Hawaii Confirmed to be 28nm and 30% smaller than GK110

[3DVagabond]

What AMD really needs to do is fix their performance in Battlefield/Frostbite engine.

Right now NVIDIA's ~$250 GTX 760 matches AMD's ~$300 7970.

...and the leaked BF4 alpha benchmarks don't look very promising for AMD's supposed advantage as a DICE partner either. http://www.bf4blog.com/battlefield-4-alpha-gpu-and-cpu-benchmarks/

IMO, BF4 performance is the game that will sell cards this fall. Sure there are other games, but none will matter as much in terms of sales volume and actual need for a graphics upgrade.

Stock for stock the 7970 and 670 have been pretty close just about forever. It needs the higher clocks of the GHz edition to start to stretch it's legs. There's no where near as much O/C headroom on the 760 to keep up with a 7970 once you start to crank the clocks.

 

[nemesismk2]

Why is there so many charts which just show the performance for battlefield 3 which i don't own and have never owned any of earlier battlefield games either. Now Crysis is a game which i have owned and enjoyed all of them from Crysis to Crysis 3.

 

[raghu78]

True. I'm still waiting for it to lay a hurting (I think that's the term used) on the 780. At ~30% smaller it will have to be something really special to even match the 780. Unless they clock it up by 20% or so above the 780. Just my opinion, of course.

AMD has had 21 months to go back and look at efficiency. You are assuming GCN 2.0 is only as efficient as GCN 1.0. thats a very naive belief. from a perf/ sq mm perspective Hawaii will beat Tahiti and GK110. on perf per clock and perf/watt it willl match or exceed GTX 780. Titan might be slightly ahead. All of this will be clarified in a few weeks.

 

[3DVagabond]

Why is there so many charts which just show the performance for battlefield 3 which i don't own and have never owned any of earlier battlefield games either.

I think it's a fairly neutral game between the brands. It started off faster on nVidia and after some driver improvements AMD improved. It also doesn't seem to have any bugs that render it unusable in comparisons between the brands. Of course there are other games to consider as well.

 

[3DVagabond]

AMD has had 21 months to go back and look at efficiency. You are assuming GCN 2.0 is only as efficient as GCN 1.0. thats a very naive belief. from a perf/ sq mm perspective Hawaii will beat Tahiti and GK110. on perf and perf/watt it willl match or exceed GTX 780. About Titan it might be slightly ahead. All of this will be clarified in a few weeks.

I'm not naive. Titan isn't even a consideration since the 780 was released.

The 780 is the only highend card that matters. An O/C'd Titan can't keep up with an O/C'd 780 because the 780 has more TDP room and is available with better cooling. Once warmed up a stock Titan throttles. No gamer in his/her right mind is going to pay the extra for a Titan over the 780. Unlike Titan the 780 can reach clocks as high as anything AMD has shown us. So, it's going to have to beat the 780 clock for clock. Being ~30% smaller, that's not going to be easy. I realize the 780 is a twice culled card. Maybe that'll give Hawaii a chance. Maybe!

 

[Erenhardt]

Being ~30% smaller, that's not going to be easy. I realize the 780 is a twice culled card. Maybe that'll give Hawaii a chance. Maybe!

20% of GK110 is gone in gtx780. If hawaii is less than 30% smaller than gk110, then effectively it is almost as big as gtx780. If Hawaii really have the best performance/mm2 then we can expect it to put up a fight to GTX780

 

[3DVagabond]

20% of GK110 is gone in gtx780. If hawaii is less than 30% smaller than gk110, then effectively it is almost as big as gtx780. If Hawaii really have the best performance/mm2 then we can expect it to put up a fight to GTX780

30% of the CC's yes, but not 30% of the chip. 7970 has ~15% more shaders than the 7950 but is only about 7% faster at the same clocks. Shaders/CCs don't scale linearly.

If you read where I started from it was posts here that basically said Hawaii is going to put a hurting on GK110, not just be competitive. Competitive, very likely. Be appreciably faster (except in those compute functions where GCN simply obliterates Kepler), I don't think so. From this report it seems like this isn't anything more than we were expecting in January, but was pushed back. I remember AMD saying they learned a lot making Pitcairn and would use those lessons with their future designs. I don't know how much you can do in 2 years to change a design, but incorporating those lessons into Hawaii might have been what pushed it back.

I'm actually more expecting Hawaii to be a compute monster than a gaming monster. It's often said that the reason AMD can't compete with +500mm² chips is because they lack the market to sell them profitably. That market is HPC. They need to address that, IMO, and I think that's the direction they are pushing into. Just speculation, of course. I certainly have no inside info. Time will tell.

 

[raghu78]

I'm not naive. Titan isn't even a consideration since the 780 was released.

The 780 is the only highend card that matters. An O/C'd Titan can't keep up with an O/C'd 780 because the 780 has more TDP room and is available with better cooling. Once warmed up a stock Titan throttles.

on cost I can agree Titan makes little sense. But on perf I have to disagree. custom BIOS for Titan with increased power target and higher max voltage is out and many Titan users on OCN are running upto 1.3 Ghz with watercooling and no throttling.

