I know this might seem like a strange topic; however, the more I look at old or extremely recent information pieces, the least sense it makes to me. The following is either fact, or highly reliable information.
- NVIDIA proposed 100 3DFX engineers to join them.
- A few key engineers joined, among which one who is responsible for NV40's Pixel Shading and Texture technology, including the L2 cache implementation afaik.
- 60%+ of the 3DFX engineers at NVIDIA work on nForce-related projects, and not on the GPU family of products. Remember 3DFX had a motherboard project in progress, which evantually failed because of the apparent impossibility to manufacture the design sufficiently massively and cheaply, partly because of a too high number of layers. (an ex-NVIDIA source told me all of the ex-3DFX engineers he knew at NVIDIA worked on the nForce chipsets)
- Ex-3DFX's contribution couldn't have been *too* significant for the nForce 1, as it is a direct derivative of the XBox chipset. They would have joined in the last 25% of the project; major contributions to the nForce2 seem more likely.
- Two major projects were underway at 3DFX; Rampage (and Sage, ofc, Spectre as a whole) and Fusion. The Rampage team was extremely dissapointed of the dissolution of the company, as it had working Rampage silicon shortly before that, and initial results were very positive.
- Fusion, on the other hand, was in its early stages and was to exclusively support FP16. This is extremely unreliable information though, but it feels quite likely to me; however, the project was in its preliminary stages, so it might have changed afterwards - it feels likely to me this is where the "FP16-FP32" compromise in the NV30 came from, as sort of a compromise between 3DFX and NVIDIA, who already had working FP32 in the NV20 for specific texture operations)
- Fusion's design was merged with NV30's, which I believe meant that initial rumors regarding the NV30 design came from 3 separate directions: The original, pre-buyoyt NV30 plans, the 3DFX Fusion project plans, and the Work In Progress for the final design at NVIDIA - which changed quite a bit due to a nice amount of fuckups. This, however, is once again speculation from my part.
- Some of Rampage's engineers went to ATI as I said, but I haven't got the slightest clue of the real proportions between NVIDIA and ATI. The only real influence I can see from Rampage in the R300 is the anti-aliasing technology, which does share many common points, although it is obviously improved in the R300 (mostly by doubling the number of ROPs per pipeline, I believe)
Now, regarding SGI:
- SGI's graphics division officially ended when they entered in a strategic alliance with NVIDIA (which is ironic, considering some of their current high-end is ATI-based) and many of their remaining engineers were given to NVIDIA for the NV20.There most likely is an influence from those engineers regarding the shading engines of the chip
- But I speculate not regarding its quad structure, which was already present in the NV10 from what I understand (using a single 4x2 quad has been a defining characteristic of all chips between NV10 and NV38, excluding the MXs that worked in a delay scheme I believe)
- Many SGI graphics engineers had apparently joined other companies - including NVIDIA, ATI and most particularly ArtX - before their strategic alliance with NVIDIA.
- Since the ArtX being buyout by ATI, NVIDIA now has less ex-SGI engineers than ATI. ATI has slightly more engineers than NVIDIA overall, or at least used to.
So anyhow, if anyone else knows something or just saw something on a forum/heard it from soemone midly reliable, please do say so! Recruitment proportions between the different IHVs, and possibly Quantum3D, would be most interesting. Oh by the way, I believe one of the uses of the new NVIDIA design lab in India might be to have testing capabilities nearer to Taiwan, reducing delays between tape-out, testing and worldwide release, including shipment to partners. Any confirmation on that would be interesting also.
Uttar
P.S.: Please?
I'm mostly curious really, this whole thing has been bugging me for a while; I seem to have a lot of elements that, put together, make a lot of sense but I don't have enough details to really be sure about the big picture.
- NVIDIA proposed 100 3DFX engineers to join them.
- A few key engineers joined, among which one who is responsible for NV40's Pixel Shading and Texture technology, including the L2 cache implementation afaik.
- 60%+ of the 3DFX engineers at NVIDIA work on nForce-related projects, and not on the GPU family of products. Remember 3DFX had a motherboard project in progress, which evantually failed because of the apparent impossibility to manufacture the design sufficiently massively and cheaply, partly because of a too high number of layers. (an ex-NVIDIA source told me all of the ex-3DFX engineers he knew at NVIDIA worked on the nForce chipsets)
- Ex-3DFX's contribution couldn't have been *too* significant for the nForce 1, as it is a direct derivative of the XBox chipset. They would have joined in the last 25% of the project; major contributions to the nForce2 seem more likely.
- Two major projects were underway at 3DFX; Rampage (and Sage, ofc, Spectre as a whole) and Fusion. The Rampage team was extremely dissapointed of the dissolution of the company, as it had working Rampage silicon shortly before that, and initial results were very positive.
- Fusion, on the other hand, was in its early stages and was to exclusively support FP16. This is extremely unreliable information though, but it feels quite likely to me; however, the project was in its preliminary stages, so it might have changed afterwards - it feels likely to me this is where the "FP16-FP32" compromise in the NV30 came from, as sort of a compromise between 3DFX and NVIDIA, who already had working FP32 in the NV20 for specific texture operations)
- Fusion's design was merged with NV30's, which I believe meant that initial rumors regarding the NV30 design came from 3 separate directions: The original, pre-buyoyt NV30 plans, the 3DFX Fusion project plans, and the Work In Progress for the final design at NVIDIA - which changed quite a bit due to a nice amount of fuckups. This, however, is once again speculation from my part.
- Some of Rampage's engineers went to ATI as I said, but I haven't got the slightest clue of the real proportions between NVIDIA and ATI. The only real influence I can see from Rampage in the R300 is the anti-aliasing technology, which does share many common points, although it is obviously improved in the R300 (mostly by doubling the number of ROPs per pipeline, I believe)
Now, regarding SGI:
- SGI's graphics division officially ended when they entered in a strategic alliance with NVIDIA (which is ironic, considering some of their current high-end is ATI-based) and many of their remaining engineers were given to NVIDIA for the NV20.There most likely is an influence from those engineers regarding the shading engines of the chip
- But I speculate not regarding its quad structure, which was already present in the NV10 from what I understand (using a single 4x2 quad has been a defining characteristic of all chips between NV10 and NV38, excluding the MXs that worked in a delay scheme I believe)
- Many SGI graphics engineers had apparently joined other companies - including NVIDIA, ATI and most particularly ArtX - before their strategic alliance with NVIDIA.
- Since the ArtX being buyout by ATI, NVIDIA now has less ex-SGI engineers than ATI. ATI has slightly more engineers than NVIDIA overall, or at least used to.
So anyhow, if anyone else knows something or just saw something on a forum/heard it from soemone midly reliable, please do say so! Recruitment proportions between the different IHVs, and possibly Quantum3D, would be most interesting. Oh by the way, I believe one of the uses of the new NVIDIA design lab in India might be to have testing capabilities nearer to Taiwan, reducing delays between tape-out, testing and worldwide release, including shipment to partners. Any confirmation on that would be interesting also.
Uttar
P.S.: Please?
I'm mostly curious really, this whole thing has been bugging me for a while; I seem to have a lot of elements that, put together, make a lot of sense but I don't have enough details to really be sure about the big picture.
