ATI HD 6970 actual benchmarks

9

96redformula

2[H]4U
Joined
Oct 29, 2005
Messages
2,578
This is from http://translate.google.com/translate?hl=en&sl=de&tl=en&u=http%3A%2F%2Fwww.hardwareluxx.de%2Fcommunity%2Ff14%2Famd-radeon-hd-6970-alle-technischen-daten-update-2-a-768017-30.html

69701jlz3.jpg


gpuzneuvxuk.jpg
 
Last edited:
These were posted in at least one or two other threads, I do believe. Their veracity is still questionable.
 
fattypants said:
These were posted in at least one or two other threads, I do believe. Their veracity is still questionable.
Click to expand...

The guy even posted the RE5 screenshot as well showing 6900 series. He said he tried the button as well and didn't see performance changes, must be dual bios. I personally can't find a reason he would be lying.
 
96redformula said:
The guy even posted the RE5 screenshot as well showing 6900 series. He said he tried the button as well and didn't see performance changes, must be dual bios. I personally can't find a reason he would be lying.
Click to expand...

If this is how the 6970 really performs then AMD has a real clunker on their hands.
This would mean that the VLI4W shader architecture is a complete disaster compared to VLI5W.
AMD managed to squeeze nearly 5870 level performance out of the 6870 using the old fashioned VLI5W shaders and only 1120 SPs compared to the 5870's 1600 SPs.
The 6870 would have done a lot more with LESS.
The 6970 would do a lot LESS with more.

Just doing some simple math:
6870 ~ 5870
1120 shaders vs 1600 shaders ~ 30% fewer shaders.
Clearly the shaders haven't changed any yet efficiency is increased nearly 30% due to front end architecture.

6970 vs 5870
1536/4 vs 1600/5
384 vs 320 ~ 20% more shader clusters in a 1536 SP projected 6970.
So if the front end of the 6970 is the same as the 6870 we would expected 30% increase regardless of shaders. Add in another 20% for more shader clusters and we would expect a near 50% increase in overall performance. But it looks like according to the 'leaks' we'll only see a 20% overall gain vs the 5870.

This means that even with an increased shader cluster count, the VLI4W shaders would be causing an overall 10% performance decrease despite improved overall front end architecture. I'm sorry but AMD engineers are not dummies. I can't see how this can possibly work out mathmatheically even going with a less optimistic overall shader count of 1536 for the 6970. They wouldn't engineer BAD shaders and INCREASE die size, and INCREASE costs to pair with the brilliant front end architecture that yielded a remarkable near 30% efficiency gain in the 6870, would they?
 
vjcsmoke said:
If this is how the 6970 really performs then AMD has a real clunker on their hands.
This would mean that the VLI4W shader architecture is a complete disaster compared to VLI5W.
AMD managed to squeeze nearly 5870 level performance out of the 6870 using the old fashioned VLI5W shaders and only 1120 SPs compared to the 5870's 1600 SPs.
The 6870 would have done a lot more with LESS.
The 6970 would do a lot LESS with more.

Just doing some simple math:
6870 ~ 5870
1120 shaders vs 1600 shaders ~ 30% fewer shaders.
Clearly the shaders haven't changed any yet efficiency is increased nearly 30% due to front end architecture.

6970 vs 5870
1536/4 vs 1600/5
384 vs 320 ~ 20% more shader clusters in a 1536 SP projected 6970.
So if the front end of the 6970 is the same as the 6870 we would expected 30% increase regardless of shaders. Add in another 20% for more shader clusters and we would expect a near 50% increase in overall performance. But it looks like according to the 'leaks' we'll only see a 20% overall gain vs the 5870.

This means that even with an increased shader cluster count, the VLI4W shaders would be causing an overall 10% performance decrease despite improved overall front end architecture. I'm sorry but AMD engineers are not dummies. I can't see how this can possibly work out mathmatheically even going with a less optimistic overall shader count of 1536 for the 6970. They wouldn't engineer BAD shaders and INCREASE die size, and INCREASE costs to pair with the brilliant front end architecture that yielded a remarkable near 30% efficiency gain in the 6870, would they?
Click to expand...

One fatal flaw: you're assuming shaders scale linearly forever. There are saturation points and throughput isn't a straightline. At one point or another, simply having more shaders isn't going to make things faster. Hence Barts rebalanced the core to have fewer shaders and more TMUs. It looks like Cayman is doing the same thing.

One of the things Cypress showed was that despite being 2 x Juniper in every functional unit, performance was not 2X faster.

