AMD: R9xx Speculation

[caveman-jim]

Ahh I guess that makes sense now.

BTW, what did Dave save this generation? Ever since the HD 48xx cards, the thing Dave has saved has remained a mystery.

I think he saved Christmas for AMD. The following is all guesswork, so ready your salt shakers.

When Ibiza became Cayman on 40nm, it became a big, un(w/y)ieldly kinda thing. Relative to RV770, to keep perspective. So it was going to be more expensive than preserving the power target of 32nm Ibiza would allow. So the power budget went up, to give the performance targets a hope of being achieved. Once this happened, the market target changed - high end enthusiast ~$300+. So the clever boys put their clever hats on, and got to being clever, in the clever manner only they know how.

The result was - extend the Juniper line up, give Cypress a tweak into being the ~$200 hero, and attack the high end from the flank with a 4-barreled salvo to the bread-basket. Thus Barts was born, and sent into the breach against GF104, aided by Cypress who was tracking the GF100, now under pressure from GF110.

At this point, Dave saved Christmas.

The guy who did Secret Santa in Markham left, and nobody wanted to do it - I mean seriously, who wants to be that guy? "Pick a name, don't blow your budget but try to be original! ATI Radeon shirts don't count." Dave, having sat around since he won the high end in September 09, the ultra high end until.... well, pretty much now, decided to give it a whirl this year, as long as someone else took it up next year.

And so Christmas was saved!

 

[Unknown Soldier]

I think he saved Christmas for AMD. The following is all guesswork, so ready your salt shakers.

When Ibiza became Cayman on 40nm, it became a big, un(w/y)ieldly kinda thing. Relative to RV770, to keep perspective. So it was going to be more expensive than preserving the power target of 32nm Ibiza would allow. So the power budget went up, to give the performance targets a hope of being achieved. Once this happened, the market target changed - high end enthusiast ~$300+. So the clever boys put their clever hats on, and got to being clever, in the clever manner only they know how.

The result was - extend the Juniper line up, give Cypress a tweak into being the ~$200 hero, and attack the high end from the flank with a 4-barreled salvo to the bread-basket. Thus Barts was born, and sent into the breach against GF104, aided by Cypress who was tracking the GF100, now under pressure from GF110.

At this point, Dave saved Christmas.

The guy who did Secret Santa in Markham left, and nobody wanted to do it - I mean seriously, who wants to be that guy? "Pick a name, don't blow your budget but try to be original! ATI Radeon shirts don't count." Dave, having sat around since he won the high end in September 09, the ultra high end until.... well, pretty much now, decided to give it a whirl this year, as long as someone else took it up next year.

And so Christmas was saved!

Pure speculation!

 

[gongo]

I just had a thought....if Cayman 40nm is as it is.....what made AMD go ahead with only 5-25% increase in perf over Cypress....sure its better than RV790...but Cypress/RV770 were great successor chips....kinda aimed low with Cayman eh? Maybe if time was not wasted in porting back to 40nm...it could have been released earlier and cheaper? VLIW4 drivers would have been better?

Then again...VLIW4 Cayman sounds very experimental, aiming low bit...reading Rage3D comments is confusing as well...VLIW4 as i know is more suited for compute tasks....so...

Looking Forward - VLIW-4 vs. VLI5

The Northern Islands story is an interesting one, where the decision to stick with 40nm caused a few interesting knock-on effects. The AMD engineers, seeing how 32nm was going to be more expensive per transistor, looked to the possibilities afforded by 40nm and came back to Product Management with a suggestion. No, not that kind of suggestion. This was the birth of Barts - a last minute addition to the line up. Let us do Cayman like this, they said, and we can give you four more GPUs to go with it! AMD is strongly hinting that the rest of Northern Islands will be VLIW-5. It's a good architecture for gaming and media use, which is the primary focus of the rest of the NI range. Power Tune is too good a technology to overlook for long, though.

http://www.rage3d.com/reviews/video/amd_hd6970_hd6950_launch_review/index.php?p=10

 

[eastmen]

I just had a thought....if Cayman 40nm is as it is.....what made AMD go ahead with only 5-25% increase in perf over Cypress....sure its better than RV790...but Cypress/RV770 were great successor chips....kinda aimed low with Cayman eh? Maybe if time was not wasted in porting back to 40nm...it could have been released earlier and cheaper? VLIW4 drivers would have been better?

