So you are suggesting that a middle-end GT300 series GPU is greater in performance than a theoretical cypress and boasts 512 bit bus width? I cannot fathom G300 for this reason, just want it now! I'll pay $649 for such a monster.
Nvidia GT300 core: Speculation
- Thread starter Shtal
- Start date
- Status
Well, that was in the context of Batman: Arkham Asylum. The question is: is it true in that game?
Since MIMD executes multiple instructions over the same amount of ALUs compared to SIMD you need more logic to decode and route these instructions. The amount of extra logic required will vary based on the design.
DegustatoR
Legend
Were did I say this?
I've said that GT212 was supposed to have performance that Cypress has right now. So ressurecting it and putting it against Juniper will not be good for Juniper and Cypress because performance >>>>>> features. I believe that if GT212 really get ressurected they'll put it against Cypress, not Juniper.
If you want my completely baseless insight on the upcoming 300 series then here it is:
$$100-200 -- "G304": 128 SPs / 32 TUs, 128-bit GDDR5, 2Q 2010 (~GTX260 perf)
$$250-350 -- "G302": 256 SPs / 64 TUs, 256-bit GDDR5, 1Q 2010 (1,5x GTX285 performance)
$$400-500 -- G300: 512 SPs / 128 TUs, 512-bit GDDR5, ??? (~2-2,5x GTX285 performance)
$$550+ -- G300 X2, ???
First you said there was no reason to resurrect an hypothetical GT212 and suggested a middle end G300 made more sense over it. Next you suggested that 1.4B trannies wont be enough for 512b. Adding all these I came to the conclusion that you were suggesting such a thing.
So you still think G300 is on track to have a real launch very soon? I hope so.
DegustatoR
Legend
Any "dumb shrink" of GT200 should have 512-bit GDDR3 bus. If it has something else then it's not a "dumb shrink".
The definition of "very soon" may vary. I do think that compared to the end of the world date set to 2012 G300 will launch "very soon", yes. Although it's pretty clear by now that it will launch later than originally anticipated.
Does that mean you think it'll be like Jan-Mar? Or even later? :|
Ok, so re-work for GDDR5 would mean its not a shrink ala RV770. So I guess GT212 (if there is/was/is such a thing) is more than a dumb shrink. And I also remembered some speculation regarding 384SP, 96TMU, 256b regarding it.
Clearly 2012 was supposed to be a joke, end of the world kind of joke. You cant take that seriously.
But it does say 24!
[quote
Less of what? You can buy thousands of them for your notebooks if you're an OEM.[/quote]
See, I'm not an OEM. for normal consumers there are exactly 0 GT21x derivate's available. A shrill contrast to the couple of hundred RV740's I can have at my door next week.
Quite an Amazing feat considering GT216x218 taped out at least one quarter before RV740.
Other thing would be that AMD admitted they made RV740 a pipe cleaner part. Sacrificing that half year with RV740 netted them almost a full year against the opponents 40nm products.
Still, 98 was better received, had a longer life and was a lot quicker than 95. The better OS.
I'll give you one reason, they fucked-up!
You're having really high hopes of GT215. Something close to G92 in the best scenario. As far as I know, those numbers aren't "overall" by a long shot. (Redwood would be about 20 to 30% faster on average)
They have to do something, GT300 might as well be a good 9 months out!
It's too damn quiet anywhere regarding NV's next generation in order to hope for a release soon.
Anyway wouldn't it be better if NVIDIA in the future concentrates on its roadmaps to release the performance part of a GPU family first and then anything else later?
Anyway wouldn't it be better if NVIDIA in the future concentrates on its roadmaps to release the performance part of a GPU family first and then anything else later?
275 was launched right next to 4890. Assuming gt300 launches late this year. Even charlie says it has been taped out, so it should launch by jan 2010 at the latest.
Even in the best case scenario, it will be almost 8-10 months since they launched a desktop consumer (non oem-only) gpu. And it looks like gt300 wont be able to sell at prices needed to recoup the massive r&d. I am pretty sure that they'll try to minimize the delay between gt300 and it's mid range derivatives this time.
Even in the best case scenario, it will be almost 8-10 months since they launched a desktop consumer (non oem-only) gpu. And it looks like gt300 wont be able to sell at prices needed to recoup the massive r&d. I am pretty sure that they'll try to minimize the delay between gt300 and it's mid range derivatives this time.
