Core i7 860 @ 4ghz
MSI P55 GD65 4gb Gskill Ripjaw 2xAsus 5770 1003/5600 Corsair HX620 psu http://trust.hexus.net/user_profile.php?user=10950
Yea. I had a look at scans prices just after the launch and they were sitting at about £420 for the cheapest 7970. Half an hour or so later and I refreshed the page, suddenly the prices hiked up to about £460-£470 :S.
I only hope that they are more reasonable with the 7950 and the rest of the 7xxx range .
Ah.. this is the thread I was looking for.
Partial 7950 benchmark and OCing results:
http://forums.overclockers.co.uk/sho...php?t=18366789
Let have some wild conjecture and extrapolation, shall we?
Assuming roughly linear scaling of 40% @ max OC (and trusting Gibbo's figures!) the stock 7950 should do about 48fps in that unigine benchmark at 1080p, putting it only 5% behind that overclocked GTX580 3GB. Given the stock clock for a GTX580 is 772, v 902 on the clocked version, it looks like it will perform similarly to, perhaps even slightly better than, a stock GTX580.
Gibbo is sneaky.. no especially direct comparisons.
The 7950 OC to 7950 stock should be as you say seeing as the scaling should be roughly the same at the higher resolution. Maxed OC is 40% higher score than stock, so stock could score ~1227 at 1920x1080 4xaa 4xaf.
This gtx580 OC review may help provide scaling for the 580 OC (though the benchmark settings are a little different):
http://www.tweaktown.com/reviews/416...ew/index6.html
There a 935/5000 gtx580 gives 1835@1920x1200 while stock 772/4000 gives 1574@1920x1200. So a 21% increase in clock speed and 25% increase in mem speed gives a 17% increase in benchmark score. Gibbo's 902/4335 is 17% faster clock and 8% faster mem, so would give somewhere between an 14% (best case - fully GPU speed dependent) and 5% (worst - fully mem speed dependent) increase in the tweaktown bench over stock.
Assuming similar scaling in the gibbo test settings (main difference is 4xaa/4xAF, which these cards should be able to cope with at the similar but slightly lower resolution) then a stock gtx 580 would score between 1135 and 1232.
So unless my maths is wrong (quite possible), the stock 7950 should indeed be a smidgen faster given GPU speeds is probably more relevant than mem speed.
To summarise:
Unigene Heaven (*predictions) - 1920x1080 4xAA, 4xAF, normal tess, high shaders.
Stock gtx580 (772/4000 GPU limited prediction): 1135*
Stock hd7950 (810/5000): 1227*
OC'd gtx580 (902/4355): 1294
overdrive OC'd hd7950 (1025/6300): 1497*
Max OC'd hd7950 (1180/7200): 1718
Max OC'd hd7970 (1225/7800): 1844
I think the 580 scores are a bit too low, we'll see.
Last edited by kalniel; 25-01-2012 at 10:34 AM. Reason: added frequencies
Interesting looking card (and even more interesting if it unlocked) but its still way to expensive esoecially compared to the prices of the card it replaces (even original 6950 launch prices). Drops £100 and unlocks and it will be perfect
It does make me think though. I wonder if AMD is trying to make sure it can clear the HD6900 series stocks before dropping the price of the HD7900 series. I can't see this price being maintained too long, as it would mean the HD7000 series would be slightly better than HD6900 series level performance, while costing the same.
it depends entirely on how well nvidia kepler does, the 5850/70 didnt really budge at all because nvidia were performing terribly so hopefully nvidia can actually meet them on performance but then AMD will cut the prices as their dies are smaller (i think still?) so possibly cheaper than nvidia chips, eitherway it will be a win win for consumers when Nvidia arrives.
Supposedly CD from semi-accurate thinks that Kelpler has a smaller die and will be faster. He also things that Kepler will win against GCN even for generations after(at least this is the gist I get from one of his replies). How he can tell that from supposedly seeing one card,I don't know. It would be bad for AMD if true.
What is quite funny though is that he has not seen the other AMD cards,and it is most likely that the other GCN ranges outside the HD7900 series are likely to be relatively smaller, as they will not form the basis of the AMD compute cards.
