NVIDIA GF100 & Friends speculation

So does anyone think that the GTX 570 will be a full GF104?

384SPs, 1GB, 256bit. It could go up against 6950 in the same way the 460 went up against the 5850. One would hope that if Nvidia are launching a full GF110 there aren't many harvestable parts and they have got the process working well.

It also leaves room for an easy introduction of a 768 SP part for a 28nm refresh, if Nvidia can squeeze another 5-6% of performance gains from architectural improvements we could see a very quick introduction of a GF200 with a 30-50% performance increase over GF110. All speculation of course, and we do have to bear in mind that Nvidia and quick don't really go hand in hand these days...
 
Ok... then its lower than 2x 6870 :p
Joking.. You are right :p
Still, it smokes the 5970 on quite some ocasions, and probably is much more future proof being a single GPU :)

But overall the 5970 is a tad ahead. But yes, that's what you get for having a dual GPU... you're more susceptible to differences in coding/drivers/optimization
 
Therefore, you can't really cross reference different reviews because their methods maybe different.
After having looked at some reviews, it looks though that power consumption really is very very similar. Might be just based on sample variance if it turns out lower or higher (though it is possible it's lower on average - after all the TDP number IS 6W lower). It's possible though power consumption is only slightly lower because the cooler does a better job and it runs not only quieter but cooler :). In any case, even with same power draw, that's still a 20% perf/power improvement.
 
So does anyone think that the GTX 570 will be a full GF104?

384SPs, 1GB, 256bit. It could go up against 6950 in the same way the 460 went up against the 5850. One would hope that if Nvidia are launching a full GF110 there aren't many harvestable parts and they have got the process working well.

If a full GF104 would be competitive with HD6950, it would a big failure on ATI part, IMO.
However, I believe we will rather see a GF114, with same type of improvements done on GF110. There must be a reason why there isnt a full GF104, just like there isnt a full GF100, and my bet is the problems are similar.
 
If a full GF104 would be competitive with HD6950, it would a big failure on ATI part, IMO.
However, I believe we will rather see a GF114, with same type of improvements done on GF110. There must be a reason why there isnt a full GF104, just like there isnt a full GF100, and my bet is the problems are similar.

Well that's what I was getting at, I mean the GF110 is essentially a GF100 with everything working and with a few architectural enhancements, so the GF114 would be the same story but with the GF104.

The 6950 is going to be a harvested part and if the performance of the 5850 is anything to go by it will be a disappointing product purely by performance standards. I think a GF114 could be competitive with the 6950, but we'll have to wait and see. Though one would hope the wait isn't too long, and Nvidia have been working on fixing GF100/GF104 simultaneously and launching quicker.
 
The 6950 is going to be a harvested part and if the performance of the 5850 is anything to go by it will be a disappointing product purely by performance standards. I think a GF114 could be competitive with the 6950, but we'll have to wait and see. Though one would hope the wait isn't too long, and Nvidia have been working on fixing GF100/GF104 simultaneously and launching quicker.

I agreed til this part... the 5850 was quite good for its performance and pricing bracket. Yes, it was significantly slower than the 5870, but it performed quite well, especially with overclocks

Of course, that ought to be a hint at what 6950 should slot in as... it can't slot too close to the 6870 so it has to perform a bit better than a 6870, and the 6970 needs to be faster still
 
Couldnt understand if you agreed or disagreed with my afirmation :p
If its the latter, the underlying architecture stay the same, didnt it? The GPCs and all that Jazz.. It was probably badly implemented on GTX480, though.
But architecture and it's implementation go hand in hand. If you come up with an architecture that is hard to manufacture and has problems, then it's broken. For the GF110 they rearranged stuff and we will have to wait and look at yield reports or avaialbilty in shops to make a judgment, if Fermi is now well and truly fixed.

That's neither here nor there. It all depends on how the measurements were taken. Also how the measurements are presented.
Also depends on individual chips the reviewers get. Pclab.pl measured the power consumption to be a few watts more than GTX480. They repeated the tests, contaced nVidia, they were suprised too.

While Guru3d got almost 30W more for GTX580. But you have to remember, they originaly got 250W for GTX480, the only ones in the whole internet. :)
 
Because some people know the numbers NV is promising and the actual allocation and/or production dates etc?


It took 1,5 month for that thing to become available.. and that revision had taped out in December already. That and a health Supply > Demand situation ascertains availability.

Newegg and overclockers uk seems to have plenty of stock at the moment. Those are the only places I checked and confirms what I thought was going to happen.
 