Park XT of the same TDP range as GT218 has 120SPs and GDDR5. ~86GT/3650 while the best we can see from GT218 is ~85GT+ (even less bandwidth too), which is a big difference.
Whatever GT218 can do, Park probably can deliver in <10W. New OEM sweet spot.
Madison LP/Pro (~4670) again does to GT216 (~8600GTS++) what Park did to 218.
Only the high end nVidia part seems more comfortable in competing (it's a 8800GS with 8800GT bandwidth after all), and that's basing on the assumption that Madison is just a 320SP part (480 = Sayonara all, kthxbai)
DegustatoR
Legend
Your guess here is as good as mine.
Right.
Well NV better take it seriously otherwise the end of the world may come to them sooner than expected. I mean, Juniper vs GT200b is kind of funny in a not at all funny for your margins way.
Why would normal consumers want anything GT21x-based? It's not like those GT21x chips bring anything big to the normal consumers. Or sure they can do DX10.1. But even RV770 had less power than needed to use DX10.1 in, say, Clear Sky. So that's pretty pointless checkbox feature for GT21xs. So while supply is limited NV's putting them where their TDP makes a difference -- i'd do exactly the same.
What was the point of releasing RV740 early and then selling 4850 instead of it because of 40G problems? There was no point in this. So NV didn't do it. That's completely OK in my book.
AMD can admit whatever it want. I have my own eyes to see what's really happened.
Well, 95 vs 98 probably wasn't the right comparision. It's probably better to compare Vista vs 7 to XP SP1 vs XP SP2 or something. We'll see what adoption rates 7 will have. But I have a feeling that they won't be much higher than that of Vista in the last couple of years.
With what? Last time I checked it was GT212 that they allegedely "fucked up" beyond the point of making it to market. So now it's suddenly not GT212, GT212 is fine and alright? Maybe they should go back to NV15 while they're at it?
You see, if you fucked up you go and fix it. You don't drop your next generation part in favour of a previous generation CANCELLED part. Cancelled so that the next generation part could be finished sooner, I might add.
Since GT212 never made it to tape out the amount of work to be done on it is the same (probably even more) as on any possible G3xx middle class part. The process is the same. So why would they ressurect GT212 instead of fixing what's fucked up with G300 chips? That simply makes no sense.
Well, I suppose there is one possibility: the ARCHITECTURE of G300 is what's fucked up so bad that they've decided to completely rework it and ressurected GT212 while they're doing it. I however find it hard to believe that NV would suddenly fuck up their next GPU architecture at the point of having first samples of GPUs built on it after having ~3 years to work on it. That's just as believable as NV quitting GPU business to concentrate it's forces on Tegra.
Redwood should be much slower than RV770, probably in the 4670 leagues of performance. I do believe that GT215 will be faster than 4670 because it certainly will be faster than 9600GT which is already faster than 4670.
"Doing something" as going back to a cancelled GPU on a previous architecture sounds like an NV hater fantasy to me, sorry.
Gotta admit, it makes sense.
DegustatoR
Legend
Unfortunately for all of us NV's latest decisions make no sense more often then they do. A year ago I would laugh my ass off to some bullshit like NV's cancelling G200 in favour of G75 on a 55nm. Now -- I don't know. Maybe they are really doing something that stupid? After all the renaming of late and general lack of serious progress in their GPUs since G80 I'm finding myself ready to believe that they've "ressurected" GT212 because they essentially has nothing against Cypress till 2Q 2010 and G300 is just so much faster (and more expensive) that it makes no sense for them to even try to put it against Cypress.
That's what i'm ready to believe, yes.
But if that'll turn out to be true I'm going to wonder what were NV's decision makers thinking when they decided to go with G300 first and G3xx middle GPU of ~Cypress specifications later. Because it was more than a year clear as hell that what NV will need against RV870 first is a middle class DX11 GPU. And GT212 cancellation was supposed to be the result of them understanding this situation and switching resources from GT212 to some "G302" chip. If it wasn't then I truly don't know what the fuck were they thinking back then.
Do you know how ridiculous that sounds? G300 is so good that nvidia can't market or sell it?
-FUDie
- Status