More rumors on NV28/NV30

[McElvis]

This from HardOCP :

Guys talk, you hear things....We had stated in another edition of the Rumor Mill a few weeks back that NVIDIA's NV30 had "come back" from the fab at .15 instead of .13. Of course, that is oversimplifying a drawn out process, but we were still apparently wrong. Just another reason why you should never trust the Rumor Mill 100%. I know we don't. Anyway, word on the street this week is that NVIDIA will be launching their NV28 core to satisfy this upcoming product cycle. From what we can gather, the NV28 will pretty much be a hopped up GeForce4 Titanium series core that will take advantage of 8X AGP. Little information beyond that is worth posting.

Lastly the Rumor Mill is saying this week that the NV30 chipset had not been taped out yet and will not be till next week, but will surely be at .13 micron and will make the Christmas buying season.

Remember this is all just rumor and speculation and should all be taken with a grain of salt...and possibly a beer as well.

Gets more confusing as time goes on...

NV30 not yet taped out - ???
NV30 available at Christmas - Is this still 'in the fall'?
NV28 to satisfy this upcoming product cycle - NV28 v R300?
NV28 just a GeForce4 with AGP8 - what happend to the extra ~20 million transisters?

Maybe nVidia are putting out all these rumors just to keep people guessing

 

[KimB]

As for "nVidia putting out rumors," I doubt it. There are enough out there as it is...there is no need for nVidia to spend any money on it.

After all, consider for a moment how many fake pictures we've seen floating around the web. How much more effort is it to create one of those than spread a simple rumor? I'd say quite a bit...which means that most rumors are people pulling numbers out of their collective ass in order to get a charge out of people being wrong.

 

[Dolemite]

I can't see how a "hopping up" can account for the additional 20+ million transistors, NV25 -> NV28, not to mention the even greater increase in complexity from NV17 -> NV18.

 

[KimB]

The coolest thing that nVidia could do with the extra 20 million or so transistors, in my mind, would be to expand the flexibility of the NV28's programmable pipelines (i.e. make it as programmable, or nearly so, as the NV30, but without the extra color depth, and possibly without as many T&L/pixel pipelines).

 

[Dolemite]

Personally, I'd like to see the rumored free M-buffer FSAA.

BTW, where exactly did these transistor numbers come from originally? I'm still a bit skeptical as I'm not sure how an 80 million transistor NV18 will play out as a value chip.

 

[KimB]

Well, you can't have free FSAA on an IMR, as FSAA always results in a memory bandwidth hit.

(except maybe with overly-excessive memory bandwidth...such as if eDRAM made it...).

 

[ben6]

NV30 has taped out. That's public information.

 

[Geeforcer]

Kyle keeps shooting himself in the foot.

I can see it now:

"Guys talk, you hear things....We had stated in another edition of the Rumor Mill a few weeks back that NVIDIA's NV30 had "come back" from the fab at .15 instead of .13. Of course, that is oversimplifying a drawn out process, but we were still apparently wrong. In yet another, more recent edition of the Rumor Mill we have stated that NVIDIA's NV30 had "not taped out yet". We were apparently wrong again; that's 0 for 2 for the folks keeping the score at home. I will now take a break from spreading baseless rumors and write a self-serving two paragraph rant at how Inqwell should adhere to the journalistic standards and stop spreading baseless rumors.â€￾

 

[FrgMstr]

NV30 has taped out. That's public information.

Please share this with me will you? I would really like to know where to get that knowledge.

Email me at kyle@hardocp.com

 

[Dave Baumann]

Kyle, its been spoken about in some of the recent investors talks NVIDIA have been doing.

 

[FrgMstr]

Kyle keeps shooting himself in the foot.

I can see it now:

"Guys talk, you hear things....We had stated in another edition of the Rumor Mill a few weeks back that NVIDIA's NV30 had "come back" from the fab at .15 instead of .13. Of course, that is oversimplifying a drawn out process, but we were still apparently wrong. In yet another, more recent edition of the Rumor Mill we have stated that NVIDIA's NV30 had "not taped out yet". We were apparently wrong again; that's 0 for 2 for the folks keeping the score at home. I will now take a break from spreading baseless rumors and write a self-serving two paragraph rant at how Inqwell should adhere to the journalistic standards and stop spreading baseless rumors.â€￾

Hmm, my foot feels just fine. Seriously, when we present information that we are not fully 100% are true, we very much make that known to our readers....unlike some other folks around our community.

Also, if we find out if we were wrong, we tell folks.

If that is shooting ourselves in the foot, guess I had best go get some steel toes as it is not going to stop as we are very rarely wrong with our rumor mill information. I will assure you it does not originate via hotmail...