This is a good product but it is only mediocre overall. If Nvidia could have released this back in April or even before the end of 1H '10, it would have been a killer product. Unfortunately, with Cayman launching soon and Antilles to follow, I don't know how well GTX580 will succeed, unless Cayman doesn't even get close to it's performance targets.

Also, I haven't had time to read through HardOCP's review but is it really 20-30% faster than GTX480? The handful of reviews I have looked through it is ~15% in the vast majority of cases.
Edit- I just got a chance to look through their review, it is certainly an interesting one...
 
Last edited by a moderator:
Well that's what I was getting at, I mean the GF110 is essentially a GF100 with everything working and with a few architectural enhancements, so the GF114 would be the same story but with the GF104.
GF104 already supports full-speed FP16 filtering, so the performance delta would be likely a bit more conservative.
 
Well that's what I was getting at, I mean the GF110 is essentially a GF100 with everything working and with a few architectural enhancements, so the GF114 would be the same story but with the GF104.

The 6950 is going to be a harvested part and if the performance of the 5850 is anything to go by it will be a disappointing product purely by performance standards. I think a GF114 could be competitive with the 6950, but we'll have to wait and see. Though one would hope the wait isn't too long, and Nvidia have been working on fixing GF100/GF104 simultaneously and launching quicker.

Let's assume 5850 was "disappointing"...

6870 is already faster than this very "disappointing" board, which itself was quite a bit faster than a GTX460, so, accounting for the GF104 improvements included in the GF110, it'll probably be in the same ballpark as 6870.
 
If a full GF104 would be competitive with HD6950, it would a big failure on ATI part, IMO.
However, I believe we will rather see a GF114, with same type of improvements done on GF110. There must be a reason why there isnt a full GF104, just like there isnt a full GF100, and my bet is the problems are similar.
I'm not sure what GF114 could bring to the table. From the 3 known architecture improvements of GF110 over GF100, we know that GF104 already supports full speed FP16 filtering. I wouldn't be surprised if it supported the other two neither (especially the cache split part seems rather trivial). Of course, other tweaks might be possible instead.
Also, GF104 never suffered from the problems GF100 did - or at least not even close to the same extent. It had no troubles reaching "reasonable" clocks, and its power draw was not really that high given the chip complexity. It didn't have the same leakage issues neither, as evidenced by its low idle power draw. That nvidia didn't manage to release a full chip though is a mystery to me and it begins to be more mysterious (maybe they'll do it once channel is cleared of GF100 products).
I don't think a full chip GF104 can match Cayman Pro - you could get HD5870 performance level with high-clocked, full chip GF104, but I expect Cayman Pro to be at least a bit faster as otherwise it would be too close to HD6870.
 
That nvidia didn't manage to release a full chip though is a mystery to me and it begins to be more mysterious (maybe they'll do it once channel is cleared of GF100 products).
I don't think a full chip GF104 can match Cayman Pro - you could get HD5870 performance level with high-clocked, full chip GF104, but I expect Cayman Pro to be at least a bit faster as otherwise it would be too close to HD6870.

I think the main problem is still too many GTX470.
As far as Cayman Pro, I will be sorely dissapointed if it isn't around GTX480 peformance, give or take about 5%, especially if the ~$349MSRP is true.
 
On previous page:
IMG0029904.gif

img0029904.gif


http://www.hardware.fr/articles/806-4/dossier-nvidia-geforce-gtx-580-sli.html
 
I'm not sure what GF114 could bring to the table. From the 3 known architecture improvements of GF110 over GF100, we know that GF104 already supports full speed FP16 filtering. I wouldn't be surprised if it supported the other two neither (especially the cache split part seems rather trivial). Of course, other tweaks might be possible instead.
Also, GF104 never suffered from the problems GF100 did - or at least not even close to the same extent. It had no troubles reaching "reasonable" clocks, and its power draw was not really that high given the chip complexity. It didn't have the same leakage issues neither, as evidenced by its low idle power draw. That nvidia didn't manage to release a full chip though is a mystery to me and it begins to be more mysterious (maybe they'll do it once channel is cleared of GF100 products).
I don't think a full chip GF104 can match Cayman Pro - you could get HD5870 performance level with high-clocked, full chip GF104, but I expect Cayman Pro to be at least a bit faster as otherwise it would be too close to HD6870.

Why would a GF104 or newer version of it go against a Cayman Pro, Cayman Pro and XT's are AMD's top end GPU's, they would go againt the top two nV GPU's.
 
Back
Top