AMD Cayman info (or rumor)

[Xoulz]

I'm sure they are tweaked and have a better PPP (performance, price, power usage) in this round. But this number-game is the "gray zone" if making a lot of confusions.

They should,'t change the first digits. They cold add something to the last digits to indicate improvements/evolutions on the same "old" 40nm node.

This is going to be a big mess, if not bigger than G92, at least as confusing. Why can't they keep it real?

What confusion? I think any kid Xmas shopping this fall will be able to see the price/performance of these cards and purchase accordingly... since the 5000 series cards are EOL.. where is the confusion?


Or, are you trying to say, for Nvidia fans trying to argue on the forums, it's hard to do so, when cards are named 6800 series and are cheaper, cooler and better than GTX 400 series at their perspective price/performance ratios?

ATi could've called these cards Fusion 1000 series.. would that be too confusing for you too?

[god_43]

They are the successors to the 5700 series. The naming scheme was shifted to make room for more midrange cards. The 6970 will be the successor to the 5870.

oooohhh thanks, that makes so much more sense now!

[god_43]

i'm getting sick of repeating myself. there is nothing flawed about tsmc's 40nm process.

wasn't it a problem with the equipment? the process itself is/was sound right?

[Sn0wm@n]

Barts was a 5 shader part but tweaked if I'm reading the spec sheet correctly.
It seems Cayman will be the 4D shader reworked chip people were expecting.

Since Cayman has (currently rumoured) 1920 shaders vs Barts 1120 and they are beefed up as well, it would account for a greater performance difference than what you would expect on the same node.

IF the rumoured TDP of 300W is true for cayman and the 6870 is 150W max then it's likely that the clock speed of Cayman will be in the same sort of region.
Best case scenario is that it's almost twice the speed of the 6870.

Cayman also has a different tessellation solution and involves of chip loading of some sort.

I hope the 6970 is fast enough for 1920x1200 crysis at 60fps and it might just be plausible too that finally, a real crysis card exists

crysis 2 you mean ????

[Jowy Atreides]

crysis 2 you mean ????

That as well but Crysis 1 mainly. Mods need love <3

[Sn0wm@n]

wasn't it a problem with the equipment? the process itself is/was sound right?

the problem itself was with the transistors at the begining and then maybe equipment ... but overtime these problems get fixed and amd has proven over time to have competent engineer who can figure out ways to improve a product based on the same node .. lower power consumption etc... so i really cant see a problem with amd getting more out of 40nm 1 year later

That as well but Crysis 1 mainly. Mods need love <3

LOL those mods do need alot of gpu love

[Chumbucket843]

wasn't it a problem with the equipment? the process itself is/was sound right?

there were rumors of electromigration issues in the vias (via defects are actually fairly common.) but that was even before evergreen launched. this is the only rumor i heard and defect density has been good so yields are not an issue as long as the chip is designed well.

40nm has gotten a bad rep mostly because of the blame game. they did undershoot capacity by a lot but that happens a lot and it would have been extremely risky to invest more into 40nm with the way the economy was doing. almost every analyst had a gloomy forecast for the IC market so they played it safe.

[flippin_waffles]

what does intel have to do with gpus? they have never been a player in that market?

I should have said 'gpu' technology development. GPU development is a key concern going forward, as there is a huge amount of compute power locked up in a modern GPU.

Parallel programming is really taking off, since parallel has to be the future. There is nearly no headroom left for single threaded computing.

Just as importantly is the direction the culture is moving, and that's High Def. entertainment. Whoever is lagging there is at a pretty big disadvantage. Intel doesn't have the hardware experience AMD and Nv have, and they definitely don't have the software experience or software eco system that the other two have.

They've tried a few times and failed miserably.

[Hell Hound]

Dedicated gpu psu.

[zerazax]

You are talking about architectural changes and new generation. What do you mean by these words, are we getting a shrink, you think?

