AMD Radeon HD 9970 Specifications Leaked – Twice as fast as GTX 780 (ChipLoco rumor)

[sushiwarrior]

Thank god non of AMD cards are compute focused, that would make them worthless for games!

FP64 rate of Hawaii is going to turn some heads for sure. Not sure what Firepro has planned for it, I wonder if there are thoughts of trying to make a Tesla competitor.

 

[iMacmatician]

FP64 rate of Hawaii is going to turn some heads for sure. Not sure what Firepro has planned for it, I wonder if there are thoughts of trying to make a Tesla competitor.

Is it true that Hawaii has a 1:2 DP rate as some have claimed?

And speaking of Tesla, maybe they're also going after Xeon Phi?

 

[poohbear]

Yea & i heard the MSI lightning version shoots out REAL lightning!!!! Finally!!!!!!!!!

 

[DiogoDX]

Is so funny to see that now overclock performance is the most important thing again.^_^

 

[RobertR1]

Is so funny to see that now overclock performance is the most important thing again.^_^

Well seeing how well my 780GTX OC's over a stock card, hell yeah! I'm all for knowing OC potential. It certainly matters to me.

 

[BoFox]

About the 780 being so much faster than 770/680 in some newer games, it shows how the sheer number of shader processors is actually being pushed harder by increasingly more demanding games in the future. This is generally the case with cards that have more SP-to-ROP/bandwidth ratio (and why we keep on seeing the ratio going up and up...).

Ok but now watch this. By Feb 2011, you could get 6950 unlocked for only $230 + $250 savings from not buying 580 provided 85% of the performance of the 580 for ~2 years before 7970 launched. Sell 6950 for $175 + $250 savings from not buying the 580 = $549 - $375 resale + savings ~ $125 to upgrade to 7970.

Once 7970 was overclocked it gave 50%+ more performance over 580 oc. Therefore, for the majority of competition between 6950 @ 6970 vs. 580, 6970 provided about 85% of the performance and then when it came to newer games, the amount of $ saved from not getting 580 + reinvesting it into a future 28nm GPU blew 580 away.

Therefore, the point for 480/580's longevity is not really that important imo since upgrading more often is the more preferred route. The same thing will happen with 7970GE vs. 780/R9 280. By the time next gen PC games come out, all 3 of these cards will be too slow but right now 7970GE has little trouble with most games since they are still PS360 console ports. The difference now is even more in favour of the 1Ghz 7970 since they are dropping to $280 which represents a $370-400 savings over a 780! You know how much $370-400 savings + 7970's resale value will get you in GPU power in the next 2 years? I bet 40-50% faster than 780 by the time next gen PC games hit and we really need the extra GPU power.

So well-said!

Ever since I used to spend up to $500 on video cards, I started to really think of the "Return of Investment" on the cards that I bought and how I could better manage the resale value for future purchases while not losing too much right before a predicted price drop. I quickly noticed that there are sweet "plateau" spots for most video cards that take a while to drop further past: $300 plateau, $200 plateau, $150 plateau, and $100 plateau, etc..

Perhaps you should become the official "financial analyst" for GPU enthusiasts in general! Like the stock market analysts.. ^_^

 

[Jaydip]

Aha.. That is a good thing. But when AMD did this with HD7000 they failed to release new arch and their driver team sux... logic!


Thank god non of AMD cards are compute focused, that would make them worthless for games!


If NV need more die space for the same performance then it's a point to talk about.
But perf/W Lol. That is used by people who try do defend NV if they are not able to beat 7970 in compute.

1.I don't remember saying anything like that.

2.Real professional people run quadro/firepro where even the top dog of AMD's firepro lineup is not a match for Quadro K5000.The Quadro K5000 is not a real successor for Quadro 6000, the real successor is coming in Q4 which will make the matter even worse.K5000 is Gk104 based while the K6000 is based on GK110.

3.As I said it is irrevelant how fast 7970 is in compute, when the real test comes it fall apart.If 7970 was enough for professionals they would not have bothered creating their firepro lineup.

