NV30 production problems

[kid_crisis]

Hmmm, I think that this speculation by the News Editor at Semiconductor Fabtech has been mentioned before, but here we go again about the low k nightmare:

The NV30 graphics chip was planned to be launched using TSMC’s 130nm copper and CVD based low k dielectric process. This would have been the first chip tape out for volume production using the most advanced foundry process currently available. RussSchultz, if the process is broken how can you fix it by tweaking your silicon?

Edit: I ask because TCMS have a less advanced but nice working .13 process on less complex chip designs.

About 2.5 months ago Jen Hsun claimed that they would not be using low-k dielectrics on the NV30 design. He did not elaborate on the statement though (i.e. did they originally plan to use low-k and then had to "fallback" to a no-low-k approach?) So if they are seeing more problems in the 0.13um process on NV30, it's probably not due to low-k dielectrics. The other big problem that foundries were having was voids in the copper interconnects that caused broken connections when temperature cycled, i.e. in reliability testing. That isn't something that would probably be known halfway through a processing run though (which is presumably where NV30 is right now, I guess).

 

[Mephisto]

A link to where you got that info would be really appreciated m8...

This is where we come at a philosophic point. All true information can be linked in the internet, otherwise, it is not trustfull, is it Alexsok?

I'm sure CMKRNL knows what he is talking about considering his past information here.

 

[RussSchultz]

RussSchultz, if the process is broken how can you fix it by tweaking your silicon?

Edit: I ask because TCMS have a less advanced but nice working .13 process on less complex chip designs.

What I suggested was that their process wasn't really broken, but improperly modeled, which could cause the back end work (synthesis and layout) to give you a chip that doesn't work or works only at very low speeds.

Redoing the synthesis and layout with the proper models (but keeping the same RTL/logical design) would fix the problem.

Or, as somebody suggested the VIAs were developing voids. (maybe) making them bigger would solve the problem, but that would require a new layout...

Or maybe the analog stuff is having problems (as its tougher to model this stuff), and they are doing tweaks like making transistors bigger, making the gates thicker, etc to account for the process issues.

 

[LeStoffer]

Thanks Russ, I understand now. 8)

BTW: TSMC offers two basic versions of 0.13-micron technology: They are both copper-based technology, but the first is with an FSG (fluorine-doped silicate glass) option while the second version is the low-k dielectrics option.

 

[Evildeus]

I don't know if Nvidia is going to have some paper launch or not in november, but the launch will be at that time:

Separately, Nvidia spokesman Derek Perez confirmed reports that the NV30 will launch at the Fall Comdex show in Las Vegas in late November. The company is considering an early-morning press conference.

http://www.extremetech.com/article2/0,3973,573873,00.asp

 

[Sabastian]

I don't know if Nvidia is going to have some paper launch or not in november, but the launch will be at that time:

Separately, Nvidia spokesman Derek Perez confirmed reports that the NV30 will launch at the Fall Comdex show in Las Vegas in late November. The company is considering an early-morning press conference.

http://www.extremetech.com/article2/0,3973,573873,00.asp

Well CMKRNL we are getting conflicting information here. I really have to take your claim with a grain of salt now my frend.

NV30 to be Revealed at Comdex Fall: Now Confirmed by NVIDIA!
Posted 10/02/02 at 7:31 pm by Anton

In early September some of our friends, who are familiar with NVIDIA’s plans told us that the highly-anticipated NV30 graphics processor will be unveiled at Comdex Fall show in Las Vegas late November (see this news-story). Now NVIDIA’s Derek Perez confirmed it at ExtremeTech.

If you read our previous news-issues about the NV30, you probably remember that the VPU features 8 rendering pipelines with 2 TMUs on each one, beyond DirectX 9 shaders support, 256-bit DDR SDRAM memory, AGP 8x and a lot more technical innovations. The core-clock should be between 350 and 400MHz, while the memory frequency will be about 700MHz. Originally NVIDIA wanted to adopt DDR-II look-alike memory for their NV30 based graphics cards, however, since JEDEC still has not approved the appropriate standard, the company has to go with the DDR-I memory (see this news-story). The yield of the NV30 is not going to be very high due to a very complex design and new manufacturing technology, thus, the company may launch a lower-clocked version of their most powerful and feature-rich graphics processor. It does make sense in the current situation, since NVIDIA has to compete with the arch-rival ATI Technologies not only in terms of performance, but also in terms of features and prices. After ATI will roll-out a number of DirectX 9 supporting RADEON 9500-series of products in mid-October (see this news-story) for $179 and $219, NVIDIA will not be able to offer a graphics chip with the same features and price. As a result, they will try to play with two versions of the NV30, offering them to different customers.

Mr. Perez said that the company was considering an early-morning press conference. This phrase makes me think that when the sun rises in Las Vegas on November 18, so does the NV30.

I should also remind you that according to the unofficial information, NV30 based graphics cards will be available for sale almost right after the announcement.

 

[BRiT]

CMKRNL is more reliable than XbitLabs.

