Leaked ATI S.I. 6870 benchmark

[Grooveriding]

grooveriding: any way you can redo your crysis bench from earlier in this thread on a single 480 with 64bit checked? Thanks

Here is a single 480 on 64 bit

Leaked 6870 for the same settings

 

[evolucion8]

That's a nice score for Crysis, GTX 480 is ain't bad but the score of the HD 6870 is impressive, seems that the HD 6870 will be a nice upgrade for my rig as a birthday gift (I went 30 seven days ago). 😛

 

[Chumster]

I Don't see it. Could you put another pic of legit heaven benches to compare.

I believe he is referring to the top section where the "2" in the Scores: field looks different than the "2" in the Min FPS: and Max FPS: field. He's partially correct - they do in fact look different. What he doesn't realize is that the top 2 fields are in bold and that the numbers are, in fact, using the exact same font. 🙂

 

[Obsoleet]

Looks decent!

 

[tviceman]

If this leaked bench was accurate the 6870 is faster than a 480 in tesselation as you predict.

I Don't see it. Could you put another pic of legit heaven benches to compare.

I'm referring to big yellow numbers. Specifically, the SCORES: 922 compared to min FPS: 22.0 and max FPS: 70.2

The "922" font is clearly different from the "22.0" and "70.2" font. Look at the 2's. Does this screen always use different fonts between the scores line and the min and max FPS lines?

I believe he is referring to the top section where the "2" in the Scores: field looks different than the "2" in the Min FPS: and Max FPS: field. He's partially correct - they do in fact look different. What he doesn't realize is that the top 2 fields are in bold and that the numbers are, in fact, using the exact same font. 🙂

Ah I see. That could be it. I'm being to lazy to look, but is that how Unigine always outputs the fonts on that screen?

EDIT: I did I search, found this page and correct me if I am wrong, but the fonts and type are identical between the min, max, and scores lines. http://www.overclock.net/benchmarki...-official-top-30-heaven-benchmark-2-a-11.html

If this is how the Heaven benchmark normally outputs it, then why would the leaked 6870 benchmark have some lines bolded and others not?

 

[Grooveriding]

Here is a single GTX 480 in Heaven

Here is the 6870 in Heaven

Looks like tesseletion performance is also superior. I also forgot to downclock the 480 so this is with a 480 running at 800/2000.

As for the tin foil conspiracies about the image itself. The benchmark spits out an html format file to give you the results when you save them. Differences in appearance could be as simple as having different display and format settings in your browser.

 

[railven]

Here is a single GTX 480 in Heaven

Here is the 6870 in Heaven

Looks like tesseletion performance is also superior. I also forgot to downclock the 480 so this is with a 480 running at 800/2000.

As for the tin foil conspiracies about the image itself. The benchmark spits out an html format file to give you the results when you save them. Differences in appearance could be as simple as having different display and format settings in your browser.

Hell I'd add any kind of anomolies to the simple fact that using the JPEG extension is adding compression. Who knows how many times this image has been "saved" and "uploaded?"

I forgot who here was arguing never use a JPG to argue IQ, I'd figure that would apply to an image's authenticity? Haha.

 

[railven]

I'm referring to big yellow numbers. Specifically, the SCORES: 922 compared to min FPS: 22.0 and max FPS: 70.2

The "922" font is clearly different from the "22.0" and "70.2" font. Look at the 2's. Does this screen always use different fonts between the scores line and the min and max FPS lines?




Ah I see. That could be it. I'm being to lazy to look, but is that how Unigine always outputs the fonts on that screen?

EDIT: I did I search, found this page and correct me if I am wrong, but the fonts and type are identical between the min, max, and scores lines. http://www.overclock.net/benchmarki...-official-top-30-heaven-benchmark-2-a-11.html

If this is how the Heaven benchmark normally outputs it, then why would the leaked 6870 benchmark have some lines bolded and others not?

Checking your link, maybe my job is blocking stuff, but I only see Uni 2.0 scores, not 2.1.

I know 2.1 doesn't output the results into your browser as one of the pictures there - could just be my short experience with it.

EDIT:
Towards the end of the thread you see some 2.1 scores:

http://www.overclock.net/benchmarki...-official-top-30-heaven-benchmark-2-a-65.html

The top numbers look "bolder" than the lower numbers in that first screen shot versus the 2.0 exmaples in the other pages of the same thread.

 

[tviceman]

Checking your link, maybe my job is blocking stuff, but I only see Uni 2.0 scores, not 2.1.

I know 2.1 doesn't output the results into your browser as one of the pictures there - could just be my short experience with it.

EDIT:
Towards the end of the thread you see some 2.1 scores:

http://www.overclock.net/benchmarki...-official-top-30-heaven-benchmark-2-a-65.html

The top numbers look "bolder" than the lower numbers in that first screen shot versus the 2.0 exmaples in the other pages of the same thread.

