AMD: R9xx Speculation

it would be nice, but this slide says otherwise ("50" and "70")...

slajd11c4g.png

What's most interesting about that is they go to great lengths to explain xx70, xx50, and xx30. No explanations for an xx90 would imply there is no xx90 in that picture. And there is likely no xx30 in that picture either as the explanation for a "30" is for a hypothetical "If an entry level part exists". As well trying to squeeze in 2 Antilles products above a speculative Cayman 6970 makes for an extremely crowded numbering scheme. So the subtext for that is that there are only xx70 and xx50 cards in that list. [EDIT - Also note how "70" is denoted for the highest performing part. So anything higher than that would be out of the question on the chart]

If we assume that it is talking about Barts, Cayman, and Antilles, that also confirms that each has their own seperate series. so 6x.., 6y.. and 6z... series. In other words due to the way it was created it would be weird if they did, for example.

6800 series
6900 series
6900 series

What would actually be somewhat hilarious is if Barts was 6600 series, Cayman is 6700 series, and Antilles was 6800 series. And Antilles wasn't a dual GPU codename afterall. :p EXTREMELY doubtful, however.

But that also leaves open the possibility of what No-X threw into the ring. That dual GPU cards would get something like a 6X50/6X70, where the X denotes a multi-GPU setup.

Anyway, if that's the case, 68xx, 69xx, 6Xxx, I'm still not liking it unless 68xx is a significant performance upgrade over 58xx. If not, then AMD is deliberately breaking their established naming scheme in order to take advantage of non-enthusiast gamers (who don't even benchmark their cards) when they don't have to. As long as 68xx is a significant performance increase over 58xx, I won't have a problem with it. Or if 68xx is the fastest single chip card but still isn't a signficant performance increase (denoting it's the best they could do with a single chip on the same process), I'd be somewhat OK with that.

Regards,
SB
 
God, i cant wait for this clusterf@#k of misinformation to end.

Same here! The level of secrecy AMD has managed to keep until now (and probably for some time still) just days away from launch is really impressive. Then again, we'll probably see the same thing all over again with Southern Islands (28nm), maybe even worse.
 
Last edited by a moderator:
I think there will be no two Antilles. There is no logical need for this. ;)

The main difference could be VRAM, since 4GB will be quite expensive there will still be a large market who only need 1GB effective VRAM while 2GB will be useful for eyefinity/high res/SSAA/newer games.

I speculate the same VRAM differentiation with Cayman Pro and XT. Barts XT and PRO will more than likely both be 1GB.
 
Link ->

6800_2.jpg


Catalyst 10.8 (Device ID) :
...
243,BARTS XT (6738),NI BARTS
244,BARTS PRO (6739),NI BARTS
245,WHISTLER XT (6740),NI WHISTLER
...
 
The same goes for iSomething, it's completely meaningless. Core i5 includes 32nm and 45nm parts; some with 2 cores, some with 4; some with integrated graphics, some without; some parts with a TDP of 18W, or 35W, or 73, 82, 87 or 95W; some with HyperThreading, some without; some with 3MB of cache, others with 4MB… or 8! How the hell is that not confusing? iSomething makes an otherwise fairly understandable naming scheme completely obscure. In fact, the best thing to do if you want to actually "navigate" Intel's lineup is to completely ignore the iSomething part and focus on the following number.

So if Intel went ahead and be rid of iX, it's some how clearer? How?

You know for sure that i7 has larger cache and better turbo regardless of what core they are using. i5 ALL have 4 threads and Turbo, regardless core counts. etc..

Without it how do you generalize the difference? If a customer asks you can simply tell them that. AMD has two-part system, you'll have to explain to them why X3/X4 (tri-core) has two brand names. Any reasonable person can figure out i7 is better just by looking at the name. Can you say the same for Athlon/Phenom? Isn't that PURELY MEANINGLESS? Why do X3 have higher number than X4? Is it really better in ALL circumstances?

