http://psinext.com/index.php?catego...mp;PHPSESSID=7b2d35c0b156e97084220eddfc306c0e
if this is old, I'm sorry
if this is old, I'm sorry
Could the RSX be more powerful than originally thought?
- Thread starter cobragt
- Start date
BOOMEXPLODE
Regular
Same reason why one SPE in PS3 will be disabled...redundancy. Otherwise they would have to throw away all the chips that had one little flaw, this way they can still use them (as long as the flaw is somewhere within the redundant areas, which is higly likely because that's the largest part of the chip).
Megadrive1988
Veteran
cobragt said:
maybe, maybe not.
RSX will be more powerful than G70 even though RSX is based on G70. that's all we know for sure. any additional pipelines that RSX might have, will be for redundancy, as mentioned.
*If* RSX ends up with 32 working pixel pipelines (instead of 24) then that will probably be at the expense of the 8 vertex shaders.
Since RSX doesn't need to differentiate itself from other GPU parts though in the same PC line-up, they would only need to deactivate a single quad, rather than the eight pipes presently needed to keep 'distance' in the PC world. So, it should theoretically be able to have a greater number of pixel pipes anyway, to say nothign of the different layout that might be applied to vertex shaders.
PS - I'm a strong believer in the current theory that G70 has a number of 'hidden' transistors - as stated in the article.
PS - I'm a strong believer in the current theory that G70 has a number of 'hidden' transistors - as stated in the article.
Ironic that someone here links to a story on another site that quite obviously originates from this site!
Hey, who told you about my secret game? :|ZoinKs! said:It's as if people start rumors just to see how long it'll be before someone else repeats it back to them.
I'm going to go further and just put some math out there.
1 - ([current process]/[past process]) = gain in transistor density with process shrink
So for the 130nm to 110nm shrink on an NV40, we're looking at a 28.5% increase in transistor count for the available area. That would/should - (where's that NV48 info?) - shrink the die area of the NV40 from 287mm^2 to 205mm^2 with it's 222 million transistors - giving G70 roughly ~ 120 mm^2 to work with above and beyond (334mm^2 minus 205mm^2 = 119mm^2).
With an assumed transistor density of 1,082,926 per mm^2 at 110nm on an equivelent process and derived from a 'dumb' shrink for the NV40, G70 should have roughly ~129 million more transistors at 334mm^2 than NV40. With an announced 302 million transistors, only 80 million those 129 million have been accounted for.
32 pipe part ready to go when R520 launches...?
1 - ([current process]/[past process]) = gain in transistor density with process shrink
So for the 130nm to 110nm shrink on an NV40, we're looking at a 28.5% increase in transistor count for the available area. That would/should - (where's that NV48 info?) - shrink the die area of the NV40 from 287mm^2 to 205mm^2 with it's 222 million transistors - giving G70 roughly ~ 120 mm^2 to work with above and beyond (334mm^2 minus 205mm^2 = 119mm^2).
With an assumed transistor density of 1,082,926 per mm^2 at 110nm on an equivelent process and derived from a 'dumb' shrink for the NV40, G70 should have roughly ~129 million more transistors at 334mm^2 than NV40. With an announced 302 million transistors, only 80 million those 129 million have been accounted for.
32 pipe part ready to go when R520 launches...?
Jaws said:
Sure, but at 90nm I could definitely see Sony with 32 pixel pipes and disabling a quad 'for redundancy' especially if that design presently exists in the chip on which it is based, and considering 65nm will be right around the corner for them. And the article mostly deals with G70 anyway, just more or less extrapolating what the possibilities might be for RSX down the line.
xbdestroya said:
G70 is 430 Mhz at 130nm and ~ 300 mil active trannies
RSX drop to 90nm with still ~ 300 mil active trannies AND and increase to 550 MHz sounds about right to me...
Anyway, aren't you forgetting from E3, RSX ~ 136 Shops/cycle and CELL+RSX ~ 51 GigaDots/sec?
With what we know of G70,
136 Shops/cycle ---> 136 instructions/cycle
G70 VS unit ~ 2 inst./cycle
G70 PS unit ~ 5 inst./cycle
24 PS (120 inst./cycle) + 8 VS (16 inst./cycle) ~ 136 inst./cycle
The only thinh that doesn't add up is that RSX should be 52 Dots/cycle not 56 Dots/cycle, i.e. 1 vec4 unit = 1 Dot/cycle.