No gamer in his/her right mind is going to pay the extra for a Titan over the 780. Unlike Titan the 780 can reach clocks as high as anything AMD has shown us. So, it's going to have to beat the 780 clock for clock. Being ~30% smaller, that's not going to be easy. I realize the 780 is a twice culled card. Maybe that'll give Hawaii a chance. Maybe!

again Titan can reach 1.2 - 1.3 Ghz with the proper BIOS and good cooling. Thats exactly what enthusiasts are running on Titan.

http://www.overclock.net/t/1406832/single-gpu-firestrike-top-30

look at the fastest Titan at 13304. the fastest GTX 780 at 10976. thats what a unlocked Titan is capable of.

 

[3DVagabond]

on cost I can agree Titan makes little sense. But on perf I have to disagree. custom BIOS for Titan with increased power target and higher max voltage is out and many Titan users on OCN are running upto 1.3 Ghz with watercooling and no throttling.



again Titan can reach 1.2 - 1.3 Ghz with the proper BIOS and good cooling. Thats exactly what enthusiasts are running on Titan.

http://www.overclock.net/t/1406832/single-gpu-firestrike-top-30

look at the fastest Titan at 13304. the fastest GTX 780 at 10976. thats what a unlocked Titan is capable of.

If you want to flash the bios and chance bricking your $1000 card and then add the expense of a custom waterloop. But that's a bit too niche to really call it a viable comparison. Especially with what the 780 can do out of the box with simple O/C'ing (Add some voltage and set a custom fan profile).

 

[imaheadcase]

Yah, even for the laymen buying a 780 o/c version is a no brainer than going Titan.

 

[Sohaltang]

If you want to flash the bios and chance bricking your $1000 card and then add the expense of a custom waterloop. But that's a bit too niche to really call it a viable comparison. Especially with what the 780 can do out of the box with simple O/C'ing (Add some voltage and set a custom fan profile).

No doubt. Most people have more money than brains and thats a lot of effort. You could prob buy 2 780's for the price of a titan with a block and water cooling loop. Plus the PITA of setup and tweaking. Im a bit of a tweaker but for this case I would just drop 1400$ and plop a pair of 780's into my rig over a titan with a water block.

 

[ICDP]

30% of the CC's yes, but not 30% of the chip. 7970 has ~15% more shaders than the 7950 but is only about 7% faster at the same clocks. Shaders/CCs don't scale linearly.

Tahiti was ROP limited and 7970 hit that wall harder/sooner than 7950. Adding more ROPs should give Hawaii a performance boost, add in the extra shaders and architecture refinements and we should end up with a card that competes with 780/Titan.

IMHO AMD will finally be able to compete on performance and give consumers a choice. This will of course help force prices to fall for Nvidia 780s.

 

[RussianSensation]

http://gamegpu.ru/action-/-fps-/-tps/battlefield-3-alpha-test-gpu.html

http://www.anandtech.com/bench/product/517?vs=509

All those benchmarks are before Catalyst 12.11s. BF3's performance got a 20-30% boost on AMD cards after the fix.

http://www.anandtech.com/show/6393/amds-holiday-plans-cat1211-new-bundle

What AMD really needs to do is fix their performance in Battlefield/Frostbite engine.

Right now NVIDIA's ~$250 GTX 760 matches AMD's ~$300 7970.

First, no one buys 925mhz 7970s anymore because for $280 you get a 1Ghz version. Secondly, you can't use 1 review that probably tested SP campaign as an indication of BF3's performance. What area of the game was tested? You can find sections where 7970 might be close to 760 but in real world multi-player on video a 7970GE easily beats a 680:
http://www.youtube.com/watch?v=qR3ewLMbywY

After that 6 hour campaign is done, you are playing MP, right?

It doesn't make sense to generalize BF3 the way you did. To make a more accurate assessment, we need to look at more reviews to see if AT's results are an outlier - they are.

7970GE > 770 in BF3 at 1440p

7970GE > 770 in BF3 in all resolutions/settings tested
http://www.computerbase.de/artikel/grafikkarten/2013/nvidia-geforce-gtx-760-im-test/26/

So I am going to call AT's review a 100% outlier on 760 matching a 7970 in BF3 at 1440P with MSAA based on the data from many other reviews on the net. This is a pure flop from AMD having an issue with Frostbite game engine for so long that now people are looking back to old benches/or the idea that AMD still sucks in BF3 stuck around.

on cost I can agree Titan makes little sense. But on perf I have to disagree.

How can we talk about performance but ignore price? Why would Titan be compared to R9 280X if the latter ends up at $550-600 and the former costs $1,000? I think it's only the most loyal NV users or wealthy folks that think it makes sense to start treating $550 and $1000 GPUs are "similarly priced." People who are going to be buying a $1,000 Titan + $100 waterblock and putting custom BIOS on it aren't cross-shopping the Titan with R9 280X. As has been mentioned, at that point might as well spend a little more and get 2x 780s over that single watercooled Titan. At OCN, you could easily find people who'd drop $1,500-2,000 on a hypothetical Titan II Ultra that was 20% faster than Titan I. Just like right now I don't think anyone expects a $250 760 to beat the R9 280X, why should a $550 R9 280X beat a $1000 Titan? It would be nice if it did, but the R9 should be aimed against after-market 780s imo. The Titan has an outrageous price that makes sense to very few.