Then again...VLIW4 Cayman sounds very experimental, aiming low bit...reading Rage3D comments is confusing as well...VLIW4 as i know is more suited for compute tasks....so...



http://www.rage3d.com/reviews/video/amd_hd6970_hd6950_launch_review/index.php?p=10

Cayman we have today may be the cayman we would have had with 32/28nm but who is to say that Cayman was supposed to be a high end chip ?

For all we know Cayman with faster clocks to do 32/28nm could have been the replacement for barts price bracket ($150-$230) but due to being on 40nm it wasn't clocked as high and was bigger than they wanted and thus it was only 25% faster.

Look at it this way, with cayman on 32nm being the size of barts. Would anyone be upset if cayman launched using the same power or less than barts and offered the performance its currently offering ?


The process guys have really screwed us this year

 

[caveman-jim]

@gongo Sorry you found it confusing. The only VLIW-4 based product is Cayman, it's the only one to get the fancy new memory controllers and RBE's, too.

AFAIK, Apply salt, etc.

 

[caveman-jim]

For all we know Cayman with faster clocks to do 32/28nm could have been the replacement for barts price bracket ($150-$230) but due to being on 40nm it wasn't clocked as high and was bigger than they wanted and thus it was only 25% faster.

Could have been the 6800 name, $200-$300 price bracket. Not a bad save, going to Barts, 6900, etc.

 

[gongo]

Cayman we have today may be the cayman we would have had with 32/28nm but who is to say that Cayman was supposed to be a high end chip ?

For all we know Cayman with faster clocks to do 32/28nm could have been the replacement for barts price bracket ($150-$230) but due to being on 40nm it wasn't clocked as high and was bigger than they wanted and thus it was only 25% faster.

Look at it this way, with cayman on 32nm being the size of barts. Would anyone be upset if cayman launched using the same power or less than barts and offered the performance its currently offering ?


The process guys have really screwed us this year

But 32nm does not mean it will be cheaper...if my reading is correct.
So Cayman @ 32nm = Cayman @ 40nm with a smaller die, lower tdp (some savings can be had here) and maybe another 50mhz increase to core clocks (gddr5 has to be limited @ nearly 5ghz at this time).....that gives another 5fps increase over 40nm Caymans @ $249/349? Still sounds low in raw fps but i guess AMD can use the perf/watt/die as a strong point...

If SIMD were cut...we could be looking at a 22/26 configs....with core clocks @ additional 40mhz...that would sound more powerful....

 

[Eolirin]

Yea maybe it was drivers but i dont believe 3 weeks is enough time to change clocks. The cards have to enter production ~1 month before launch. This is to first stockpile some quantity of cards for launch day, then the time for testing, packing, shipping(if by sea then its give or take a week total shipping time) and then cards usually reach retailers a few days in advance. So they would have had to make changes to a LOT of cards which were already produced by the time(probably in thousands if not tens of thousands)

Er, that is what I said right?

 

[eastmen]

But 32nm does not mean it will be cheaper...if my reading is correct.
So Cayman @ 32nm = Cayman @ 40nm with a smaller die, lower tdp (some savings can be had here) and maybe another 50mhz increase to core clocks (gddr5 has to be limited @ nearly 5ghz at this time).....that gives another 5fps increase over 40nm Caymans @ $249/349? Still sounds low in raw fps but i guess AMD can use the perf/watt/die as a strong point...

If SIMD were cut...we could be looking at a 22/26 configs....with core clocks @ additional 40mhz...that would sound more powerful....

Your looking at it wrong.

1) Cayman at 32nm would be the size of barts. Which means that the original plan would have been to have cayman slot in at the sub $250 price point. So the 6970 would be $250.

2) A smaller die means more working chips per wafer which means its less expensive , lower tdp means savings across all board components also.

3)Why would it need more fps at the $150-$250 price point. I'm sure if 32nm was out this year AMD would have had a high end gpu product out to take higher price points and those could have been around cypress size

 

[UniversalTruth]

 

[Arnold Beckenbauer]

Old fakes aren't funny.

 

[gongo]

@gongo Sorry you found it confusing. The only VLIW-4 based product is Cayman, it's the only one to get the fancy new memory controllers and RBE's, too.

AFAIK, Apply salt, etc.

That is whats confusing...will AMD support the single VLIW4 product in their lineup? So more renaming for the HD6000 series..not good. I like to hear about the new RBE...what it does differently? The new MC i guess is the one driving GDDR5 to ...really fast...and overclocking is fairly easy too...