Barts appears to be Cypress' sweet spot in terms of shaders + TMU + ROPS + the entire rebalance.

P.S. 3dMark11 might utilize the 5th unit in 5D, hence effectively, Cypress has similar TFLops as Cayman

However, gaming is where the differences are, and the performance there seems much more in Caymans favor
 
At this point, I don't know what the hell is going on........It's like something from the twilight zone.
 
heflys20 said:
At this point, I don't know what the hell is going on........It's like something from the twilight zone.
Click to expand...


lol good way of putting it.

though its the typical rumor mill crap. only a few more days till we get the legit information..
 
ZerazaX said:
One of the things Cypress showed was that despite being 2 x Juniper in every functional unit, performance was not 2X faster.
Click to expand...

Strictly speaking, memory bandwidth was far from doubled. Then again, I doubt that's actually the bottleneck with Cypress.
 
I'm curious about what new feature set the 6970/6950 will bring to the table, if any.

Rumor has it the AMD cards in the wild now are purposefully "slowed". It will be interesting to see the reviews.
 
Since we're all tossing pictures into this mess of rumors, I'll toss my own pictures as well.

Let's say we're to believe ALL and EVERY rumored benchmark that has come out in the last 72 hours about the Radeon 6970, this would then THEORETICALLY position the Radeon 6970 like so:

currentpossibility.jpg


It doesn't make much sense does it?

However, let's say all these people managed to fake these benchmarks (or AMD used that lawsuit money Intel paid them and paid these people to fake these benchmarks), then the Radeon 6970 should look like this:

whatitshouldbe.jpg


Think about it for a minute people.

Think long and hard.

Why would AMD position the Radeon 6970 so close to the 5870 and 480, when the 6870 is practically at the heels of the 5870 in terms of performance?

All these pictures that's coming out is not making any shred of logical sense. Even the chart I made above basing them on these rumors and the recent reviews of the GTX 580 and 570 is not making much sense to me either.

There is no way AMD would shoot themselves in the foot to release a subpar high-end video card that barely outperforms the last generation's high-end card-- 5870. It's like stating that Ford only added 50 horsepower to a 2010 Mustang and called it a 2011 Mustang. It doesn't make sense at all if AMD wanted to do that.

I say we shut up for now and wait until December 15. If AMD shoots themselves in the foot, then ready the pitchforks and torches, and I'll meet you guys at AMD's front lawn. If AMD managed to pull something out of left field and catches us by surprise, then we will celebrate with beer, wine, caviar, and the latest episodes of Big Bang Theory played over HDMI connected to a PC with a 6970 in them.

But, for now, rumors are rumors, and nothing more until they become fact.
 
abdulrehmansuleman said:
May be he is just laughing on how such accurate benchmarks got posted before NDA. :D
Click to expand...

As I recall it, Charlie has no love for nVidia. I doubt he would be laughing about AMD owning themselves. My guess is the cards are going to be faster than the leaks and have a compelling feature set.

In any case, we will know soon enough.
 
So its true, ATI really did decrease the shader count, WTF were they thinking. Anyway, I'm not too happy about these early leaks, guess I'll have to wait until the real reviews come.

1600 to 1536 = FAIL... Seriously, AMD/ATI? Why?
 
octoberasian said:
Since we're all tossing pictures into this mess of rumors, I'll toss my own pictures as well.

Let's say we're to believe ALL and EVERY rumored benchmark that has come out in the last 72 hours about the Radeon 6970, this would then THEORETICALLY position the Radeon 6970 like so:


It doesn't make much sense does it?

However, let's say all these people managed to fake these benchmarks (or AMD used that lawsuit money Intel paid them and paid these people to fake these benchmarks), then the Radeon 6970 should look like this:

Think about it for a minute people.

Think long and hard.

Why would AMD position the Radeon 6970 so close to the 5870 and 480, when the 6870 is practically at the heels of the 5870 in terms of performance?

All these pictures that's coming out is not making any shred of logical sense. Even the chart I made above basing them on these rumors and the recent reviews of the GTX 580 and 570 is not making much sense to me either.

There is no way AMD would shoot themselves in the foot to release a subpar high-end video card that barely outperforms the last generation's high-end card-- 5870. It's like stating that Ford only added 50 horsepower to a 2010 Mustang and called it a 2011 Mustang. It doesn't make sense at all if AMD wanted to do that.