 

[nitromullet]

All those benchmarks are before Catalyst 12.11s. BF3's performance got a 20-30% boost on AMD cards after the fix.

http://www.anandtech.com/show/6393/amds-holiday-plans-cat1211-new-bundle



First, no one buys 925mhz 7970s anymore because for $280 you get a 1Ghz version. Secondly, you can't use 1 review that probably tested SP campaign as an indication of BF3's performance. What area of the game was tested? You can find sections where 7970 might be close to 760 but in real world multi-player on video a 7970GE easily beats a 680:
http://www.youtube.com/watch?v=qR3ewLMbywY

After that 6 hour campaign is done, you are playing MP, right?

It doesn't make sense to generalize BF3 the way you did. To make a more accurate assessment, we need to look at more reviews to see if AT's results are an outlier - they are.

7970GE > 770 in BF3 at 1440p

7970GE > 770 in BF3 in all resolutions/settings tested
http://www.computerbase.de/artikel/grafikkarten/2013/nvidia-geforce-gtx-760-im-test/26/

So I am going to call AT's review a 100% outlier on 760 matching a 7970 in BF3 at 1440P with MSAA based on the data from many other reviews on the net. This is a pure flop from AMD having an issue with Frostbite game engine for so long that now people are looking back to old benches/or the idea that AMD still sucks in BF3 stuck around.

yep, already covered on page 1.
http://forums.anandtech.com/showpost.php?p=35503620&postcount=21

 

[BallaTheFeared]

look at the fastest Titan at 13304. the fastest GTX 780 at 10976. thats what a unlocked Titan is capable of.

Don't be fooled by inflated x6 scores.

 

[exar333]

Based on this news, I seriously think it will be better for folks to grab the 7970s NOW when the price is low. I agree that we will see these new cards at 10-20% better performance for 40-50% more money...

The bundles and prices now for AMD cards is pretty tough to beat, unless you just want performance. Then you get a GTX 780 or Titan IMHO.

 

[Elfear]

Based on this news, I seriously think it will be better for folks to grab the 7970s NOW when the price is low. I agree that we will see these new cards at 10-20% better performance for 40-50% more money...

The bundles and prices now for AMD cards is pretty tough to beat, unless you just want performance. Then you get a GTX 780 or Titan IMHO.

I highly doubt the 9970 (or R9 280X or whatever it's called) will only be 10-20% faster than the 7970Ghz. I'm betting it will match the 780 for a cheaper price.

Your first point still stands though, the price increase won't be linear with the performance increase. $280 1Ghz 7970s are a great deal, as are the $180 7950s.

 

[hawtdawg]

It will be roughly in the 780 range, maybe a tad faster after a few driver updates, and cost 499. That's my prediction.

 

[BallaTheFeared]

Admittedly I don't think AMD will beat GK110, however I do think they can bring the prices back down to normalcy.

GTX 780 is going up in price, not down right now with these new aftermarket cards. That's a trend AMD can halt rather quickly, even reverse.

I'd like to see the 780 at $500, and a 40 ROP GK110 at $370/$400 but that's just me being wishful it's not like I'd commit to either. Same node refreshes never appeal to me personally. I think this will just get AMD in the hunt, whereas now they're not competitive hence the redic prices from Nvidia.

 

[Vesku]

From a gaming perspective the 7950 is the sweet spot right now, 8 models at $230 or less on Newegg today.

I've personally been completely happy with a mildly OCed 7950 @ 1080P for almost a year now. I won't even consider a GPU change until 20nm cards show up and I'd certainly not recommend anyone get more than a 7970GHz/670 unless they are driving more pixels than 1900x1200.

The only interesting things to me about Hawaii will be efficiency and GPU compute, because that will signal what sort of 20nm parts AMD will be fielding next year.

But even at low 400mm2 size, which is what I expected from Hawaii, there is the potential for AMD to be offering a nice 780 alternative for those who run high resolutions. Those high resolution people who are locked into Nvidia can hope a competitive Hawaii will reduce 780 prices.

 

[seitur]

Admittedly I don't think AMD will beat GK110, however I do think they can bring the prices back down to normalcy.

GTX 780 is going up in price, not down right now with these new aftermarket cards. That's a trend AMD can halt rather quickly, even reverse.

I'd like to see the 780 at $500, and a 40 ROP GK110 at $370/$400 but that's just me being wishful it's not like I'd commit to either. Same node refreshes never appeal to me personally. I think this will just get AMD in the hunt, whereas now they're not competitive hence the redic prices from Nvidia.

Yep I agree.

On the other hand Nvidia users will have to actually start to choose to buy those competetive Radeons more often rather than hoping it will bring down prices of Nvidia cards they then will buy.
If consumers want to bring down inflated prices of GPUs generally that is.