Yeah it could be a number of things so I'll go with the leaks being completely plausible. Assuming they're accurate, it looks like the 6870 is gearing up to be a great high end card.

 

[railven]

I have a question, not sure if it was asked/answered. I know currently ATI uses the 5 instruction counter for their shader count (1600 SP == 320x5) does that mean the count will change with this new architecture since it's shifting to a 4? Or did I miss the memo?

 

[Skurge]

I have a question, not sure if it was asked/answered. I know currently ATI uses the 5 instruction counter for their shader count (1600 SP == 320x5) does that mean the count will change with this new architecture since it's shifting to a 4? Or did I miss the memo?

Thats a good question actually. I think I might just switch to to X4 for the shader count, but it sounds like it's a little more complicated than that.

 

[MrK6]

Rumor has it that Cayman is 1920 SP (480x4); but that hasn't been substantiated by AMD, so who knows?

 

[lopri]

Tempered81 posted about it early in this thread.

http://forums.anandtech.com/showpost.php?p=30369125&postcount=205

To keep size & TDP practical, 'Cayman' is only 3/4ths of the full 28nm design in many ways. 3 of 4 SIMD groups, 3 of 4 Tessellating units. 3x out of a possible 4x higher Tessellation power over cypress. Contrastingly, IMO, Barts is a full 1/2 of the original NI flagship design: 1280SP. As a result, the performance gap between 'Barts' XT (360alu) and 'Cayman' XT (480alu) should be smaller than gap between 5770 Juniper XT (160alu) and 5870 Cypress XT (320alu). 3rd gen video decoding UVD3, Eyefinity4 (+). The stream processor redesign is there. Vector 5d > 2+2 incorporated t-unit.

 

[Screech]

Rumor has it that Cayman is 1920 SP (480x4); but that hasn't been substantiated by AMD, so who knows?

480x4 definitely seems to align pretty well with a lot of the other rumors floating about. If all the stuff about the new 4-sp block being 95% as fast as the 5-unit block but 80% less die area, then 1920 /480x4 should be about 95% as fast as a theoretical 2400-sp 5xxx card (so "2280 sp" 5xxx series) WITHOUT taking into account improvements to the tesselator, etc. A 2280 sp 5xxx series would have 42% more shaders than the existing 5870, which certainly puts it, theoretically, in the ballpark of the improvements the few benchmarks out now seem to suggest, if it (the 6870) really is 25% faster than the gtx 480. (damn, the number 480 is going to take on multiple meanings now ! 😉

If you only count shaders as contributing to die area, 480x4 would have a 20% increase in die size for this increase in performance, which would be pretty sweet. That said, I think the die will increase more due to other things (4 isnt really 20% smaller than 5, memory, a bunch of other stuff that isn't only shader related, stuff I don't understand, etc.)

My apologies for the incoherent rambling, this is all quite interesting to me.

 

[AdamK47]

I'll take 3 in CrossFireX

 

[Silverforce11]

Cayman XT is 256bit with high speed GDDR5, rated for 7Gbps. The rest is fud, ATI is trying to crack down on leaks by giving fake numbers and specs to AIBs and see who is leaking. They have told their partners not to leak anymore until the official paper launch (its not helping the demand for 5800 GPUs).

Cayman Pro is not going to be as fast as the gtx480.. but it doesn't need to be, its aimed at 5870 performance, but for cheap.

 

[RussianSensation]

Silverforce11, by all current rumors, 6870/90 won't launch until 2011. Looks like there is nothing to get excited about for 3 months.

 

[Grooveriding]

Silverforce11, by all current rumors, 6870/90 won't launch until 2011. Looks like there is nothing to get excited about for 3 months.

http://translate.google.ca/translate?hl=en&sl=tr&u=http://www.donanimhaber.com/ekran-karti/haberleri/AMD-Radeon-HD-6800-serisi-Kasim-ayinda-lanse-edilecek.htm&ei=oiWJTICHO8y1nAe-0ZXBCw&sa=X&oi=translate&ct=result&resnum=1&ved=0CBcQ7gEwAA&prev=/search%3Fq%3Dhttp://www.donanimhaber.com/ekran-karti/haberleri/AMD-Radeon-HD-6800-serisi-Kasim-ayinda-lanse-edilecek.htm%26hl%3Den%26client%3Dfirefox-a%26hs%3Dj76%26sa%3DG%26rls%3Dorg.mozilla:en-US😱fficial%26channel%3Ds

Most recent rumors point to 6870 next month in October.

6970 in November.

6990 in December.

AMD likely wants to capitalize on the holiday season the same way they did with 58XX series last year.

