Leaked ATI S.I. 6870 benchmark

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

GaiaHunter

Diamond Member
Jul 13, 2008
3,650
218
106
That is what you get for multitasking...

I'll correct it in a second.

The GLs are supposed to be the Firestreams actually.
 
Last edited:

Dribble

Platinum Member
Aug 9, 2005
2,076
611
136
Wasn't expecting that much more performance. If that pans out in the real world it'll be pretty impressive.

Would like to seem more then a synthetic benchmark - if you've got the card then surely you'd test with everything?
 

bryanW1995

Lifer
May 22, 2007
11,144
32
91
well, it's not like they set the bar too high last time. all 6870 needs to do is be ~ 10% slower than 5970 and this "1/2 step" will actually be a full step. hopefully nvidia is more timely this time and we'll get some great competition and, more importantly, reasonable prices.
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
well, it's not like they set the bar too high last time. all 6870 needs to do is be ~ 10% slower than 5970 and this "1/2 step" will actually be a full step. hopefully nvidia is more timely this time and we'll get some great competition and, more importantly, reasonable prices.

I'm curious to see what Nvidia can do with 40nm. It seems like they pushed the limits on power and size already. But, I'm sure they won't sit on their thumbs and do nothing. I don't think Nvidia likes having the second fastest GPU.
 

HurleyBird

Platinum Member
Apr 22, 2003
2,759
1,455
136
I'm curious to see what Nvidia can do with 40nm. It seems like they pushed the limits on power and size already. But, I'm sure they won't sit on their thumbs and do nothing. I don't think Nvidia likes having the second fastest GPU.

I think Nvidia will probably discontinue GF100 products for desktop (but not the pro market) and will try to stetch GF104 as far as possible. They'll slot GTX 490 (2x Gf104) between HD6870 and HD6970, slot GTX475 (full spec GF104) between 6770 and 6850, and the current GTX460 probably around 6750. GF106 and 108 will be slotted wherever they can be based on the performance even if the parts have low margins.

It does seem that they are stretched to their limits, and it will be hard for them to compete with the 6xxx generation, at least at the high end. If they're lucky AMD will drop the ball again on tesselation performance and they can play the future proof card, but I wouldn't count on it.

I think about the best thing Nvidia can do is compete with whatever they can still make money on, scrap whatever they can't, and work like mad on their 28nm part.
 

Ika

Lifer
Mar 22, 2006
14,264
3
81
Wow, 35% is huge. looks like nvidia could be playing catchup for another six months again...
 

at80eighty

Senior member
Jun 28, 2004
458
5
81
while more than effing impressive, its still a synth. bench - I'm now wondering if i should go for a 6870 and then pick up the NI when it comes out; or go the hog and pick up a 6970 and ride out the first gen of NI. hm
 

Cookie Monster

Diamond Member
May 7, 2005
5,161
32
86
I think Nvidia will probably discontinue GF100 products for desktop (but not the pro market) and will try to stetch GF104 as far as possible. They'll slot GTX 490 (2x Gf104) between HD6870 and HD6970, slot GTX475 (full spec GF104) between 6770 and 6850, and the current GTX460 probably around 6750. GF106 and 108 will be slotted wherever they can be based on the performance even if the parts have low margins.

It does seem that they are stretched to their limits, and it will be hard for them to compete with the 6xxx generation, at least at the high end. If they're lucky AMD will drop the ball again on tesselation performance and they can play the future proof card, but I wouldn't count on it.

I think about the best thing Nvidia can do is compete with whatever they can still make money on, scrap whatever they can't, and work like mad on their 28nm part.

The thing is that TSMC's 28nm is due sometime late next year. So this is why we are seeing next gen parts from AMD using 40nm process tech. nVIDIA probably too has another 40nm refresh for the high end since the time frame for 28nm is quite far away.

So its possible that we could see something along the lines of 12SM (48SP each just like GF104), 3GPC chip (96TMU, 48ROP) from nVIDIA or even (8SM with 64SP each) which could end up smaller and use less power than the GF100 although GPGPU performance might take a little hit.

But it all depends on how NI is going turn out and will be interesting to see which aspect of its architecture has changed. Synthetic benchmarks don't really mean much so im waiting eagerly for the real world benches.
 

HurleyBird

Platinum Member
Apr 22, 2003
2,759
1,455
136
So its possible that we could see something along the lines of 12SM (48SP each just like GF104), 3GPC chip (96TMU, 48ROP) from nVIDIA or even (8SM with 64SP each) which could end up smaller and use less power than the GF100 although GPGPU performance might take a little hit.

I have no doubt Nvidia could produce something that is somewhat faster than than the current GF100. The very least they could do is bring in some of the GF104 design principles, and even then they can probably do a bit further tweaking. However, if their largest most power hungry, performance at-all-costs design cannot challenge SI there really isn't much of a reason to bring it out to market. We've seen Nvidia cut their losses before when they discontinued GT200b once it could not compete in either price or performance with cypress.
 
Feb 19, 2009
10,457
10
76
There's 7Gbps GDDR5 chips in mass production, confirmed a few months ago. The chips are also denser, which means on a 256bit bus, ATI could easily be aiming for a 2GB release as standard. However, i don't think they will, because the target pricing is around $440 USD, which is not much more than the 5870 at release. Probably will be a 1gb and later, a 2gb version.

Vantage is always the first leak as its the easiest for OEM workers to do. Others require full games and time. Expect some real game benchmarks this week however.