 

[ben6]

Hey Kyle , was nice meeting you at San Francisco, though you didn't know who I was in the Cirque room. Was a blast meeting some of the people who I talk to in forums once in a while. Did you make your filight? when me and doc overclock were leaving the Fairmount, you hadn't come down for the waiting car... Nvidia has had a investor's meeting last week or this week where it was mentioned NV30 was taped out.

Ben

 

[Geeforcer]

Ben and Dave are correct.

Kyle,

I apologies if my parody offended you. It just that you posted 2 NV30 related "rumors" in a row despite the availability of the official information to the contrary, as 10 minute of research would have shown you. Perhaps you should pay more attention to company's conferences and less to these "reliable" sources.

 

[MuFu]

Which part of "rumour mill" do people not understand?

AFAIK, there was an internal rumour at nVidia itself that suggested that NV30 had TO'd on 0.15u. I'd love to think was paranoia in the buildup to ATi's R300/RV250 launch , but it's more likely a confusion of parts (not great if that happens in the bedroom, believe me...); nVidia mentioned that the only 0.13u part this fall would be NV30 in that presentation, so it's likely that the NV28 (supposedly NV25 + AGP 8x + ???) is 0.15u.

So news of a 0.15u ASIC returning from TSMC spreads, people talk over lunch, the guy that fills coffee machine doesn't know shit about anything, he tells his dog, the dog tells the cat and the cat tells an nVidia engineer who has a great love of animals and BANG! Suddenly "NV30 came back from the fab at 0.15u".

In my experience, having talked to, err... certain people, there is a great deal of segregation at nVidia and at ATi - the software and I/O guys just get the register specs and electro-mechanical data to work with, which is all they really need. Architecture/process information is restricted to a few select individuals.

MuFu.

 

[Ailuros]

Hey don´t underestimate my dog *snicker*

 

[MuFu]

*woof!* 8 pipes *woof!* 2 TMUs per pipe *woof! MEATVISIONâ„¢ Technology *woof woof!*

 

[FrgMstr]

Kyle, its been spoken about in some of the recent investors talks NVIDIA have been doing.

Well, I am not an investor. I make it a issue with the person that handles my finanacials that we not invest in any tech companies we report on, therefore reducing any type of conflict of interest that might arise.

I appreciate the fact that it NVIDIA has said that NV30 is taped out, but I have yet to see it produced here beyond rumor. Apparently you guys have the facts, and if so, I simply asked for them to be stated with proof.

Who said it? When? Where?

You guys are just repeating your own rumors in my opinion....

 

[FrgMstr]

Ben and Dave are correct.

Kyle,

I apologies if my parody offended you. It just that you posted 2 NV30 related "rumors" in a row despite the availability of the official information to the contrary, as 10 minute of research would have shown you. Perhaps you should pay more attention to company's conferences and less to these "reliable" sources.

Well hey, three guys at Beyond3d Forums said it was true so I guess that is the gospel.

NVIDIA will not confirm any of this at this time, so I guess it is not that official.

 

[Typedef Enum]

I suggest you go to their site and listen to the last quarterly report. Whether or not the PR guys are going to reiterate that info. has nothing to do with the how legit. the info. is. Generally speaking, they aren't going to say squat about unannounced products.

 

[MuFu]

Ben and Dave are correct.

Kyle,

I apologies if my parody offended you. It just that you posted 2 NV30 related "rumors" in a row despite the availability of the official information to the contrary, as 10 minute of research would have shown you. Perhaps you should pay more attention to company's conferences and less to these "reliable" sources.

NVIDIA will not confirm any of this at this time, so I guess it is not that official.

Yeah? Interesting... I wouldn't take what they say to their investors to be necessarily corrrect, although you guys seem to have a lot of problems with large corporations and investment FUD, so I guess it's possible they aren't hiding the facts.

Don't forget that [H] reported this weeks ago. Alot can happen in a few weeks! A few weeks ago, R300 was stuck at 250MHz and setting nearby objects on fire.

My bet is that they had a team working on an 0.15u NV30 "contingency" ASIC simultaneously. Having said that, an automated shrink wouldn't have required too many R&D resources (limited eng time - only the financial cost & fab time will be painful) at the cost of some performance, so a separate team might not have been necessary. Either way... it could well have taped out on two processes by now. Either that or there was confusion with an NV28 tape-out, as I mentioned earlier.

When I heard of the rumour at nVidia, I also learnt that it stirred up quite a bit of worry, because the NV30 design relies on a high clock to reach performance targets. Their at-speed target is ambitious, but their performance-per-clock target is not. Hence the possibility of a very successful 0.13u R300 "counter-attack" on ATi's part. That's the way it seems to me, anyway.

MuFu.