AMD's Radeon HD 6870 benchmarked? (updated more screens)

[saaya]

From 3Dcenter

i like those charts... but they havent exactly been accurate in the past... according to that chart ati will use the 6000 series to push their asps up, which i doubt tbh... if the chart is accurate, then the 6800 series is barely an improvement over the 5800 series, and the 6700 series can only reach 5830 performance? if thats true, then thats bad...
bart has to be at least as fast as gf104, if not faster...

Because not everyone knows how to read the Chinese original source

nobody knows how to use google translate?
and nobody can understand the screenshots because they are posted on a chinese website?

[RaZz!]

According to these scores, 6870 will be ~35% better than 5870 (muuuuuch more in intense tesselation situations), and ~20% better than gtx480. Seems like a nice follow up

For reference this is what scores a gtx480 w/ extreme tesselation & 1920/1080

*snip*

35% faster? it's 17/18 (5870) vs. 36 (6870) fps in unigine, no?

plus it's 1920x1200 (more pixels) vs 1902x1080 (higher fov?). no idea how much of a difference this makes.

6870 looks promising though. my 4850 feels a bit old when playing bc2 - i'm quite gpu limited there, and looking forward to replace my gpu with one of the 6000 series cards.

[JohnJohn]

35% faster? it's 17/18 (5870) vs. 36 (6870) fps in unigine, no?
.

Thats why i said "(muuuuch more in intense tesselation situations)" , the 35% number i took it from vantage scores and crysis

[Olivon]

i like those charts... but they havent exactly been accurate in the past... according to that chart ati will use the 6000 series to push their asps up, which i doubt tbh... if the chart is accurate, then the 6800 series is barely an improvement over the 5800 series, and the 6700 series can only reach 5830 performance? if thats true, then thats bad...
bart has to be at least as fast as gf104, if not faster...

That's not about performance rating Sascha, that's just about prices and segments

[RPGWiZaRD]

Haha so many people fail to read that graph, it's only about pricing and you know ATI/AMD pushes prices often a lot more aggressively than Nvidia even if they hold the performance crown whereas ATI/AMD seems to understand there's a certain amount cash people are willing to spend while Nvidia simply prices the cards according to their performance. It's common for ATI/AMD to replace new cards are roughly similar prices as previous series but with added performance obviously. For Nvidia they only got expensier and expensier when they were at top (top card was sold for $500 then 550-600 for the next series, then $650 etc) but since they don't have the performance crown atm, they are pricing their cards rather according to ATI/AMDs pricing.

I think it's the best way to go atm with aggressive pricing even if they could say charge 500~530 EUR or $550~599 for a HD6870, they'll make life a lot harder for NVIDIA especially for GTX 480 which is a card that was probably targeted to sold at even higher than what it is, I'm sure NVIDIA wanted to price GTX 480 gladly at even $650 if they could. Now if 6870 is as good as these early "rumors" say then GTX 480 (and 470 most likely too) will be have to face price cuts and this is not good for Nvidia at all. Now ATI/AMD might not gain as much profit from selling at slightly lower price but they'll hurt Nvidia more instead and make Nvidia lose more market share and they will gain more market share which is probably easily worth a $50 lower price on HD6870. Talk about HD4xxx - HD6xxx series being ATI/AMD's era in discrete graphics market. At HD4xxx nv could still keep up well but that was the series started what ATI/AMD has built up a strong ground around. Soon they'll have to switch to an entire new architecture and then there's always more risks especially as NV might get their stuff right that time around as well but AMD/ATI has been very smart these last 3 gens.

[Calmatory]

Hehe. Higher prices here we come.

