News: The 7800gtx isn't really the G70....

[Tim]

Either way, I thought for sure Mesce was going to close the thread on me.

That just totally made my day ROFL!

[cadaveca]

Nowhere in there does it say the 7800gtx is the NV47, just that codenames of G70/G80 surfaced before the NV47 could be released. It should also be noted that NVidia stated the NV47 was cancelled completely, and therefore that makes the fact that it turns out the 7800gtx IS the NV47 news, and since no site has posted the truth, that makes this New News... Othersites only posted suspicions, I posted everything with proof.

Geez dude...relax...slow down and read my post:

This is not new news, but the proof and format is new. Well done Diltech! i had thought something was fishy when i got rivatuner yesterday....guess we can confirm it now.

[iS.Terminator]

weird thing to do

yeah.......

[MaxxxRacer]

this is fascinating.. can anyone with more "acess" confirm this story?

[magg]

Props for following through on this, Dil-tech.

I watched through your thread on EOCF, I must say it was pretty funny

I responded a few times, but it was more interesting to just watch. If this isn't the G70, then what should be the specs of it? 32pipes / 90nm / ___ shaders / etc...?

[DilTech]

Thanx magg, I've put quite a bit of work into all this, I'm definitely gonna ignore anyone who doesn't seem to get the whole point of this.

The way I see it , I stumbled onto something that a few sites only hinted at, but no one ever got really indepth. I'm also stuck sittin here burning cds all day for a local dj, so I've got nothin better to do than look things up while I sit here and change cds out, thus I was curious... This time, curiousity didn't kill the cat, but rather got it stuck chasin a mouse thru a cheese maze... Hopefully soon enough we'll find that rodent... I'm gonna talk to my "source" when I go work at my sidejob next week, if anyone can tell me something, it's him...

Either way, where's shamino or OPP... they'd probably know something about all this!

[RaZz!]

Is it not possible that some components of the cancelled NV47 were carried over to the G70? I mean, they had some good engineering obviously, why start from scratch? I think the G70 IS totally based off the NV47 core that was never actually built. Why change what the chip is reported as in development tools? It's probably too much work for no real gain considering they were half way done when they stopped.

That makes logical sense, moreso in my opinion than than having unused technology lying around, the board of directors would fire them all for wasting bucks otherwise.
UnG

that's wat i thought, too.

after i got my 7800gt i was wondering why rivatuner reports it to nv47. but there are many people out there who say, that the previous name of the g70 was nv47.

[DilTech]

that's wat i thought, too.

after i got my 7800gt i was wondering why rivatuner reports it to nv47. but there are many people out there who say, that the previous name of the g70 was nv47.

Read thru my original post, it's not just rivatuner, it's their NVidia's own development tools too. The codename NV47 is hardcoded into the supposed "G70" core, the specs match the rumored "NV47" specs, and we have a korea stamp on a core that's suppose to have been produced by taiwan fabs run by TSMC...

[speed bump]

Hmm interesting, I wonder if its possible though that the NV47 had a weird defect and Nvidia decided to rengineer it to the g70 and then someone in the coding department forgot to update it to g70 also. A good way to check is if this blows up huge will these should start being reported as g70s in rivatuner becuase Nvidia wants to cover there engineering department.

[cadaveca]

Well, i'll tell ya what...15.6 reports my core as G70....

[Tim]

coding department forgot to update it to g70 also.

Highly unlikely

[P_1]

That is a lowball trick if it is right. This is why I buy last-gen cards becuase of scandals like this.

i dont see this as low at all, why would ati need to look at what nvidia is doing to compete with them. They should be making as much advances as they can instead of looking at what advances they can cut out ala sm3.0 and fp32 hdr.

[Kunaak]

dumb question...

but who cares?

I mean, they can call it a R521 ATi2 for all I care.

it has 24 pipelines, it has 8 VS, its clocked high.
performs high...
it's a definate leap over the 6800 series...

it's a definate "next generation" card.
so, why would it matter what they call it??

[Turok]

Wait a sec... Hmmmm

If its based on the nv45 and the nv47 is a improved nv45 (6800 Ultra) with more pipes and stuff, then thats why the texture quality problem is simmilar to that of the 6800 Ultra when first released.

BTW, didnt the nv45 have problems with texture compression in games like Doom3 and FarCry (They looked less compressed) and was the main cause of why the nv45 had an advantage in these games?
nVidia sacrificed image quality for performance, wile ATi stayed with image quality and performed simmilar or even better.

The r520 may have a chance here. Maybe afther they fix the High res problem, at high resolutions the 7800 may lose performance and the r520 may outperform it at the same image quality

That reminds me how the Prescotts back then had more pipelines than the FX and because of the excessive pipelines they were slower.
Maybe more pipelines isnt better, and ATi is doing the right thing by sacrificing the expensive pipes for more clocks.



I got an idea:

Has anyone tryed testing Pipes vs Mhz?
How large was the jump by unlocking 12 pipes to 16 pipes on the x800 pro?