 

[raghu78]

FP64 rate of Hawaii is going to turn some heads for sure. Not sure what Firepro has planned for it, I wonder if there are thoughts of trying to make a Tesla competitor.

sushiwarrior the hopes of AMD taking the GPU crown vanished when Matt Skynner said GK110 is nearly 30% larger than Hawaii wrt die size. with 430 - 440 sq mm AMD would find it difficult to match Titan on a clock for clock basis. I see AMD matching GTX 780 on a clock for clock basis and coming at USD 550 with BF4. I think they would sell very well .

 

[96Firebird]

Is so funny to see that now overclock performance is the most important thing again.^_^

Yup. Overclocking the 7970 was probably the best thing it had going for it.

 

[RussianSensation]

Well seeing how well my 780GTX OC's over a stock card, hell yeah! I'm all for knowing OC potential. It certainly matters to me.

He is pointing out the constant double standard on our forum. 460's and 780's overclocking counted big time but 7950/7970's overclocking and 6950's was deemed as "luck of the draw", "not guaranteed", etc. The green team always downplays any overclocking advantages from AMD but when their cards are overclocking very well (460/470, 780), then they start dropping NV OC vs. AMD stock benchmarks.

 

[The Alias]

He is pointing out the constant double standard on our forum. 460's and 780's overclocking counted big time but 7950/7970's overclocking and 6950's was deemed as "luck of the draw", "not guaranteed", etc. The green team always downplays any overclocking advantages from AMD but when their cards are overclocking very well (460/470, 780), then they start dropping NV OC vs. AMD stock benchmarks.

lol beat me to it, but it's funny how that happens .

 

[boxleitnerb]

I showed the hwc benchmark where the GTX 780 was running at same core clock and memory clock as GTX 770. so it had 50% higher bandwidth with 50% more cores and 50% more ROPs. the diff was still below 30%.if you say the GTX 780 was bandwidth bottlenecked then so was GTX 770.

Fair enough. Yet their result doesn't make sense, so I would be very very careful with their benchmarks. Especially when we have other reviews (PCGH, HT4U) that show values that make much more sense.

you took a stock GTX 770 with stock power target against a GTX 780 OC with power target maxed out. the stock GTX 770 could be throttling for all you know. thats such a bad comparison. max out power target and voltage. push to max clocks. if clocks are similar draw comparisons. btw hwc even verified that the card is not throttling.

Fixed clocks for both, no throttling. You can make a clock for clock comparison at many values (800, 900, 1000... MHz), not just "max clocks".

They say they "verified" it, but still they get a result that is highly unlikely. Other reviews I have linked to give much more probable results clock for clock. You should not just look at numbers but also question their validity and ask "does it makes sense"? Don't accept everything so easily. Thinking for yourself and being critical is key here.

you are argumentative. I repeat you are yet to show a single benchmark of GTX 770 and GTX 780 at 1.3 Ghz clocks in a game at same settings and prove the 50% perf diff. :whiste:

A clock for clock comparison is possible at multiple clocks, not just 1.3 GHz. Apparently you don't want to accept that and instead bang on that specific frequency. And as I said: The values you posted just don't make sense and other reviews contradict them.

Btw:
For a comparison at the same clocks with extreme OC you absolutely have to disable the boost or at the very least use a mod bios. On my watercooled Titans with a modded Bios I have set the power target to 130% and the maximum value I have observed during benchmarking is around 125%. And I was running "only" 1200 MHz@1.2V at 40-45°C which saves quite a bit of power compared to 1.3 GHz at higher temps. As far as I can tell, the reviews you linked to didn't use a mod bios and had a max power target of 115%. That may be sufficient in benchmark A, but not in benchmark B. Increasing memory speed to almost 7.5 Gbps adds another 5-10% to your power target usage compared to base clocks. So what I'm saying here is that I'm pretty sure that all those reviews with these extreme overclocks hit the power barrier to some degree or another. You would have to continuously monitor/log the clock values during the entire benchmarking process for all cards involved to arrive at a valid result.

