Interesting comment on Columbia and NV30 from Anand

[McElvis]

Got this from Anand's 'Inside Via' article -

Although a number of graphics related announcements came out of the S3-VIA joint venture at Computex, none of them were particularly exciting. The graphics business is quite lucrative and it makes a lot of sense for a chipset manufacturer to dabble in it since the fastest growing sales are of chipsets with integrated graphics. VIA currently plans on becoming much more competitive in the graphics sector with the release of their forthcoming Columbia GPU, a DX9 part due out late this year. We've heard claims as wild as NV30-like performance out of Columbia when it ships but seeing that NV30 is little more than numbers on paper right now, it's quite simple to make such claims; when push comes to shove, it will be interesting to see whether Columbia ends up being the C3 of graphics chips or something that exceeds all expectations.

Interesting to note :
1/ Columbia is a full DX9 high (possible) performance card, released THIS year.
2/ Anand states 'seeing that NV30 is little more than numbers on paper right now'.

Link : http://www.anandtech.com/printarticle.html?i=1637

 

[king_iron_fist]

VIA surely haven't got the tech to make that kind of gpu by themselves and I doubt S3 could do it....

I'm hoping there's a powervr link somewhere .

 

[ActionNews]

Hmm..didn't VIA say in an interview, that they do not have the technology with S3 to make a highend-graficscard? As i remember they said that S3 is good for Mainboard-chips and integrated chips but for highend-cards they need a new partner and i remember VIA said they want the Kyro-technologie !

Hmm...end of the year? Perhaps PowerVR Series 5? Might be possible as i remember a PowerVR-Roadmap saying Series 5 iss ready end of this year and www.hardware.fr claims to know that Series 5 is a DX9-chip !

But it's all speculation !

CU ActionNews

 

[Gollum]

Hmm...end of the year? Perhaps PowerVR Series 5? Might be possible as i remember a PowerVR-Roadmap saying Series 5 iss ready end of this year and www.hardware.fr claims to know that Series 5 is a DX9-chip !

Its, speculation, but that would be nice! If PVR could finally enter the market with a high-end part, then the past 3-4 years of waiting might have finally been worth it!

 

[amk]

Haven't Columbia rumours being knocking around for a very long time? As in, not long after STM licensed PVR S4 and S5, and long berfore ST decided to jettison its graphics division? Wouldn't that make the idea of Columbia being PVR based a little far fetched? If IMG would announce STM buying licenses, why wouldn't they announce VIA buying them?

Maybe it's BitBoys based

 

[Hellbinder]

Let me share a little *insider* information with you....

S3graphics licenced AFR technology from ATi. The Columbia is really....

8 Savage 2000 chips on one board running AFR.

They retaped the 2000 of course, T&L is actually working now. 8)

 

[LittlePenny]

Anand sounds a bit like ben6.

 

[KimB]

1/ Columbia is a full DX9 high (possible) performance card, released THIS year.

Interesting, but given the terrible graphics reptuations of VIA and S3, no way is it going to be any good in the real world (even if it does look decent on paper).

2/ Anand states 'seeing that NV30 is little more than numbers on paper right now'.

Because it's not released yet. I don't believe for a moment that this is the case to the engineers over at nVidia. Supposedly NV30 has already taped-out, which would put them in line for a late August/September release, unless they do have a large number of errata to fix (which is not out of the question in the least).

 

[Joe DeFuria]

Because it's not released yet. I don't believe for a moment that this is the case to the engineers over at nVidia.

Well, I think his point was that comparitively, the R-300 is not "released" yet either, but it surely is more than "just paper." I would certainly HOPE the engineers at nVidia have some prototype silicon they are working with...otherwise nVidia is further behind than most of us anticipated.

It seems fairly obvious at this point that nVidia has briefed some developers and journalists on the NV30. Probably the architecture, and the "expected specs" (clock-speed, etc.),.

nVidia though, doesn't have from what I can tell, a good track record of shipping products that meet the their internal "pre-launch anticipated specs." Particularly for new cores. I'm pretty confident that nVidia will ship NV30...in "some" form, within a few months of when the R-300 ships. In my opinion, it's not really a question of "when", it's more of a question of what that "form" is. Will it be the same specs and chip they are briefing to the press, (which Anand has been impressed with) or will it be toned down (either in performance or availability of chips) for the sake of shipping the product.