You have a good point there. these guys (both of them) are trying to repeat something similar to nVidia's G92-mess by using new numbers (68xx, 69xx, GTX 560, 580 etc ..) on th improvements (evolution) of "old" 40nm and fool consumers that they are getting a new generation.

I can't see why they are playing the number-game (68xx, 69xx, GTX 560, 580 etc ..) without a shrink to 22nm. They will soon run out of the number, in with this mess.

There is a lot of issues there, my friend. Why should they try to "fake" a new generation by new numbers, while it's a evolution/improvement based on the same node(40nm in this round).

I mean they should use the same (58xx, 59xx, GTX 462, 482, or something similar), because they don't have a shrink in this round.

Now you're just trolling

Did you care when RV670 -> RV770 occured on the the same node? Changes were pretty minor too, just the scale was bigger. Or RV770 -> 870? Even fewer architectural changes, but it was shrunk to 40nm

Hell, where was the outrage when R600 -> RV670 was given a 3xxx moniker despite being only a 55nm shrink?

Or how about G92 from 8800s to 9800s to the GT 200's?

If anytihng, AMD is MORE consistent with its naming:

2900s -> 3800's (shrink to 55nm)
3800 -> 4800 (architectural changes, scaling up)
4800 -> 5800 (shrink to 40nm)
5800 -> 6xxx (architectural changes, and Cayman looks to be scaled up)
Looks like tick-tock to me, and that 6000 -> 7000 will be based on NI but with a shrink to 28nm

What are you going to say when GTX580 is just Fermi, but fixed? Oh but sorry, its not a new architecture nor is it a shrink. But it's Nvidia so its okay right?

And since when was there a rule that a shrink is the requirement for a new generation name? It's AMD's problem if they run out of numbers - not yours. And frankly, a company is probably happy to have this "problem" since it means they're pushing out enough products.

Frankly I think you're just here trying to stir the pot

[LordEC911]

I don't think the larger size will be a real concern; 40nm has pretty much been mastered now. As for uArch, I don't think it'll be majorly different from Barts. Upgrading the uArch for Barts was enough of an investment, which means that Cayman will probably feature (on-chip) high-end/professional features (double-precise float-point, for example) and perhaps a new memory controller as well.

I'm looking forward to proper numbers and a real comparison between Cypress, Barts, & Cayman.

Barts also is using Redwood's MC instead of Cypress/Juniper, which supposedly saves ~50% PHY.
Which is part of the reason why the memory doesn't have much headroom even though they are 5Gbps chips.

Cayman will be 4d.

[spursindonesia]

i'm getting sick of repeating myself. there is nothing flawed about tsmc's 40nm process.

I'm not asserting it from technical standpoint, since i'm way out of my league in this matter technically. What i mean by "buggered" is that the process hasn't been as smooth & as forgiving in product availability & capability compared to the ones before it. There might be technical reasons (the via issue, the discrepancy beetween nVidia's simulation result & real world one), but i digress.

there were rumors of electromigration issues in the vias (via defects are actually fairly common.) but that was even before evergreen launched. this is the only rumor i heard and defect density has been good so yields are not an issue as long as the chip is designed well.

40nm has gotten a bad rep mostly because of the blame game. they did undershoot capacity by a lot but that happens a lot and it would have been extremely risky to invest more into 40nm with the way the economy was doing. almost every analyst had a gloomy forecast for the IC market so they played it safe.

Then that seems to be a quite good explanation, though i must say it's quite weird that nVidia's bigger GF 104's cards can be so widely available & easily moving down in price structure compared to the smaller Cypress's cards.

Thx for the insights anyway, appreciate your info & analysis there.

[Sn0wm@n]

nvidia lowering the pricing on its GF104 doesnt mean they are making a boat load of proffit ... it only means they know they wont win market share if they dont lower their prices

and how many days left till cayman antilles news???