[Olivon]

I think it's the best way to go atm with aggressive pricing even if they could say charge 500~530 EUR or $550~599 for a HD6870, they'll make life a lot harder for NVIDIA especially for GTX 480 which is a card that was probably targeted to sold at even higher than what it is, I'm sure NVIDIA wanted to price GTX 480 gladly at even $599 if they could. Now if 6870 is as good as these early "rumors" say then GTX 480 will be have to face price cuts and this is not good for Nvidia at all. Now they might not gain as much profit from selling at slightly lower price but they'll hurt Nvidia more instead and make Nvidia lose more market share and they will gain more market share which is probably easily worth a $50 lower price on HD6870.

I agree RPGWiZaRD

NVidia is losing this 40nm round and they don't have many things to deliver until 28nm next chapter.

[RPGWiZaRD]

Hehe. Higher prices here we come.

But still if that graph is true then 450€ for a HD6870 which would correspond to $499.99 newegg price for the cheapest standard designs which is the same what GTX 480 sells for atm and if these benchmarks are true then it would be like perhaps 20% faster and MOST LIKELY draw significantly less power, I don't think that's such a bad deal, they could charge even more IMO.

[Manicdan]

Hehe. Higher prices here we come.

i think were going to find out if i can skip 2 generations

[informal]

Flashforward one year from now : Radeon 6000 prices in Q3 2011 are the same as launch prices in Q3 2010.Déjà vu all over again .

[Florinmocanu]

My guess is that Nvidia is heavily re-working the GF100/GF104 architecture, without much intention to release a dual gpu (which would probably be beaten by the future 6970).

It may release, but my guess is that their main focus is on properly re-working the architecture to be able to stand against AMD, maybe even on 40nm. What AMD did with SI in terms of performance improvements Nvidia can and i guess will do as well, they have to.

[Manicdan]

it seems nvidia is more concerned about the server market, trying to sell as many cuda cores as possible instead of gpus. i wonder if they will keep sacrificing gaming for cuda.

[Calmatory]

My guess is that Nvidia is heavily re-working the GF100/GF104 architecture, without much intention to release a dual gpu (which would probably be beaten by the future 6970).

It may release, but my guess is that their main focus is on properly re-working the architecture to be able to stand against AMD, maybe even on 40nm. What AMD did with SI in terms of performance improvements Nvidia can and i guess will do as well, they have to.

I'm afraid they can't, unless there's a clear bottleneck in Fermi's gaming performance somewhere, and Nvidia can fix that, and actually decides to do so considering that 40 nm round is near to it's end anyway. I'd guess tons of FUD and marketing, if not anything else.

If it is actually true that AMD would focus on tesselation performance(guess it's their weakest link now), then they have quite much found the bottleneck wich allows the architecture to scale well with the improvements, leading to nice gains.

..and as Manicdan pointed out, Nvidia doesn't necessarily even care about 40nm that much. It's focus is somewhere else. Actually it has been all this time with Fermi.

[Olivon]

I'm afraid they can't, unless there's a clear bottleneck in Fermi's gaming performance somewhere, and Nvidia can fix that, and actually decides to do so considering that 40 nm round is near to it's end anyway. I'd guess tons of FUD and marketing, if not anything else.

If it is actually true that AMD would focus on tesselation performance(guess it's their weakest link now), then they have quite much found the bottleneck wich allows the architecture to scale well with the improvements, leading to nice gains.

..and as Manicdan pointed out, Nvidia doesn't necessarily even care about 40nm that much. It's focus is somewhere else. Actually it has been all this time with Fermi.

Right too Calmatory.
NVidia is doomed on 40nm ant they have no time to do a 40 nm refresh on Fermi 40 nm.
A chip revision maybe in order to decrease power consumption and calore but that's all.
NVdia had no B plan. That was a terrible mistake and ATI wins the cake.

[Levish]

35% faster? it's 17/18 (5870) vs. 36 (6870) fps in unigine, no?

plus it's 1920x1200 (more pixels) vs 1902x1080 (higher fov?). no idea how much of a difference this makes.

6870 looks promising though. my 4850 feels a bit old when playing bc2 - i'm quite gpu limited there, and looking forward to replace my gpu with one of the 6000 series cards.