24 PS + 8 VS ~ 136 inst/cycle, 56 Dots/cycle
If the VS units are upgraded to ~ 4 inst/cycle, then,
24 PS + 4 VS ~ 136 inst/cycle, 52 Dots/cycle
In any case not everything is adding up!
Jaws said:G70 is 430 Mhz at 130nm and ~ 300 mil active trannies
RSX drop to 90nm with still ~ 300 mil active trannies AND and increase to 550 MHz sounds about right to me...
Anyway, aren't you forgetting from E3, RSX ~ 136 Shops/cycle and CELL+RSX ~ 51 GigaDots/sec?
With what we know of G70,
136 Shops/cycle ---> 136 instructions/cycle
G70 VS unit ~ 2 inst./cycle
G70 PS unit ~ 5 inst./cycle
24 PS (120 inst./cycle) + 8 VS (16 inst./cycle) ~ 136 inst./cycle
The only thinh that doesn't add up is that RSX should be 52 Dots/cycle not 56 Dots/cycle, i.e. 1 vec4 unit = 1 Dot/cycle.
24 PS + 8 VS ~ 136 inst/cycle, 56 Dots/cycle
If the VS units are upgraded to ~ 4 inst/cycle, then,
24 PS + 4 VS ~ 136 inst/cycle, 52 Dots/cycle
In any case not everything is adding up!
LOL, ok ok good points
Well, RSX will remain a mystery until a bit later it seems. But G70, for it's part, seems to be packing some extra heat.
If I were Sony though I would definitely take a bath on the cost of 90nm parts and yields if I was confident I would have 65nm within my grasp shortly after that. They were willing with the GS afterall. Nagasaki was supposed to come on line 65nm, right? I wonder what the status is with that place right now...
xbdestroya said:I'm going to go further and just put some math out there.
1 - ([current process]/[past process]) = gain in transistor density with process shrink
So for the 130nm to 110nm shrink on an NV40, we're looking at a 28.5% increase in transistor count for the available area. That would/should - (where's that NV48 info?) - shrink the die area of the NV40 from 287mm^2 to 205mm^2 with it's 222 million transistors - giving G70 roughly ~ 120 mm^2 to work with above and beyond (334mm^2 minus 205mm^2 = 119mm^2).
With an assumed transistor density of 1,082,926 per mm^2 at 110nm on an equivelent process and derived from a 'dumb' shrink for the NV40, G70 should have roughly ~129 million more transistors at 334mm^2 than NV40. With an announced 302 million transistors, only 80 million those 129 million have been accounted for.
32 pipe part ready to go when R520 launches...?
You're saying G70 with 334 mm2 ---> 416 mil trans?
I get 361 mil trans...
NV40
287 mm2 -> 222 mil trannies
If NV40 was scaled to 334mm2 die as G70 but still on 130nm process,
287 mm2 -> 222 mil trannies
334 mm2 -> 258 mil trannies
If that die of 334 mm2 was drooped to 110nm from 130nm but the die size remained at 334 mm2,
(130/110)^2 ~ 1.4x as many tranisitors could be added to the die of 334 mm2.
So,
NV40 scaled to 334 mm2 and dropped to 110 nm,
258 x 1.4 ~ 361 mill trannies
Jaws said:
No no, you gotta read over my post again - what I'm reaching is 351 million transistors on G70. I think that portion of my post you bolded you took to mean I was scaling the NV40 up - I was just saying at 334mm^2, the G70 should have 129 million more transistors than the 222 possessed by NV40.
I agree that sentence of mine could be minsinterpreted though!
PS - It would be the style of both Sony and NVidia to release those 'official' shader/op RSX specs at E3 and then throw in a little boost come TGS a la 4 more VS pipes.
BOOMEXPLODE
Regular
xbdestroya said:
The "old" news and it's thread is here. Presumably all this discussion as it relates to G70 could take place there.
Steering things back on course: how could this "rumour" relate to RSX? So what if G70 really has 32 pipelines, hypothetically the other 8 pipelines HAD to be disabled to achieve reasonable yields. Wouldn't the same be true for RSX as well?