100% efficient so I am guessing it's about 98% efficient?

 

[Teasy]

Mosys mem is still DRam cells, refresh hidden by some nice caching tech.

Yeah I know that 1T-SRAM isn't actually SRAM and instead is just a more efficient kind of DRAM. I only meant that actual SRAM is the more efficient memory tech when compared to DRAM (more efficient from a performance point of view anyway).

Wildstyle

You missed the ] on your opening

. Could you edit that to make it easier to read please?