[antiacid]

..maybe i missed something? but the new cards from ati....well they suck? except for 2 benchmarks, they are slower than the 5800 series by like 10-15%? i don't see why everyone is praising them so much?

and THIS ladies and gents, is why AMD did not advertise their 6870/6850 so much. John Doe sees the product ID go up, but performance going down. John Doe is puzzled. John Doe doesn't know that those new cards are meant to replace the 5770 and 5850, not the 5870/5970. John Doe thinks "AMD sucks ten thousand dicks!" until caymen XT shows up. Let's hope AMD's strategy pays off with a $/perf argument that appeals

[ToxAvenger]

John Doe doesn't install a $200 graphics card into his system and go hey, how come it's slower than the $375 card I bought 3 months ago, come on.

BTW, when is the last time AMD advertised anything?

[god_43]

and THIS ladies and gents, is why AMD did not advertise their 6870/6850 so much. John Doe sees the product ID go up, but performance going down. John Doe is puzzled. John Doe doesn't know that those new cards are meant to replace the 5770 and 5850, not the 5870/5970. John Doe thinks "AMD sucks ten thousand dicks!" until caymen XT shows up. Let's hope AMD's strategy pays off with a $/perf argument that appeals

im not john doe, i haven't really followed the 6xxx series news (5870 still works great), amd/ati's naming scheme has not changed for like ever....why would now be different? but as was pointed out to me before; i now know that these are mid-range cards and not 58xx series replacements.

[STEvil]

im not john doe, i haven't really followed the 6xxx series news (5870 still works great), amd/ati's naming scheme has not changed for like ever....why would now be different? but as was pointed out to me before; i now know that these are mid-range cards and not 58xx series replacements.

Actually they do replace the 5850/5870 in performance. A bit less in some situations, a bit more or a lot more in a couple with potential for increases in some of the slower results (mostly due to tesselation).

A 6870 is about $100 cheaper than a 5870 for me as well.. based on ncix.com pricing.

So you get basically identical performance to the mid-high mainstream performance available before for cheaper and unless you bought a 5850/5870 before the 6850/6870 were available, you dont have anything to complain about..

Had they called the 6870 a 6970 I would definately agree it would have been a bad naming, but as it is the naming scheme is basically fine. Would have been better to say 6840 and 6860.. could have put out 2gb versions called the 6850 and 6870.
'

On a side note, depending on how MLAA is done it could be ported to GPU's previous to the 6 series and even available on nV hardware possibly. That'd be nice.

[Lanek]

You can allready use it on HD5xxx, there's a tweak you can find on the net using the 10.10a ..

This is how look my CCC after apply the tweaks... ( 2x 5870 )

Cat AI options:


MLAA:


And yes, Nvidia could and will surely release his own version soon ...

[japamd]

The HD 6990 will bring untold performance to the table through the use of a pair of Cayman GPU cores and additional features we can’t divulge at this time.

Sideport will make a comeback?

Or support for 12 monitors in a single card?

[Jowy Atreides]

Sideport will make a comeback?

Or support for 12 monitors in a single card?

Havok fysics or whatever variant of whacky spelling makes fun of fooseex most.
Open source Phayeeesooox

Seriously though, I read some marketing bumnf on the 6 series having on card physics processing support on aria.

[Manicdan]

You can allready use it on HD5xxx, there's a tweak you can find on the net using the 10.10a ..

This is how look my CCC after apply the tweaks... ( 2x 5870 )

Cat AI options:

And yes, Nvidia could and will surely release his own version soon ...

awesome, what about 4800s?
i might have to try this out later.

[Lanek]

Not tested on 4000 ... and i doubt it will work .... Too much difference on the filterings used, but i don't know.

[zalbard]

awesome, what about 4800s?
i might have to try this out later.

OpenCL doesn't work well with them, does it?

[Manicdan]

OpenCL doesn't work well with them, does it?

is that what handles the MLAA? im pretty sure they dont have it

[Iconyu]

is that what handles the MLAA? im pretty sure they dont have it

Yeah it's shader compute, I imagine that you'd need 5.0 to do it fast enough.