Edit: 28nm = late 2011, so no new Fermi refresh or Fermi 2 until then. NV scrapped the dual gf104 option as it exceeded 300W.
 

kalniel

Member
Aug 16, 2010
52
0
0
So the 17,3% must be for the 800MHz 512SP over the Default GTX480 and we get another 7,1% from the O/C of 50MHz.
Okay I see what you mean. The next page of the preview clears it up by giving the default 480SP score:

default 480SP = X9815

default 512SP = X11371

Therefore gain is 15.85% at default (non-identical) clock speeds

Overclocked 512SP = X12173

and is therefore a gain of 24.02% over default 480SP.

However, the 6870 might score around X12000, so they are still close.
 

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
I think about the best thing Nvidia can do is compete with whatever they can still make money on, scrap whatever they can't, and work like mad on their 28nm part.

Regardless of Northern Islands, I think there is too much missed opportunity between the release of GF100 and the availability of 28nm to not respin and tweak GF100.
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
There's 7Gbps GDDR5 chips in mass production, confirmed a few months ago. The chips are also denser, which means on a 256bit bus, ATI could easily be aiming for a 2GB release as standard. However, i don't think they will, because the target pricing is around $440 USD, which is not much more than the 5870 at release. Probably will be a 1gb and later, a 2gb version.

Vantage is always the first leak as its the easiest for OEM workers to do. Others require full games and time. Expect some real game benchmarks this week however.

Edit: 28nm = late 2011, so no new Fermi refresh or Fermi 2 until then. NV scrapped the dual gf104 option as it exceeded 300W.


I have no doubt that there will be 1GB and 2GB cards, but 2GB might not be what they launch with. 2GB 5870's are out there, hopefully 2GB 6870's come soon-ish after launch. While I'm in the '1GB is still enough' crowd, I don't know how long it will be enough.

Looking at AT's GTX460 review, the 460 uses more power than a 5850 but less than a 5870, so a dual GF104 card is probably possible. But, they may have to do what AMD did, down clock it or disable some parts. If they do that, will it be worth releasing? Nvidia has to walk a tight rope, not slow it down too much that it's slower than the 5970, but not use too much power. I have a feeling whatever they release will not be a 5970 killer, but more of an equal card. AMD seems to have an advantage, even against the GF104, in performance per watt. AMD is right up against that 300 watt wall, so I think it would be pretty tough for Nvidia to overtake AMD and stay under 300 watts. I mean, AMD has the FPS per watt advantage, so I think Nvidia would have to work some voodoo... not that it wouldn't be possible, I just don't see them making a 5970 killer though.

Regarding the 6870, I think AMD is really going to clobber Nvidia on the high end. AMD has had the fastest card for some 10 months now, and with a 6870, and I'm sure some sort of x2 card coming, they're really taking it to Nvidia. AMD's small scalable strategy is really paying off. They seem to be more nimble, able to get things out fast. I'm curious to see if Nvidia changes directions somewhat or if they stick to the large GPU strategy.
 

bryanW1995

Lifer
May 22, 2007
11,144
32
91
keep in mind that amd is not going quite so small this time. 6870 is going to be 400mm+, and if amd believes that they can have the single gpu crown for another 12-18 mos then who knows how big they'll go? 425mm2, 450mm2? They just went small ball before because they couldn't compete with nvidia at the top.
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
keep in mind that amd is not going quite so small this time. 6870 is going to be 400mm+, and if amd believes that they can have the single gpu crown for another 12-18 mos then who knows how big they'll go? 425mm2, 450mm2? They just went small ball before because they couldn't compete with nvidia at the top.

I assume you are right, but we may not me. Who knows how big AMD is going, are they adding 20% to their current GPU's or are they going Nvidia-big? Will it not really be larger at all and just be rearranged internally with some new shiney parts under the hood? Are these Vantage scores really meaningful, and we'll see a much smaller real world gain? Who knows, though I am curious to see how it plays out.
 

Soleron

Senior member
May 10, 2009
337
0
71
I assume you are right, but we may not me. Who knows how big AMD is going, are they adding 20% to their current GPU's or are they going Nvidia-big? Will it not really be larger at all and just be rearranged internally with some new shiney parts under the hood? Are these Vantage scores really meaningful, and we'll see a much smaller real world gain? Who knows, though I am curious to see how it plays out.

Charlie says SI is a ~380mm^2 part. So bigger but not by that much. He thinks the 35% boost is "in the ballpark".
 

bryanW1995

Lifer
May 22, 2007
11,144
32
91
wow, 380mm^2 and 35% boost would be great. that sounds more like what NI would have been in the first place at 28 nm however, almost too good to be true.

amd is fast approaching "public enemy #1" status. If they're able to squeeze nvidia out or, even worse, force them into intel's arms we are going to see extremely high gpu prices in the future. I really hope nvidia has something up its sleeve.
 

zokudu

Diamond Member
Nov 11, 2009
4,364
1
81
In fact, a bit *TOO* amazing. I'm sure people won't be disappointed with a hair faster than the GTX480 512, with livable power and noise characteristics.

If that happens I'll probably pick up a 470. Hoping that some day I'll have a choice of video card vendors for Linux, but for now I'm locked into NV.

I have to say the FGLRX driver has been working flawlessly for me on my laptop and has done well for all my needs and on my future linux machines I don't think I would hesitate to get an ATI card anymore. Perhaps your uses are different from mine but have you tried ATI linux drivers recently?