NV45?
- Thread starter overclocked
- Start date
overclocked
Veteran
Can someone maybe explain the difference in IBM´s vs TSMC different "offerings" and what´s the´pros/con betwen their offerings that affect the chip´s we are seeing and will see.
I've seen some NV45 docs on power consumption, sample availability and PCB layout.
Going by those things I'd say that the NV45 is to NV40 what NV38 was to NV35. (Identical PCB layout, more or less identical power requirements, samples allready available or at least will be within days.)
Going by those things I'd say that the NV45 is to NV40 what NV38 was to NV35. (Identical PCB layout, more or less identical power requirements, samples allready available or at least will be within days.)
overclocked
Veteran
Manufacturing process?
Evildeus said:Yes, i did see, but i was talking about consummers not Nv handsAnte P said:You just quoted the answer to your question..(last part of the quote)
All I've heard is that the Q3 figure found on some roadmaps during comdex still remains valid. (The latest, a week old, internal roadmap I saw had the exact same dates as the public ones at comdex)
NV45
Bus interface: PCI-E 16x
Memory width: 256 bit
PCB: P211
Performance: 3x 5950U (same roadmap says 2x 5950 for NV40U)
Sample Availability: April/May
Retail Availability: Early Q3
The PCB chart looked identical to the chart for the NV40 PCB. But then again the 5950 and 5900 PCBs changed in the very last minute before launch, as did the 5800 IIRC so perhaps the chart I saw was just for these early in house samples.
Can't share the original docs because then the source would be pretty obvious (AIB-branded pictures).
Got some more stuff on the other NV4x as well:
NV41
Bus interface: PCI-E 16x
Memory width: 256 bit
PCB: P260
Performance: 2x 5950U
Sample Availability: May
Retail Availability: Early Q3
NV43
Bus interface: PCI-E 16x
Memory width: 128 bit
PCB: P216, P229 and P228
Performance: 2x 5700U
Sample Availability: June
Retail Availability: Mid Q3
NV44
Bus interface: PCI-E 16x
Memory width: 64 bit
PCB: P261 and P280
Performance: 3x 5500
Sample Availability: August
Retail Availability: Mid Q3
That's about it.
Judging by these statements is it possible NV is either using a 0.11 fab or possibly even they’ve squeezed the transistors/traces on the NV40 down to make the die size smaller (hybrid 0.13 with 0.11 features?). The latter option would probably make the GPU produce more heat and scale less. I didn’t think low-k 0.13 would need a significantly larger die size for an equivalent GPU compared to IBM’s 0.13. In any case if NV wants to use the same low-k 0.13 that ATI is using to get better scaling, I would think they likely are going to end up with a die size ~ 40% larger -- equivalent to the additional transistors.
Given the poor overclocking results so far (hard time hitting 450MHz) I’d say the NV40 is pretty well EOL as far as scaling goes unless they do a complete re-layout and/or go low-k. NV had their 5700U at 475MHz pretty comfortably on 0.13 so either very large die sizes are hurting clock rates or NV, as noted, may be using a tweaked 0.13 process to get a smaller die size.
Scaling dead at 130-nm, says IBM technologist (EE Times) All Intel got at 0.09 with the Prescott was more heat. This also means that the next generations of GPU’s are likely going to cost significantly more if they are in the ~ 300 million transistor range, as die sizes are going to get really huge without some significant process breakthroughs in the next few years.
Ante P said:
Are these the same ones that say NV40 will be 600MHz?
PaulS said:
I'm still wondering about the NV44 which has 64bit bus, 3x faster than the 5500 which has a 128bit bus.
________
MEDICAL MARIJUANA PATIENT
Last edited by a moderator:
Similar threads