 

[brybir]

Cayman XT is 256bit with high speed GDDR5, rated for 7Gbps. The rest is fud, ATI is trying to crack down on leaks by giving fake numbers and specs to AIBs and see who is leaking. They have told their partners not to leak anymore until the official paper launch (its not helping the demand for 5800 GPUs).

Cayman Pro is not going to be as fast as the gtx480.. but it doesn't need to be, its aimed at 5870 performance, but for cheap.

Well the nature of the internet makes it also plausible for me to say that you are being paid by Nvidia to try to counteract the information release by spreading FUD.

Do I have any evidence? No. Do I care? Not really.

My point is is that the nature of the internet makes all of almost everything speculation until official release, and sitting around and saying one thing or another is FUD, as an anonymous source, does nothing but derail the fun we are all having guessing about upcoming features.

But who knows, maybe I am being paid by AMD to spread FUD about the people I think are spreading FUD about AMD in order to get excitement for 6XXX up and in turn attempt to limit Nvidia's sales in the near future? /infiniteloop

 

[happy medium]

http://translate.google.ca/translate?hl=en&sl=tr&u=http://www.donanimhaber.com/ekran-karti/haberleri/AMD-Radeon-HD-6800-serisi-Kasim-ayinda-lanse-edilecek.htm&ei=oiWJTICHO8y1nAe-0ZXBCw&sa=X&oi=translate&ct=result&resnum=1&ved=0CBcQ7gEwAA&prev=/search%3Fq%3Dhttp://www.donanimhaber.com/ekran-karti/haberleri/AMD-Radeon-HD-6800-serisi-Kasim-ayinda-lanse-edilecek.htm%26hl%3Den%26client%3Dfirefox-a%26hs%3Dj76%26sa%3DG%26rls%3Dorg.mozilla:en-US😱fficial%26channel%3Ds

Most recent rumors point to 6870 next month in October.

6970 in November.

6990 in December.

AMD likely wants to capitalize on the holiday season the same way they did with 58XX series last year.

Latest I heard was the 6770 on Oct 24th a dual 6770 in December and the 6870 and the rest of the low end in early 2011.

 

[Headcase_Fargone]

AMD likely wants to capitalize on the holiday season the same way they did with 58XX series last year.

Did they really capitalize on the holiday season last year? Honest question. I seem to recall there being little to no stock of the things until early 2010 on account of TSMC issues.

 

[MrK6]

Did they really capitalize on the holiday season last year? Honest question. I seem to recall there being little to no stock of the things until early 2010 on account of TSMC issues.

Well, wouldn't they want to make a better effort this year?

 

[Grooveriding]

Did they really capitalize on the holiday season last year? Honest question. I seem to recall there being little to no stock of the things until early 2010 on account of TSMC issues.

I'd say they capitalized as best as possible as the 5XXX series cards were selling out at e-tailors within minutes of coming into stock for months. I'm sure they could of done better but considering the huge demand for the 5XXX series and the TSMC issues. It is safe to say, they sold every last card that they could produce in that period.

When 6870 comes next month we'll likely see the same of course, they'll sell out as fast as they come into stock and 6970 and 6990 through November and December will be the same.

It's usually the case with a new launch, especially with this being prime buying season and AMD is launching a new generation.

That said, there will likely be tons of AMD 5XXX series and nvidia gtx 4XX series on sale for super cheap with most buyers trying to get their hands on the new 6XXX AMD Radeons.

 

[Headcase_Fargone]

I'd say they capitalized as best as possible as the 5XXX series cards were selling out at e-tailors within minutes of coming into stock for months. I'm sure they could of done better but considering the huge demand for the 5XXX series and the TSMC issues. It is safe to say, they sold every last card that they could produce in that period.

When 6870 comes next month we'll likely see the same of course, they'll sell out as fast as they come into stock and 6970 and 6990 through November and December will be the same.

I'm well aware that they were selling out as soon as they hit the shelves. I was in the market for a new video card at the time, actually, so it was very frustrating. I just wonder how much they were able to actually capitalize on the time of year that they released with such a limited stock available for sale.

Hopefully TSMC can do a better job this time around and give them plenty of stock to ship out at release. I can't imagine that not being the case considering the 40nm process is so mature now compared to last year.

 

[RussianSensation]

Most recent rumors point to 6870 next month in October.

That's not what I read. The link you posted is from Aug. 26th.

http://vr-zone.com/articles/-rumour...bility-on-october-25th-more-details/9772.html

Oct 25th is the launch of 6700 series, with 6800/6900 to follow later, with some cards arriving in 2011. September, October, November - so at least 3 months+.

Plus, "launch" and retail availability are 2 different things. Even if 6800 is launched in November, it might not ship to stores until December, etc.

I have no idea where you get info that 6870 will launch next month. The only card launching at the end of October is HD6770 (mid-range) != 6870 (high end)

ATI won't have high end cards for almost another quarter.