OTH,with massive cuts in AMD marketing it could have affected his internal AMD sources too, although if he is using ones from the OEMs it would not change things.
im not sure is CD is taking the mick or not, im siding towards the former because he seems fairly peed off by the fact of plagiarism etc. Its hard to tell if any tech will be dominant for many generations, AMD are essentially in the same situation as Nvidia was with the 400 series, i.e its a substantial change from their normal design and yet so far amd have pulled out decent performance with great power effciency.
Nvidia get to improve their chip design a crap load so it wouldnt suprise me if Nvidia win this year but i equally expect AMD to completely decimate next year, same as their CPU division when pile driver is out.
TBH, I think it is prudent to wait until these cards are released and we can see the performance in reviews. It could be that they are the next 8800GT,HD4870 or HD5850 cards.
But there are a few issues though.
1.)Did CD see stock clocked cards or uber-overclocked edition ones??
Remember the 1GHZ GTX560TI at launch which disappeared soon after?? The same goes with those uber-overclocked edition GTX460 1GB cards which were hardly in stock for yonks.
2.)What games did he see the cards tested in? Was it something like 3DMark?? What settings and resolutions??
There are certain games which do very well on Nvidia cards and some not soo much.
For example,look at the Just Cause 2:
http://www.xbitlabs.com/articles/gra...0_7.html#sect0
The game is very AMD friendly and at low resolution,the HD7970 3GB is 63% faster than a GTX580 3GB.
OTH,look at SC2 with AA:
http://www.xbitlabs.com/articles/gra...0_7.html#sect0
The GTX580 3GB is much faster than an HD7970 3GB.
First it is last year they were meant to be released,then February or April this year.
If the GK104 was up and running why was there no demonstration at CES?? The secrecy would seem weird as they would not need to divulge performance or price.
Don't remind of the 8800GT, it physically to buy an 8800GTS for £180 then a month or so later they release that at around the same price .
I feel like it's inevitable that the best Nvidia card will outperform the 7970, purely because of the time difference of the release, if they don't release their fastest until may-ish, that will be a whole 4 months extra they will have to optimise their card.
I really want to see how the mid-range cards stack up against each other. The 78xx and the 660/650. Those kind of cards tend to get the bulk of the market.
Only if they've got working silicon already, *and* that silicon can provide the performance within cost and power constraints. Sometimes the product simply isn't up to it.
Look at the 6990 v GTX590. Nvidia had working silicon - it was using the same GPU silicon as it's well reviewed GTX580. And it had ~ 3 weeks from the launch of the 6990 to tweak its product, which was based on existing production silicon. But the power draw constraints of a consumer graphics card meant they had to heavily downclock the GPUs, and the GTX590 was slower than a stock 6990, whilst drawing significantly more power. Worse, the 6990 had an OC mode built in which made it even faster, and the power draw still only went up to matching the GTX590.
I've highlighted the fact that nvidia had good silicon for the GTX590 because it's kind of important: that was a proven, functional GPU. AFAIK no-one has (officially) seen Kepler running yet - no announcements or demonstrations at CES, no press breifings - just some leaked benchmark figures on a forum somewhere, and a couple of heat-spreader shots. I'm not doubting that the silicon exists, but so far we have no proof, and no indication of how close to being production-ready it is. And nvidia need to get that sorted before they can even start thinking about how they respond to the 7970...
There is another consideration too. Unless the 28NM process is more expensive and even has lower yields when compared to 40NM,AMD still has wiggle room for the HD7900 series. Tahiti itself is in-between Cypress and Cayman in size which makes comparable to both the GF104 and GF114.
That's a givenNo way of telling - there haven't been any complaints yet, but it could be either too early, or nVidia don't want to piss off the fabs at this stage.and even has lower yields when compared to 40NM
You mean budget wiggle room? Given R&D etc. I think actual manufacturing cost is pretty non-important for the HD7900 series - it's not a volume part. They charge what the market will bear regardless of actual cost - it's the high volume parts that will make the money, and having a high end part that is perceived as fast and efficient is more important in that respect than one that has a high margin.AMD still has wiggle room for the HD7900 series. Tahiti itself is in-between Cypress and Cayman in size which makes comparable to both the GF104 and GF114.
There are currently 1 users browsing this thread. (0 members and 1 guests)