• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

ATI cuts 6950 allocation

Rroff said:
I'd love to think that was really the 6950 and the 6970 had say 1920 SP... but as I said before I have a good source who says the 6970 is 1536 SP.

EDIT: Or maybe I "made that all up" like I "normally" do and just happened to get a fairly arbitary number correct? :confused:
Click to expand...

Rroff said:
We can't count it out yet... if that did turn out to be the 6950 (which seems unlikely) then the 6970 would quite feasibly be in the 30+% faster than 580 ball park.

As I said before tho, my source is the same as my info on the 58xx cards - which anyone can look back and find I was correct about.
Click to expand...

Again I'll point out, firstly 1536 was in NO WAY arbitrary, at all, secondly LOTS of people were guessing 1536 even before the magical really early date you said it(weeks/months after others had guessed so).

Your source on the 5870, you say stuff like that, then say go look it up, then no one does, and in the future you'll claim that as "proof" you were right, as you were now.

THe 5870, was expected by EVERYONE to be a 256bit 1600 shader card, what magical new piece of information did you guess at that no one else did?

PS my claims(which I can also back up) were guessing the 6870 would be around 255mm2(literally couldn't have been more spot on there ;) ), the 6970 would be 394mm2(was marginally out) and that the transistor count would be between 2.6 and 2.7billion on it. I would have to check but I also assumed 900Mhz, circa 250W, 2gb mem and a 1920 shader count(which the 1536 4d shaders would be the equivilent of.

As for the 1536 being arbitrary, your guess came WELL after Charlie released the 3d structure info, and LOADS of people guessed as simple as 1920/5 x 4 = 1536. Theres a speculation of specs thread on SA that predates your 1536 guess, with a LOT of people guessing 1536 shaders, and there was guessing in other threads WELL before that.
 
Last edited:
6970computemarkxmc9.jpg


88761476.png


one says 1600 SP and the other says 1536 SP, also look that the release date and texture fillrate.
 
Last edited:
Yeah Gareth we have been over this, he has updated to a test build of GPU-Z that shows the correct specs for the 6970, that would be the one showing 1536 SP's. And the official release date is the 15th.
 
drunkenmaster said:
Again I'll point out, firstly 1536 was in NO WAY arbitrary, at all, secondly LOTS of people were guessing 1536 even before the magical really early date you said it(weeks/months after others had guessed so).

Your source on the 5870, you say stuff like that, then say go look it up, then no one does, and in the future you'll claim that as "proof" you were right, as you were now.

THe 5870, was expected by EVERYONE to be a 256bit 1600 shader card, what magical new piece of information did you guess at that no one else did?
Click to expand...

I don't recall seeing anyone claiming 1536 SP (other than where I got my info from) before I said it - but happy to be proved wrong. Certainly no one here was making such claims but I can believe it may have been mentioned months ago elsewhere as these guys would have known for months.

When I was claiming 1600SP for the 5870 the only people making that claim were myself and charlie pretty much everyone else was saying different numbers until charlie came out with it and then suddenly everyone changed their tune.

EDIT: To be fair I think you were claiming 1600 SP would be the logical step forwards about that time as well as IIRC you've always held the same view RE architecture changes for ATI/nVidia generationally.
 
Rroff said:
I don't recall seeing anyone claiming 1536 SP (other than where I got my info from) before I said it - but happy to be proved wrong. Certainly no one here was making such claims but I can believe it may have been mentioned months ago elsewhere as these guys would have known for months.

When I was claiming 1600SP for the 5870 the only people making that claim were myself and charlie pretty much everyone else was saying different numbers until charlie came out with it and then suddenly everyone changed their tune.

EDIT: To be fair I think you were claiming 1600 SP would be the logical step forwards about that time as well as IIRC you've always held the same view RE architecture changes for ATI/nVidia generationally.
Click to expand...

Because a general doubling has happened every generation since, errm, I'm not sure it was so long ago, was it the GF 256 that was the first 1 pipeline card , then 2, 4, 8, 16, then we got a little inbetween dx9 and dx10 and adding shaders rather than pipelines for AMD. Even the 2900xt to 4870 wasn't unexpected, 320 to 800, the 2900xt was clearly cut down to fit at 80nm(it was meant for 65nm) it was probably a 400shader part to start with.

But you're basically saying the only people claiming it was me, and this really well known guy who publishes his info in the interweb for everyone to see. I'm not sure you can claim public knowledge released by Charlie as "you were right about it".

The whole 32/28nm thing will throw a spanner in the works, realistically you'd have expected the equivilent of 2400 shaders at 32nm and 3200 at 28nm. Nvidia will WANT 1024 shaders in their next gen, but will they for the first time in a long while use their heads and go slightly smaller to bring yields up, who knows, head says they can't be stupid enough to do it AGAIN, heart says, Dear leader really is that arrogant.

http://www.semiaccurate.com/forums/showthread.php?t=3316

First mention in the very first post of a 1536 shader count, its not arbitrary, I hoped/thought they might increase number of clusters per SIMD to reduce overhead but staying at 16x clusters per simd has many advantages, and 4x16= 64, multiples of 64 were not randomly plucked from thin air.

I'm about 98% sure other people had mentioned 1536 as a 4d equivilent of 1920 shaders in the days/weeks before that but honestly I can't be bothered to find it.
 
Last edited:
I added an upto in there as theres no guarantee the new architecture is more efficent every time.

Also I'm still thinking the guy is benching on "balanced" power profile or something or the drivers really do suck as these numbers are approx. 20% under what we should be seeing for a 1536SP card.
 
drunkenmaster said:
Because a general doubling has happened every generation since, errm, I'm not sure it was so long ago, was it the GF 256 that was the first 1 pipeline card , then 2, 4, 8, 16, then we got a little inbetween dx9 and dx10 and adding shaders rather than pipelines for AMD. Even the 2900xt to 4870 wasn't unexpected, 320 to 800, the 2900xt was clearly cut down to fit at 80nm(it was meant for 65nm) it was probably a 400shader part to start with.

But you're basically saying the only people claiming it was me, and this really well known guy who publishes his info in the interweb for everyone to see. I'm not sure you can claim public knowledge released by Charlie as "you were right about it".

The whole 32/28nm thing will throw a spanner in the works, realistically you'd have expected the equivilent of 2400 shaders at 32nm and 3200 at 28nm. Nvidia will WANT 1024 shaders in their next gen, but will they for the first time in a long while use their heads and go slightly smaller to bring yields up, who knows, head says they can't be stupid enough to do it AGAIN, heart says, Dear leader really is that arrogant.
Click to expand...

Thing is its quite clear to you because you have a decent level of tech knowledge and probably there were other people making the same claims I don't haunt every tech forum... but at the time the vast number of people were making some very varied claims about the SP count on evergreen, very few people were claiming 1600 SP at that time. My information honestly didn't come from Charlie, but I do believe he did post about it a couple of days before I came out with it.

EDIT: I also doubt we will see 2000+ SP cards from AMD theres a number of reasons some relating to pipeline latency/queue depth which means the efficency drops quite a bit away from a linear performance increase as you add extra SPs (I'm not really an expert on this but I can comprehend the general ghist of the issue), which becomes an issue at those kinda sizes, I believe this is why they have been driving to increase efficency over raw hardware scaling with the 6 series.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom