DT: Next-generation 28nm GPUs Could Be 45 Percent Faster

[Termie]

Lol, did you read the link you posted?

12.5 > 25.4 = 100%
10 > 30 = 200%

The 4870 completely obliterated the 3870, it was over 100% faster in a lot of scenarios, and nowadays even more so

Actually, I made a mistake and posted the same page twice. Here's the 3870->4870 link I meant to post:

http://www.anandtech.com/show/2556/13

What interesting is that by the time of the 5870, the gap between the 3870 and 4870 had widened dramatically.

Here's the later benchmark again: http://www.anandtech.com/show/2841/17

The difference was the use of AA. They didn't use it in earlier benchmarks, but in later benchmarks it became standard. The 3870 basically couldn't run it.

So this actually helps clarify one issue: new cards often have enhanced "pipes" that can be utilized to do things that just couldn't be attempted on older cards. But in an apples-to-apples comparison of older games or older standards, the increased power doesn't show up as dramatically.

What does this all mean? When these new 7000-series cards come out, we're unlikely to see more than a 50% gain in the games we play today, at the settings we commonly use today. Cards were once benchmarked at 1280x1024. Now they're benchmarked at 5760. They were once benchmarked at 0xaa/4xAF, now it's often 4 or 8xAA, plus 16xAF, plus HBAO, plus blur, plus hi-res textures, etc. We don't buy new cards to play our old games. We buy new cards to play new games, or to implement features that didn't exist a few years ago. And that's where major breakthroughs are made in the quality of gaming.

 

[Medu]

3870->4870 is a poor example as the 3870 was basically a 2900XT on 55nm and the 2900XT was a very troubled chipped. IIRC it had to do AA in the SP. Also the SP were very space inefficient and were redesigned for the 4000 series which enabled ATi to increase the number of SP and TU by 2.5X while only increasing the die size by about 30%.

 

[Idontcare]

No, as others have said the 45% refers to the transistors, and your math is off.

I'm sure IDC will chime in about the importance of dynamic power consumption vs idle power consumption/leakage (which is what they are likely referring to when they state a 45% improvement based on the same leakage per gate).

I tried but I think it might have been lost in translation

 

[Idontcare]

Dynamic power should decrease too, because capacitances get smaller and less current flows when switching occurs.

Of course, that goes without saying, but what is notable is that they intentionally elected to not mention anything about it.

Talking up the static leakage improvements while entirely neglecting to mention the dynamic power consumption improvments (however great or meager they may be) in a conversation regarding clockspeeds advancements node-over-node is telling in its own right.

 

[Spjut]

I'd say there are very few, really: http://www.anandtech.com/bench/Product/176?vs=162



Another example of a crippling bottleneck would be the inability to use a new DX version - again, the 4870/4890 (and the GeForce 200 series) are held back in many ways, both in quality and performance, by being limited to DX10 in BF3.

I had thought that the HD4000 series, with their DX10.1 support, would hold out better than their Nvidia DX10 counterparts..Certainly doesn't seem to be the case with BF3 however

 

[Golgatha]

I'd pay good money for a single card that's 45% faster, has 2x the memory of current cards (or at least 2GB), and runs in the same thermal specs as the current generation.

 

[flopper]

I'd still go with the top single gpu being ~ 10-20% slower than 6990.

if so, 2 cards will be pretty decent in eyefinity.

 

[LiuKangBakinPie]

Actually, I made a mistake and posted the same page twice. Here's the 3870->4870 link I meant to post:

http://www.anandtech.com/show/2556/13

What interesting is that by the time of the 5870, the gap between the 3870 and 4870 had widened dramatically.

Here's the later benchmark again: http://www.anandtech.com/show/2841/17

The difference was the use of AA. They didn't use it in earlier benchmarks, but in later benchmarks it became standard. The 3870 basically couldn't run it.

So this actually helps clarify one issue: new cards often have enhanced "pipes" that can be utilized to do things that just couldn't be attempted on older cards. But in an apples-to-apples comparison of older games or older standards, the increased power doesn't show up as dramatically.

What does this all mean? When these new 7000-series cards come out, we're unlikely to see more than a 50% gain in the games we play today, at the settings we commonly use today. Cards were once benchmarked at 1280x1024. Now they're benchmarked at 5760. They were once benchmarked at 0xaa/4xAF, now it's often 4 or 8xAA, plus 16xAF, plus HBAO, plus blur, plus hi-res textures, etc. We don't buy new cards to play our old games. We buy new cards to play new games, or to implement features that didn't exist a few years ago. And that's where major breakthroughs are made in the quality of gaming.

the newer cards had the bandwidth advantage which was added for dx 11. When allowed to leverage its DX11 advantage, Cypress actually manages to reach the vaunted performance doubling goal that was supposedly established for it during the development cycle...at least when AA is disabled. This is encouraging in the perspective of future DX11 titles, and how they'll treat owners of DX11 hardware.

Cypress ends up being about 50% faster than RV790 at stock clocks in games. Slicing away the memory bandwidth advantage and the gap narrows down with AA&AF, sitting at a less comfortable 40%.

 

[RussianSensation]

New rumors are rolling in:

HD7000 series high end will have 384-bit memory bus and possibly 3GB of VRAM. Preliminary pricing is $500 for the Tahiti XT (January 2012 launch date) and $400 for Tahiti Pro (February 2012).

Could it be that ATI's old pricing structure is back (i.e., once again gunning for single-GPU performance crown)? $500 high-end single GPU may be packing serious heat

Mid-range cards are expected to follow March-April 2012, priced at $200-300.

http://semiaccurate.com/2011/11/17/more-bits-on-hd7000northern-islandsgcn-leak/

 

[tviceman]

New rumors are rolling in:

HD7000 series high end will have 384-bit memory bus and possibly 3GB of VRAM. Preliminary pricing is $500 for the Tahiti XT and $400 for Tahiti Pro.

Could it be that the ATI of old is back (i.e., once again gunning for single-GPU performance crown)? $500 high-end single GPU may be packing serious heat

http://semiaccurate.com/2011/11/17/more-bits-on-hd7000northern-islandsgcn-leak/

Charlie's last article was about how much of a failure Tegra3 is going to be (eerily familiar to his articles during and about GF100, GF104, and nvidia's financials), and how in trouble Nvidia is in. So, in other words, he is still a sucky unreliable source of information.

 

[Grooveriding]

New rumors are rolling in:

HD7000 series high end will have 384-bit memory bus and possibly 3GB of VRAM. Preliminary pricing is $500 for the Tahiti XT (January 2012 launch date) and $400 for Tahiti Pro (February 2012).

Could it be that ATI's old pricing structure is back (i.e., once again gunning for single-GPU performance crown)? $500 high-end single GPU may be packing serious heat

Mid-range cards are expected to follow March-April 2012, priced at $200-300.

http://semiaccurate.com/2011/11/17/more-bits-on-hd7000northern-islandsgcn-leak/

Ooof $500, could be harkening back to the X800XT-PE days. :thumbsup:

Not good for us, unless that $500 means some serious oomph.

If they can get 45% higher clockspeeds in tandem with whatever the architecture brings, would be nice.

I'd like to see a big jump for once again, meaning 7970 is twice as fast as 6970 or very close to it.

 

[Silverforce11]

There's plenty of rumors coming from AIBs that Tahiti XT is ~= 6990 performance.

Not going to bother speculating or believing it til its ready for shipping though.

 

[tigersty1e]

New rumors are rolling in:

HD7000 series high end will have 384-bit memory bus and possibly 3GB of VRAM. Preliminary pricing is $500 for the Tahiti XT (January 2012 launch date) and $400 for Tahiti Pro (February 2012).

Could it be that ATI's old pricing structure is back (i.e., once again gunning for single-GPU performance crown)? $500 high-end single GPU may be packing serious heat

Mid-range cards are expected to follow March-April 2012, priced at $200-300.

http://semiaccurate.com/2011/11/17/more-bits-on-hd7000northern-islandsgcn-leak/

If that site is wrong on this info, I'm never going there again and will bash it forever.

 

[Skurge]

New rumors are rolling in:

HD7000 series high end will have 384-bit memory bus and possibly 3GB of VRAM. Preliminary pricing is $500 for the Tahiti XT (January 2012 launch date) and $400 for Tahiti Pro (February 2012).

Could it be that ATI's old pricing structure is back (i.e., once again gunning for single-GPU performance crown)? $500 high-end single GPU may be packing serious heat

Mid-range cards are expected to follow March-April 2012, priced at $200-300.

http://semiaccurate.com/2011/11/17/more-bits-on-hd7000northern-islandsgcn-leak/

Hmmm, when is Ivy Bridge coming out again? Cause if those cards launch in good quantities and ivy bridge is close behind, I can see a lot of IB systems with these cards being sold.

 

[Coydog]

The upcoming vid cards have gotten me curious, and as such, held off and went and grabbed a SSD for my system.

 

[dangerman1337]

New rumors are rolling in:

HD7000 series high end will have 384-bit memory bus and possibly 3GB of VRAM. Preliminary pricing is $500 for the Tahiti XT (January 2012 launch date) and $400 for Tahiti Pro (February 2012).

Could it be that ATI's old pricing structure is back (i.e., once again gunning for single-GPU performance crown)? $500 high-end single GPU may be packing serious heat

Mid-range cards are expected to follow March-April 2012, priced at $200-300.

http://semiaccurate.com/2011/11/17/more-bits-on-hd7000northern-islandsgcn-leak/

So Pitcairn will be CGN arch as well? I hope it'll do slightly better than GTX 580. If the performance figures for Thaiti XT (6990-ish performance) are true.

 

[Olikan]

Charlie's last article was about how much of a failure Tegra3 is going to be (eerily familiar to his articles during and about GF100, GF104, and nvidia's financials), and how in trouble Nvidia is in. So, in other words, he is still a sucky unreliable source of information.

for me atleast, when it comes to nvidia charlie usaully nail it (and with a lot of hate, don't know why). But when it comes with ati he screw up alot

anyway, I take the 384-bit memory bus rumor anyday than xdr2 memory BS

 

[boxleitnerb]

There's plenty of rumors coming from AIBs that Tahiti XT is ~= 6990 performance.

Not going to bother speculating or believing it til its ready for shipping though.

To talk about 6990 performance is quite sketchy. At peak efficiency (100% scaling) or what? This is not always reliable as we all know because scaling can suck at times.

 

[Idontcare]

If that site is wrong on this info, I'm never going there again and will bash it forever.

Addicts always say they can quit whenever they want to, they just don't want that day to be today

 

[lavaheadache]

Actually, it wasn't even close. Despite the common perception that most generations are a 100% jump, they are usually closer to 50%:

HD3870 -> HD4870: http://www.anandtech.com/show/2556/13. 60 percent.

HD4870 -> HD5870: http://www.anandtech.com/show/2841/17. 50 percent.

This is depite the fact that the 5870 had twice the theoretical computational power of the 4870 (in shaders).

The new chip would be just about on target at 45%. Here's a thread on the topic: http://forums.anandtech.com/showthread.php?t=2200297

I looked at a few of the bechies you linked to and saw quite a few that were 100% or so improvements.

 

[flopper]

There's plenty of rumors coming from AIBs that Tahiti XT is ~= 6990 performance.

Not going to bother speculating or believing it til its ready for shipping though.

if true, it be a nice single card to own
still saving for one

 

[crisium]

Let's pretend they do bluntly mean "45% faster". Though I do believe they might mean clock frequencies - forget that for the moment.

1920 res used as it's the most common. Let's look at the new generation vs the top dog of the last.

First card of new gen vs old: GTX 280 over 8800/9800 GTX by 48%
Fastest card of new gen vs old: GTX 285 over 8800 GTX by 59%
http://www.techpowerup.com/reviews/Leadtek/GeForce_GTX_285/26.html

First: Radeon 5870 over GTX 285 by 20%
http://www.techpowerup.com/reviews/ATI/Radeon_HD_5870/30.html
Fastest: GTX 480 over GTX 285 by 41%

First: Radeon 6970 = GTX 480
http://www.techpowerup.com/reviews/HIS/Radeon_HD_6970/29.html
Fastest: GTX 580 over GTX 480 by 18%
http://www.techpowerup.com/reviews/ASUS/GeForce_GTX_580_Direct_Cu_II/24.html

If the first new top card is 45% faster than the 580, than that's a performance gain we haven't seen since the GTX 280 hit. Even if we take 45% as the "final top card", then 45% is still in line from 480 over 285.

Not sure if the "only 45%" complaints are valid at all. If true, it will be better than the most recent two generations (way better than the current).

Not sure why there is the 4870 vs 5870 thing going on, as the 4870 was never the fastest card at any point in time anyway. You gotta compare top dog to top dog, imo. I know not every generation had a nm shrink, but a new generation is a new generation nevertheless.

Just a fyi, take whatever you want from this.

 

[LiuKangBakinPie]

Let's pretend they do bluntly mean "45% faster". Though I do believe they might mean clock frequencies - forget that for the moment.

1920 res used as it's the most common. Let's look at the new generation vs the top dog of the last.

First card of new gen vs old: GTX 280 over 8800/9800 GTX by 48%
Fastest card of new gen vs old: GTX 285 over 8800 GTX by 59%
http://www.techpowerup.com/reviews/Leadtek/GeForce_GTX_285/26.html

First: Radeon 5870 over GTX 285 by 20%
http://www.techpowerup.com/reviews/ATI/Radeon_HD_5870/30.html
Fastest: GTX 480 over GTX 285 by 41%

First: Radeon 6970 = GTX 480
http://www.techpowerup.com/reviews/HIS/Radeon_HD_6970/29.html
Fastest: GTX 580 over GTX 480 by 18%
http://www.techpowerup.com/reviews/ASUS/GeForce_GTX_580_Direct_Cu_II/24.html

If the first new top card is 45% faster than the 580, than that's a performance gain we haven't seen since the GTX 280 hit. Even if we take 45% as the "final top card", then 45% is still in line from 480 over 285.

Not sure if the "only 45%" complaints are valid at all. If true, it will be better than the most recent two generations (way better than the current).

Not sure why there is the 4870 vs 5870 thing going on, as the 4870 was never the fastest card at any point in time anyway. You gotta compare top dog to top dog, imo. I know not every generation had a nm shrink, but a new generation is a new generation nevertheless.

Just a fyi, take whatever you want from this.

The 4890 vs 5870
http://www.beyond3d.com/content/reviews/54
40 percent on dx 9 and 10

On dx 11 double the performance

 

[Nemesis 1]

That's what I got out of it as well. Purely talking about clockspeeds.

Same here . Myself if both NV /AMD next generation are 70% faster I would be thrilled.

 

[tigersty1e]

Not sure why there is the 4870 vs 5870 thing going on, as the 4870 was never the fastest card at any point in time anyway. You gotta compare top dog to top dog, imo. I know not every generation had a nm shrink, but a new generation is a new generation nevertheless.

We're not saying the 4870 was the top dog. We are trying to predict the 7970 performance, so we need AMD's past performance jump numbers because for the most part they are consistent. Including Nvidia cards in the prediction will just throw everything off because some years they don't hardly beat amd (280 over 4870) and some years they are late to release (480 vs 580)