Could anyone compare the following:

7800 GTX *speed:
Clocked at 450/1200 with 24 pipes
--------------------Vs--------------------
Underclocked at 200/800 with 24 pipes

AND

7800 GTX pipes:
Clocked at 450/1200 with 24 pipes
--------------------Vs--------------------
Underclocked at 450/1200 with 16 pipes



Compareing these *speed would be like compareing a nv47/"g70" 450/1200 vs a r520 possibly 700/1400

IF the gap between 200/800 Mhz and 450/1200 Mhz is greater in % performance than the gap between the 16 and 24 pipes, THEN the r520 may have a chance of winning.

If someone is going to try this, please make a new thread about it

[Vapor]

Me thinks G70 is a renamed NV47 (of which they already had a bunch produced) and was considered good enough (and different enough) to be renamed in a completely different series. G71 (the short-rumored NV48) will be a revamp of this on 90nm with an unknown number of pipes while G80 is the old NV50. No scam here, just codename renaming and resorting.

I agree with other people here in saying that this was done because of ATi's delays.

[Vapor]

I got an idea:

Has anyone tryed testing Pipes vs Mhz?
How large was the jump by unlocking 12 pipes to 16 pipes on the x800 pro?

Could anyone compare the following:

7800 GTX *speed:
Clocked at 450/1200 with 24 pipes
--------------------Vs--------------------
Underclocked at 200/800 with 24 pipes

AND

7800 GTX pipes:
Clocked at 450/1200 with 24 pipes
--------------------Vs--------------------
Underclocked at 450/1200 with 16 pipes



Compareing these *speed would be like compareing a nv47/"g70" 450/1200 vs a r520 possibly 700/1400

IF the gap between 200/800 Mhz and 450/1200 Mhz is greater in % performance than the gap between the 16 and 24 pipes, THEN the r520 may have a chance of winning.

If someone is going to try this, please make a new thread about it

I'm not going to sort through what you said--it makes little sense and I'm quite sober right now--but I know my 7800GTX was completely unstable below 320MHz, this is likely true for others.

I don't follow your math either, why such a HUGE clock drop? Why a memory clock drop? Are you just spitting out random numbers?

EDIT: and what does this have to do with a card that we have little information about other than the name (R520/X1800)???

[DilTech]

NV48 was only a 16 pipelined part, it was the refresh to the 6800nu, not a higher end NV47, despite the massive confusion that was caused... My guess is that the G71 is actually what the G70 was originally suppose to be, as NVidia saw an opening where they could simply release the NV47 and make massive sales doing so, as ATi wouldn't have a counter for almost a full quarter.. Looks like it paid off.

A for Kunaak, the reason this actually matters is this. The word was the NV47 and G70 were being worked on Side-by-Side, which there-fore would mean that since this is the NV47, the G70 still exists in another form. Likely in the form of the ultra, the 90nm part rumored to be 32pipes....

Now, we all heard the news that "the ultra was cancelled", but then again, so was the NV47, yet we see it here today. This is ALOT of work to simply make the NV47 look like the G70, now isn't it?... Something is going on in NVidia land, the question is what...

[Mesce]

That reminds me how the Prescotts back then had more pipelines than the FX and because of the excessive pipelines they were slower.
Maybe more pipelines isnt better, and ATi is doing the right thing by sacrificing the expensive pipes for more clocks.

Its nothing like cpu's what so ever. Basicly a cpu has one pipeline, on that pipeline it has so many stages, the Prescotts have a load of stages somewhere in the 20's which allows for massive clock speeds but it takes longer delivering them. The Amd is a lot more efficient, a lot lower stages and now we are seeing higher clocking Amd's into 2.8 and soon 3ghz with the fx-59.

[cadaveca]

um...heh.


G70 codename=Geforce 7 series. It's may just simply be another name. The "G71" is most likely the Ultra, 32PP,10VS. This chip is the NV50, which was rumoured to be cancelled way back in december/january, like the NV47. At the time, The first R520 units started floating around looking for cooling solutions, and nVidia found that they needed to bring something faster than the NV47/NV50, and they scraped the NV48 @ the same time...remember the 512mb 6800Ultra's? those were NV48.

Here's where it gets strange...to me it seems as though nVidia paid TSMC to fubar ATI's chips, so that Nvidia would have time to finalize thier 90nm designs.

A wild accusation, i know, but with ATi moving to new fabs recently, giving them easy projects like mobile phone chips, it seems to me that they are testing the waters looking for a replacement for TSMC.

nVidia has shyed away from TSMC with final production of chips due to lack of security, i think. They have always been tight about thier products, but alot of info has been leaked in the last 18 months or so. The most obvious source would be someone related to the goings-on at that level, and TSMC to me is most obvious becasue of relations to both companies, as well as the major CPU designers.


Anyway, becasue of the leaks, nVidia have taken the new chips elsewhere, and becasue of the loss, TSMC have been very tight-lipped about ATI's current project...possibly chastised by the nVidia move, and maybe also becasue of thier problems in producing the ATI chips.


But hey, what do I know?