BTW did AMD really fixed the capacitor buzzing with Cypress..as i read in reviews...because i can hear the buzzing with Cayman...launch FF14 benchmark, its there when the screen is black and loading...oh my.

 

[UniversalTruth]

That is whats confusing...will AMD support the single VLIW4 product in their lineup? ...

Yes, they are supposed to do so.

 

[mczak]

That is whats confusing...will AMD support the single VLIW4 product in their lineup?

I'd be very surprised if we don't see other chips using VLIW4 down the road.

 

[Squilliam]

I think he saved Christmas for AMD. The following is all guesswork, so ready your salt shakers.

When Ibiza became Cayman on 40nm, it became a big, un(w/y)ieldly kinda thing. Relative to RV770, to keep perspective. So it was going to be more expensive than preserving the power target of 32nm Ibiza would allow. So the power budget went up, to give the performance targets a hope of being achieved. Once this happened, the market target changed - high end enthusiast ~$300+. So the clever boys put their clever hats on, and got to being clever, in the clever manner only they know how.

The result was - extend the Juniper line up, give Cypress a tweak into being the ~$200 hero, and attack the high end from the flank with a 4-barreled salvo to the bread-basket. Thus Barts was born, and sent into the breach against GF104, aided by Cypress who was tracking the GF100, now under pressure from GF110.

At this point, Dave saved Christmas.

The guy who did Secret Santa in Markham left, and nobody wanted to do it - I mean seriously, who wants to be that guy? "Pick a name, don't blow your budget but try to be original! ATI Radeon shirts don't count." Dave, having sat around since he won the high end in September 09, the ultra high end until.... well, pretty much now, decided to give it a whirl this year, as long as someone else took it up next year.

And so Christmas was saved!

To think I had a part in Dave's save for 2010! I am incredibly humbled by these events. See, I sent Dave a package of goodies not too long ago which included a Black Mesa pass as well as other goods. Since the secret Santa has to give out more gifts than everyone else in order to make sure everyone gets one, Dave suddenly had a surplus of giftable items for him to distribute around the office. That enabled Dave to step into the breach and become the Markham secret Santa when the previous guy absconded from his role.

Now going back to that Black Mesa pass. Dave obviously knew that since his Prius has a mileage distortion field it is able to cross dimensions in order to conserve gasoline. Everyone knows that the hybrid systems don't do squat for mileage without it, which is why only pure hybrids work. Anyway, being the technical fellow he is and with his trusty Black Mesa pass afixed firmly to his vehicle, he crossed into the Half-Life Universe in order to access secret technology from Black Mesa which enabled him to go back in time and hand ATI the secrets to advanced GPU development and thus the R300 was born which was the ultimate Dave save which saved ATI all the way back then. In addition to this Dave is the only known person to have heard Gordon Freemans voice and it was his advice which enabled ATI to come up with the VLIW 4 technology seen in Cayman.

Phew, Dave Saves again.

 

[caveman-jim]

That is whats confusing...will AMD support the single VLIW4 product in their lineup? So more renaming for the HD6000 series..not good. I like to hear about the new RBE...what it does differently? The new MC i guess is the one driving GDDR5 to ...really fast...and overclocking is fairly easy too...

I fully expect to see more VLIW4 products in the future.

BTW did AMD really fixed the capacitor buzzing with Cypress..as i read in reviews...because i can hear the buzzing with Cayman...launch FF14 benchmark, its there when the screen is black and loading...oh my.

That can be a number of things, from PSU to card, so it's hard to say whether its a smoothing issue at the VRMs, not enough lacquer, or a need for a smoothing cap across the PCI-E power lines on the PSU. /shrug I don't hear the squeal/buzz, unless I'm running 6000FPS+ on Filter Tester.

 

[digitalwanderer]

Anyone else just feeling a bit disappointed overall with this launch?

 

[swaaye]

Anyone else just feeling a bit disappointed overall with this launch?

Disappointment is located over in the review thread.

 

[nutball]

Anyone else just feeling a bit disappointed overall with this launch?

Not really. It's a solid product. It'll sell by the truck-load and make AMD a lot of money, which is what they need if the want to get the company out of the financial mire it's been in for a while now.

What is nice though is that the clatter from the AMD Technical Marketing Division (Netherlands Branch) has quieted down a bit recently.

 

[Squilliam]

Anyone else just feeling a bit disappointed overall with this launch?

Dave saves yet again and you feel disapointed