 

[thilanliyan]

Yup. Overclocking the 7970 was probably the best thing it had going for it.

My 7950 seems to play games pretty well too, but that may just be luck of the draw. :awe:

 

[96Firebird]

My 7950 seems to play games pretty well too, but that may just be luck of the draw. :awe:

Nope, you're just delusional.

:awe:

 

[raghu78]

Fair enough. Yet their result doesn't make sense, so I would be very very careful with their benchmarks. Especially when we have other reviews (PCGH, HT4U) that show values that make much more sense.

k. here it is. your pcgh benchmark shows a 23% perf improvement in crysis 3 for a 30% clock increase from 902 to 1176 mhz. so 23% perf increase for 30% clock increase = 23 / 30 = 0.7666 scaling per Mhz clock increase

http://www.hardocp.com/article/2013/08/14/galaxy_geforce_gtx_780_hof_edition_review/5

the hardocp review shows a whopping 34% performance increase over a stock GTX 780. so for a roughly 400 mhz increase from 900 mhz to 1300 mhz (44.4% clock increase) we see a 34% perf increase. 34 / 44.44 = 0.765

so you see pcgh data and the hardocp show similar perf scaling. btw you know that GTX 770 perf was incorrect because driver versions changed GTX 780 performance from the 2 charts you mentioned.

I can understand your point because the benchmarks which pcgh ran were 4X MSAA while hardocp ran 2X SMAA. so you are showing cases where the GTX 770 is showing ROP limitations. but unfortunately at 2560 x 1600 the pcgh benchmark shows a 1175 mhz GTX 780 at 32 fps. so that setting is not really smooth on even a GTX 780.

anyway I can say that avg 50% diff is just not happening between GTX 780 and GTX 770 at same clocks. maybe 30 - 35%. the fact is GK110 does not have the perf per cuda core of GK104 which is a gaming focussed GPU with excellent balance of front end resources , cuda cores and ROPs.

 

[ocre]

k. here it is. your pcgh benchmark shows a 23% perf improvement in crysis 3 for a 30% clock increase from 902 to 1176 mhz. so 23% perf increase for 30% clock increase = 23 / 30 = 0.7666 scaling per Mhz clock increase

http://www.hardocp.com/article/2013/08/14/galaxy_geforce_gtx_780_hof_edition_review/5

the hardocp review shows a whopping 34% performance increase over a stock GTX 780. so for a roughly 400 mhz increase from 900 mhz to 1300 mhz (44.4% clock increase) we see a 34% perf increase. 34 / 44.44 = 0.765

so you see pcgh data and the hardocp show similar perf scaling. btw you know that GTX 770 perf was incorrect because driver versions changed GTX 780 performance from the 2 charts you mentioned.

I can understand your point because the benchmarks which pcgh ran were 4X MSAA while hardocp ran 2X SMAA. so you are showing cases where the GTX 770 is showing ROP limitations. but unfortunately at 2560 x 1600 the pcgh benchmark shows a 1175 mhz GTX 780 at 32 fps. so that setting is not really smooth on even a GTX 780.

anyway I can say that avg 50% diff is just not happening between GTX 780 and GTX 770 at same clocks. maybe 30 - 35%. the fact is GK110 does not have the perf per cuda core of GK104 which is a gaming focussed GPU with excellent balance of front end resources , cuda cores and ROPs.

I can totally agree with the last statement. The gk104 is a potent gaming chip. The architectures do differ in layout and there is always compromise. Pros and cons.

Scaling up performance is not easy. Many smaller chips have fantastic performance per watt, performance per core, performance per mm^2. But building on the architecture to more complex and larger designs usually does not scale perfect or linear.
Regardless, the gk110 still ends up a rather efficient chip. And Nvidia still has not got the max they could out of the design. There is still more that could be squeezed out, SMX, cuda cores, and TDP.