--|BRiT|

 

[Sabastian]

CMKRNL is more reliable than XbitLabs.

--|BRiT|

Actually it is NVIDIA’s Derek Perez that CMKRNLs post conflicts with.

 

[BRiT]

Actually it is NVIDIA’s Derek Perez that CMKRNLs post conflicts with.

No. If you read carefully, you will see Derek Perez confirms they will launch the processor at Comdex. It says nothing about having working hardware or being available to consumers at that time. Anything other than them launching the NV30 is nothing but speculation.

This is much akin to 3dfx launching the VSA-100, Voodoo5 at Comdex.

--|BRiT|

 

[CMKRNL]

The problem is in manufacturing, not in the design. It's unclear as to what extent the delay will be, but worst case scenario is a full layer spin.

 

[Luminescent]

What about all the hardware tidbits x-bit labs is touting for the NV30, are they in any way reliable?

 

[antlers]

I don't think the part about "available almost right after the announcement" in the xbitlabs story claimed to be quoting Derek Perez, although I may be mistaken.

Also, "almost right after" could still be 3 months. Mr. Perez is in marketing, after all.

 

[Luminescent]

What about the 256 bit memory bus? I wonder who x-bit's sources are.

 

[Sabastian]

Actually it is NVIDIA’s Derek Perez that CMKRNLs post conflicts with.

No. If you read carefully, you will see Derek Perez confirms they will launch the processor at Comdex. It says nothing about having working hardware or being available to consumers at that time. Anything other than them launching the NV30 is nothing but speculation.

This is much akin to 3dfx launching the VSA-100, Voodoo5 at Comdex.

--|BRiT|

Hrm on second glance(well if I look around a bit.) it does seem like a rather obscure peice of news. I would think that nvidia would have some hard hitting PR somewhere other then Extreme Tech.(Actually I like these guys.) But there really is no offical PR to back up this article. Further at the bottom of the Xbit piece there is a disclamer of sorts. RE: "I should also remind you that according to the unofficial information, NV30 based graphics cards will be available for sale almost right after the announcement. " But time will tell. A paper launch followed by three months of nothing will not impress anyone.

 

[CMKRNL]

If anyone is expecting volume shipments after Comdex, I can guarantee this will not happen. Even if the manufacturing issues cost them a minimal amount of time, we're still talking about FIB'ed first silicon at best in the Comdex timeframe. Assume at least one spin prior to production on a chip of this complexity. Ramp is expected early next year.

I don't want to comment on the rumor'ed specs other than to say that I haven't been as convinced of its ability to trounce R300 (performance wise) as some others have.

 

[Luminescent]

Well, if any of our previous speculations are correct, the NV30 will certainly not trounce all over the R300, although it seems to be more flexible. As to it's performance with respect to the R300 on current software, it is yet to be seen whether or not the NV30 will exceed the precidents.

 

[BenM]

I certainly hope NVIDIA doens't paper launch NV30. They did for nForce, and now nForce2, AMD doing it with their newest Athlons, and ATI with the AIW 9700...it's getting real old, real quick.


Personally, I was pumped about NV30 a couple of months ago. But since the hype and rumor mill has been saturating the net with speculation from the time NV25 launched, NV30 has become a "meh" issue. Hopefully when it actually arrives it won't be so "meh" in performance.

 

[JF_Aidan_Pryde]

I don't know guys, you'd be pretty darn hard pressed not to get all supersticious when it comes to 3D. The cursed Bitboys, cursed Comdex annoucements and oh boy the VSA shit storm all over again, coutresy of 100 elite ex-3dfx engineers.

In death they will fight, to drag their once bitter foe down the crapper..
LOL

 

[LeStoffer]

The problem is in manufacturing, not in the design. It's unclear as to what extent the delay will be, but worst case scenario is a full layer spin.

Would you mind getting back to us, when you source knows about the extent of the delay?

If anyone is expecting volume shipments after Comdex, I can guarantee this will not happen. Even if the manufacturing issues cost them a minimal amount of time, we're still talking about FIB'ed first silicon at best in the Comdex timeframe. Assume at least one spin prior to production on a chip of this complexity. Ramp is expected early next year.

It makes perfect sense. Thank you.

I don't want to comment on the rumor'ed specs other than to say that I haven't been as convinced of its ability to trounce R300 (performance wise) as some others have.

Interesting indeed. I obviously don't know about the specs of the NV30 but so far I haven't see any reliable info on the NV30-architecture that would support the reasoning that NV30 will kill R 9700 of easily. (I guess we only have Creatives word for it after all).

 

[Randell]

Interesting indeed. I obviously don't know about the specs of the NV30 but so far I haven't see any reliable info on the NV30-architecture that would support the reasoning that NV30 will kill R 9700 of easily. (I guess we only have Creatives word for it after all).

And Creative have to justify not going with the the R300 after all. 1 million units shipped already, yikes.

I didn't think there could possibley be that level of demand for a new >$300/£300 GPU/VPU.