My prediction: I expect NV30 to be "launched" at about the same time that R-300 ships. We'll see if nVidia includes clock speeds and pricing at the time of the NV30 launch, or if it's more "technology" oriented. The type of "launch" that NV30 has should give us an indication of how close or far away it is.

 

[KimB]

Don't forget that nVidia has only once, in recent history, announced product specs significantly before the ship date, and nVidia has only once been innacurate about the final product specs. This was with the TNT.

I don't know about you, but I cannot be certain whether or not their more recent chips have fallen short of or exceeded expectations. Granted, it is very likely that many of them fell short of expectations, as that seems to be the norm in this industry. But, that still doesn't mean that you can generalize about a whole company based on one product announcement that fell way off the mark.

 

[Roger Kohli]

Why would VIA be intersted in the ST stuff if they already have a DX9 part in the works? Maybe they were just looking at it for integrated graphics because of its bandwidth efficiency?

 

[Joe DeFuria]

Actually, there were TWO products that fell way off the mark that we know of: the second time was much more recent with the NV2A (x-box).

In addition to those two confirmed incidents, I strongly "suspect" that the original GeForce256 fell victim to the same thing. Just about everyone was surprised how low those chips were clocked...and how power hungry they were...having problems with motherboards most chips did not because of power draw. The boards were also notoriously scarce for a longer time than ususal. In other words, it's hard to believe that the GeForce256 in its shipping form was what nVidia wanted it to be. Though indeed we can not know for "sure."

And we probably won't ever know about Nv30 for sure either, unless some people leak any "projected specs" if nVidia has in fact briefed them with this info. But it will be interesting to see if the chip generally "exceeds, meets, or fails to meet" the expectaions of the web reviewers like Anand, which should give us some insight. The pricing and availability of the chips will also give us some insight.

 

[jb]

Dont forget if they have to respin the chip, that will add time. Hard to get everything right on something with that many trasistors right the first time

 

[psurge]

jb, yeah that's what i was thinking too.

But...
If it is a massively parallel programmeable design, with a bunch of identical execution units, perhaps it's not as huge an undertaking as it seems. I mean for the chip after this one with 200million transistors,
how are they going to model/debug/verify the chip? I'm assuming that besides being nice for the 3d developers, programmability makes the chip designers job easier as well...

Serge

 

[Laa-Yosh]

Well, the S3/Via guy we've talked to during CeBit was hinting at something like this.
"There's enough room for more players in the highend market", he said.

 

[Oompa Loompa]

VIA surely haven't got the tech to make that kind of gpu by themselves and I doubt S3 could do it....

I'm hoping there's a powervr link somewhere .

Zeotrope, Paramount, and Columbia were in S3's roadmap even before they were acquired by VIA.

 

[Geeforcer]

If so, where does this leave PVR? If this rumors are true, Via is going ahead with internaly developed chips instead.

 

[KimB]

Dont forget if they have to respin the chip, that will add time. Hard to get everything right on something with that many trasistors right the first time

Supposedly the NV30 has had a "record number of tapeouts," and I'm pretty certain that this was about a month ago, so they're quite a ways from their first time by now...

 

[Oompa Loompa]

Supposedly the NV30 has had a "record number of tapeouts," and I'm pretty certain that this was about a month ago, so they're quite a ways from their first time by now...

No, NVIDIA has had a record number of tapeouts. A tape-out is the approval of a physical design layout that is sent for masking, to produce the mask set for manufacturing. If a design needs revisions that's one thing, but "tape-out" refers to the first time it is sent out to the mask house.

 

[Ailuros]

If so, where does this leave PVR? If this rumors are true, Via is going ahead with internaly developed chips instead.

a) I think we can assume after the rather long period of silence that the ST Micro/Via negotiations either never took place or were blown prematurely.

b) If VIA should be hinting on a possible future cooperation with PVR I doubt it will be earlier than early next year. There wouldn´t be a reason though to mention Columbia though since it is a codename coming from S3´s roadmap.

c) I don´t think PowerVR has changed it´s last roadmap. According to more than one rumours Series5 is supposed to be dx9 compliant. Wether VIA will buy a licence or not from them is a chapter of it´s own. After all VIA loves to blow deals