You must remember that the gk110 is different than the gk104. When you build up, often its more complex and you must also build out. Also with every design there is compromise and with the Gk110 having to serve multiple functions, there is no way to get around it. With a three SMX per GPC design vs two for the gk104, you should expect some differences. Just as i expect Tahiti and Pitcairn to not line up perfectly

 

[exar333]

Well, now that the next GPU from AMD is confirmed at 28nm, these 2x performance numbers can be thrown out the window. How about 10-20%? :/

http://www.anandtech.com/show/7342/amd-teases-next-enthusiast-gpu-still-28nm-23-smaller-than-gk110

 

[Elfear]

Well, now that the next GPU from AMD is confirmed at 28nm, these 2x performance numbers can be thrown out the window. How about 10-20%? :/

http://www.anandtech.com/show/7342/amd-teases-next-enthusiast-gpu-still-28nm-23-smaller-than-gk110

Unless GCN 2.0 has some serious efficiency gains I highly doubt it.

 

[exar333]

Unless GCN 2.0 has some serious efficiency gains I highly doubt it.

Clock speed bumps, maybe a few more hardware resources. I doubt it will be hugely impressive, but who knows. I am thinking it likely will be like NV's 4xx to 5xx upgrades.

 

[Elfear]

Clock speed bumps, maybe a few more hardware resources. I doubt it will be hugely impressive, but who knows. I am thinking it likely will be like NV's 4xx to 5xx upgrades.

Sure hope not. 15% is rather pathetic after almost two years (or 15 months since the Ghz edition).

 

[bystander36]

He is pointing out the constant double standard on our forum. 460's and 780's overclocking counted big time but 7950/7970's overclocking and 6950's was deemed as "luck of the draw", "not guaranteed", etc. The green team always downplays any overclocking advantages from AMD but when their cards are overclocking very well (460/470, 780), then they start dropping NV OC vs. AMD stock benchmarks.

There will always be different opinions on what matters most and you will see those from both sides that push the OC comparison, while others will not. What I hate is when people compare OC vs stock.

 

[el etro]

GPUdb contains new entries for the new series, check them there!

Sure hope not. 15% is rather pathetic after almost two years (or 15 months since the Ghz edition).

I agree with your argument, but i think that the new card will not have a great bump on perf/mhz, but will have a very much improved performance per watt this time. Maybe the new flagship consumes close to the same as the original 7970(IMO is a great win if they can win like this here).


-------------------------------------

Once we know the die size of the GPU, i will try new speculations about the new performances of aspects of the VGAs:

- Improved reference cooling, making them 10-20% less noisier, awesome temperature levels even on stress;
- Frame pacing performance not improved, but will have not as many driver problems like when 7900 launched;
- Card size 1 inch larger than 7970;
- Price starting with $550 for 9970 and $450 for 9950;
- 9970: Performance =~ Titan / || 9950: Performance =~ 780
- TDP 265/250W;
- Crossfire scaling improved(in fps meters);
- Compute performance increased 10% against GCN 1.1(Bonaire, HD 8000s) series.

 

[Granseth]

Isn't it reasonable to think that this update for AMD is mostly about getting an updated GCN out for the mobile space.

 

[rgallant]

I think what ever amd release's the first reviews will include a highly clocked gtx 780 somewhere[just showed up at the door so had to include it], but show the amd card at stock with beta drivers.
-show power draw on a stock 780 vs 97xx not the oc one that beats the amd card.
-game reviews shows nv beating amd 23fps vs 19fps in gaming.

 

[The Alias]

I think what ever amd release's the first reviews will include a highly clocked gtx 780 somewhere[just showed up at the door so had to include it], but show the amd card at stock with beta drivers.
-show power draw on a stock 780 vs 97xx not the oc one that beats the amd card.
-game reviews shows nv beating amd 23fps vs 19fps in gaming.

that's kind of what happened with the 6850 and 460 and if memory serves 7970 and 580