[Turok]

I'm not going to sort through what you said--it makes little sense and I'm quite sober right now--but I know my 7800GTX was completely unstable below 320MHz, this is likely true for others.

I don't follow your math either, why such a HUGE clock drop? Why a memory clock drop? Are you just spitting out random numbers?

EDIT: and what does this have to do with a card that we have little information about other than the name (R520/X1800)???

I was comapreing Mhz vs Pipes
Was just saying that if you grap a 7800 GTX and compare 16 pipes less than stock and 250 more/less Mhz than stock, and the performance difference is grater with Mhz than pipes, then the r520 could be faster.

If the 7800 GTX cant be clocked that low, then someone could try compareing half the performance loss or gain
Ex:
Comapre 450/1200 to 575/1300 vs 20 pipes to 24 pipes
Then you would compare the difference in performance between stock and 4 pipes less and 125/100 Mhz less or more than stock

and I know... its a bit off topic because its not related to nv47 and g70 names. Just a thought

[Entity_Razer]

The Pipes vs Mhz difference was tested here a while ago over the same thing (ATi's next gen GPU ) and it seemed clocks rule over pipes IIRC.

So Ati could still beat the NV47/G70 thingy if it appears to have a greater gain when clocked high with less pipes then a lot of pipes clocked lower. Also more pipes = more bandwidth needed to feed the pipes, more MHZ also = more bandwidth needed but not as much because more MHZ can also mean more "normal" calculations like volumetric fog, pixel shaders, vertex shaders etc. While more pipelines = more bandwidth needed because of the increase in texture demand. (and overall data demand also off course)

BTW concerning this G70= NV47 thing....

I don't carre actually... I mean same for ATI they can call their R520 R895472264542484X1337 for all I care, as long as it performs.
BUT
off course one does wonder, if this is the NV47, what happend to the G70. THe G70 = G80? or.....
I mean the G71 could just be a respin of this core to 90nm allowing higher clockspeeds, better cooling .....

Off course, say they DID scrap the NV47, they where halfway someone said. This means they had the main archtitecture completed, including the major components but not "small" things that where to be implementad in a later stage.
Now if you decide to make a completly new GPU then why waiste a architecture already done?
I mean they can still use the "backbone" of the existing archtecture they had and just builth further on that but off course making fundamental adjustments that would allow for the new features/ justify the new name/build/core version.

[Vapor]

I was comapreing Mhz vs Pipes
Was just saying that if you grap a 7800 GTX and compare 16 pipes less than stock and 250 more/less Mhz than stock, and the performance difference is grater with Mhz than pipes, then the r520 could be faster.

If the 7800 GTX cant be clocked that low, then someone could try compareing half the performance loss or gain
Ex:
Comapre 450/1200 to 575/1300 vs 20 pipes to 24 pipes
Then you would compare the difference in performance between stock and 4 pipes less and 125/100 Mhz less or more than stock

and I know... its a bit off topic because its not related to nv47 and g70 names. Just a thought

I know what you were trying to do....but there are way too many flaws in your statements and logic, I'll leave my post at that.

[Turok]

The Pipes vs Mhz difference was tested here a while ago over the same thing (ATi's next gen GPU ) and it seemed clocks rule over pipes IIRC.

So Ati could still beat the NV47/G70 thingy if it appears to have a greater gain when clocked high with less pipes then a lot of pipes clocked lower. Also more pipes = more bandwidth needed to feed the pipes, more MHZ also = more bandwidth needed but not as much because more MHZ can also mean more "normal" calculations like volumetric fog, pixel shaders, vertex shaders etc. While more pipelines = more bandwidth needed because of the increase in texture demand. (and overall data demand also off course)

BTW concerning this G70= NV47 thing....

I don't carre actually... I mean same for ATI they can call their R520 R895472264542484X1337 for all I care, as long as it performs.
BUT
off course one does wonder, if this is the NV47, what happend to the G70. THe G70 = G80? or.....
I mean the G71 could just be a respin of this core to 90nm allowing higher clockspeeds, better cooling .....

Off course, say they DID scrap the NV47, they where halfway someone said. This means they had the main archtitecture completed, including the major components but not "small" things that where to be implementad in a later stage.
Now if you decide to make a completly new GPU then why waiste a architecture already done?
I mean they can still use the "backbone" of the existing archtecture they had and just builth further on that but off course making fundamental adjustments that would allow for the new features/ justify the new name/build/core version.

Thanks for the reply
Where was it proven so I can check the performance difference between the pipes and Mhz?

Back to nv47/"g70"....
I also dont care how its called as long as it performs better.
I do care that ATi survives tho, because more competition = better cards for both sides.

Its still a good find tho

[Cooper]

DilTech nice work
But as it was sad here before the G70 and NV47 are pretty similar (if not the same). IMO it`s just marketing move to attract customers. And now when nVidia released this new product (not revolutional, but evolutional) it has plenty of time to make something really new.

[n00b 0f l337]

Hmmmm so if what ive read is true it really explains r520 being pushed and delayed and nvidia may just have had the same problems... i geuss its just sorta hard to get a huge leap ahead on competition.