[speculation] i could see 35% faster in tesselation specific intensive stuff, in DX 9/10 its probably not going to be more than 15-20% faster [/speculation]

I was hoping the 6870 would come in at a lower price point than the 5870 but looking at this chart doesn't really reflect this.

[eric66]

sucks to be nvidia right now lol

[LightSpeed]

Well, considering that since the 48xx their architecture has been the same (58xx was just doubled with some tweaks), im sure in these couple of years they have found means to rework it and eek out more performance from it

However, Fermi got released a few months back, im sure itd take them a while to do a refresh

[madcho]

I set the same settings that was show in that test with 5870 1Go, My cpu is a bit slower but i don't think this is a problem here with this so low fps.



It's just awesome if it's not fake.

Right too Calmatory.
NVidia is doomed on 40nm ant they have no time to do a 40 nm refresh on Fermi 40 nm.
A chip revision maybe in order to decrease power consumption and calore but that's all.
NVdia had no B plan. That was a terrible mistake and ATI wins the cake.

this one ?

[Manicdan]

[speculation] i could see 35% faster in tesselation specific intensive stuff, in DX 9/10 its probably not going to be more than 15-20% faster [/speculation]

I was hoping the 6870 would come in at a lower price point than the 5870 but looking at this chart doesn't really reflect this.

i agree, both with your speculation and what i was hopping for.

its looking like AMD is doing a better job of cutting up cores to fit all the different needs. so they feel confident they can make a massive chip for 200W, and scale it down for all the others, and still fit the x2 in 300W with some slight adjusting.

they use to be all about the small chips, and if they keep up with this modular design they might be about the biggest too.

[Calmatory]

Based on the other thread, 6870 running with 1920x1200 w/ 4xAA should beat GTX 480 which runs with 1920 x 1080 w/o AA(?) on Unigine?

Too good to be true. :|

[madcho]

Based on the other thread, 6870 running with 1920x1200 w/ 4xAA should beat GTX 480 which runs with 1920 x 1080 w/o AA(?) on Unigine?

Too good to be true. :|

unigine is favorable to fermi because fermi is faster in FP16, geometry and tesslation.

That benchmark test that things in a heavy load. A 5870 is faster that a 470 in real games, so i believe it's not a surprise that tesslation & geometry was hardly optimised in cayman.

And i love the 40 TMU added too

[RPGWiZaRD]

Well it makes sense that any internal changes they made was mostly focused on improving tesselation performance cuz that's been HD5xxx's weak link. If I was in charge of development I'd clearly opt for improving tesselation performance by small internal architecture changes and otherwise just add more SPs and TMU and maybe slight clock frequency boost cuz that's easy, safe & fast way of improving performance for cards on tight schedule. By the looks of it that's what they've done.

Looks like their marketing team will have fun times ahead, tesselation performance, lower power consumption than competitor, performance advantage...

[madcho]

Geometry was incredible faster in fermi. I Hope ATi has done some work in that space.

[RPGWiZaRD]

Geometry was incredible faster in fermi. I Hope ATi has done some work in that space.

True but somehow I doubt they've been able to address this into 6xxx series (improving tesselation performance is already a great step) and most likely leaving geometry performance improvements for Northern Islands 28nm. A small increase maybe but I don't expect much. As I see it they won't need greatly improved geometry performance and still do very well with SI, NI on the other hand will need it and with 28nm it'll be easier (stay within reasonable die sizes and TDP).

[Florinmocanu]

Oh come one, read more careful. The GTX 480 scores are with 4xAA and 16 AF.

So, at 1920 by 1080, 4x AA, 16AF

GTX 480 = 30.5 average fps, HD 6870 is 36.6 fps

http://www.hexus.net/content/item.php?item=25391&page=6





According the the leaks, HD6870 is faster than GTX 480 in tesselation/DX11, at the same settings. It's quite obvious actually.