I say we shut up for now and wait until December 15. If AMD shoots themselves in the foot, then ready the pitchforks and torches, and I'll meet you guys at AMD's front lawn. If AMD managed to pull something out of left field and catches us by surprise, then we will celebrate with beer, wine, caviar, and the latest episodes of Big Bang Theory played over HDMI connected to a PC with a 6970 in them.

But, for now, rumors are rumors, and nothing more until they become fact.
Click to expand...

Although it's a strange coincidence that rumours from different sources are pointing to a gtx 570ish performance and benchmarkreviews has even an article that it will compete with the gtx 570.

http://benchmarkreviews.com/index.php?option=com_content&task=view&id=13177&Itemid=8
 
Blacklash has spoken and that is all what needs to be said. Come on guys, we are enthusiast....right? ATI has been on a roll with their graphics cards since the release of th 4800 series. Now tell me why in the hell would they release a product that wouldn't perform significantly better than last generation. Rather it's TDP, FPS improvements, etc. I can gaurantee that these cards will not disappoint. My biggest fear is pricing. I haven't upgraded in over 2 years and I have $500 to plop down on some uber-goodness performance. So let us hope that AMD price these things right!:D
 
Well i believe that device ID 6718 tends to be(might be) an 6950 ;)

Further proof
From 10.12

AMD Radeon 6950 - HD 6900 Series" = ati2mtag_NICayman, PCI\VEN_1002&DEV_6718
AMD Radeon 6970 - HD 6900 Series " = ati2mtag_NICayman, PCI\VEN_1002&DEV_6719
 
There was rumors earlier in the week of sample 6970 going around with a crippled SP bios bringing down to 1536. I hope this is the case.
 
Blacklash said:
I'm curious about what new feature set the 6970/6950 will bring to the table, if any.

Rumor has it the AMD cards in the wild now are purposefully "slowed". It will be interesting to see the reviews.
Click to expand...

Unlikely. Today's benches come from regular dudes who bought retail cards from a legit shop in Germany (which might be in trouble now...).
I can't imagine that AMD would ship crippled cards into retail and rely on the average Joe to unlock it later.
 
But the Average Joe isn't supposed to have the card, and (supposedly) the new Catalyst drivers come out tomorrow. Who knows?........
 
Understood, but ducati750ss used the driver that came with the card in a retail box on a retail CD afaik, Unlikely that that one is crippled, one would think.
 
Yah, I doubt we're suddenly going to see a huge boost in performance. As such, color me completely unimpressed. I really thought they were going to do a lot better with this card, but from the looks of it it's actually just competing against the GTX570. Waiting for proper benchmarks for the final judgment of course, but these latest rumors and benchmarks have left me disappointed.
 
I'm a little skeptical of some of Ducati's assertions, TBH. He claims to have purchased the card-which means he must have connections-but then proceeds to throw his source under the bus by benching it. Just weird.

Honestly though? My head is about to explode trying rationalize/internalize all these rumors.....
 
He (and others) bought the card at a local retailer who decided to sell the cards prematurely. Nothing mysterious about this.
 
If the 6970 does turn out to be a great card, I'd expect to see more overclocked GTX 580s with custom coolers popping up soon. Right now, I believe Sparkle is the only one with a custom cooler.
 
I'm skeptical as well. The GTX 580 is already on the table - the 6970 either needs to beat it at an equivalent or moderately lesser price point, or it needs to come close to it at a significantly cheaper price. I'd love to see a return to the ATI 4000 series pricing. Of course, we still have the 6990 to be aware of. I wish they would release it concurrently with the 6970 so we'd know if 2x 6970 or 1x 6990 would win on price and/or performance.
 
movdx said:
Well i believe that device ID 6718 tends to be(might be) an 6950 ;)

Further proof
From 10.12

AMD Radeon 6950 - HD 6900 Series" = ati2mtag_NICayman, PCI\VEN_1002&DEV_6718
AMD Radeon 6970 - HD 6900 Series " = ati2mtag_NICayman, PCI\VEN_1002&DEV_6719
Click to expand...
Now that's hilarious.
That guy just did all the benching stuff, we just spent a page discussing it, and it turns out it's a 6950 (not a 6970) sitting right where it belongs, in terms of performance.
 
x3sphere said:
If that doc hasn't been tampered with, it still points to a driver mismatch.

The INF is identifying the card as a 6950...
Click to expand...

Not sure what you mean. That document shows the 6970 as being the 6718, same model number for the card being used by ducati.
 
You must log in or register to reply here.
Back
Top