As for GTX/GTS, all you have to do is tell the customer GTX is for performance parts with excellent gaming experience (and huge power draw), GTS is for casual market. With the current AMD numbering how do you explain it, especially when there's multiple generations on the shelves?
If a 5850 is a performance part, is 5830/4850/5770 a performance part or was it once a performance part just reached EOL ?-- which means higher power draw, fewer features for the same price. It's hard to tell for the untrained eyes if there's only numbers, isn't it?

Why do you praise AMD's Phenom/Athlon approach while condemning Intel/NVIDIA's?
 
Well, GPU-Z maybe correctly is reading device ID, but maybe the 6800 part is just a guess?
Or we are left with possibility of barts being 6770&6850, Cayman 6870&6950 ...
2 screens, one with Barts id, the other with Cayman id, both saying 6800 series
 
Well, GPU-Z maybe correctly is reading device ID, but maybe the 6800 part is just a guess?
Or we are left with possibility of barts being 6770&6850, Cayman 6870&6950 ...
2 screens, one with Barts id, the other with Cayman id, both saying 6800 series

Why is everyone so desperate to downgrade Cayman to a 68x0 part? Do you guys want it to suck that badly?
 
Well, GPU-Z maybe correctly is reading device ID, but maybe the 6800 part is just a guess?
Or we are left with possibility of barts being 6770&6850, Cayman 6870&6950 ...
2 screens, one with Barts id, the other with Cayman id, both saying 6800 series

How in the world would it read device ID correctly for caymanXT and reading 6800 series is just a guess ??
 
How in the world would it read device ID correctly for caymanXT and reading 6800 series is just a guess ??

because it CAN read the DeviceID, but it's w1zzard's guess as to which part belongs to which deviceID.
For new card reviews, a special version of GPU-Z is released by AMD to reviewers to correctly identify the cards.
 
because it CAN read the DeviceID, but it's w1zzard's guess as to which part belongs to which deviceID.
For new card reviews, a special version of GPU-Z is released by AMD to reviewers to correctly identify the cards.
When I was creating ComputeMark, "DeviceID" and "Radeon <whatever> Series" string were very easy to get throught the WMI (ComputeMark uses WMI). Maybe Wizz uses the WMI too.

I wonder when some1 with new Radeonz will try ComputeMark and Submit computing score to the web ;-)
 
So if Intel went ahead and be rid of iX, it's some how clearer? How?

You know for sure that i7 has larger cache and better turbo regardless of what core they are using. i5 ALL have 4 threads and Turbo, regardless core counts. etc..

Really? Core i7-620M has 4MB of cache, Core i5-750 has 8MB. Besides, how are people supposed to know that i5 means 4 threads (except it can be 2 cores + HT or 4 cores without HT, which is entirely different) and Turbo, but that pretty much everything else can vary wildly? Remove iSomething, focus on the numbers, and remember that bigger is better: that is simpler. As far as I know, the number series are consistent, the 7xx CPUs all share the same features, as do the 8xx ones, etc.

Without it how do you generalize the difference? If a customer asks you can simply tell them that. AMD has two-part system, you'll have to explain to them why X3/X4 (tri-core) has two brand names. Any reasonable person can figure out i7 is better just by looking at the name. Can you say the same for Athlon/Phenom? Isn't that PURELY MEANINGLESS? Why do X3 have higher number than X4? Is it really better in ALL circumstances?

The dual brand (Athlon/Phenom) tells enthusiasts whether there is any L3, except for mobile parts, sadly, but obviously most people don't know any of that, or even what L3 is. So yes, that's confusing, but my point was just about the X2/3/4/6 part of the naming scheme.

Then again, lately AMD has been deemphasizing all those brands and just focusing on the much, much simpler VISION brand. I also suspect that as they introduce new products next year (Bobcat, Llano, Bulldozer) current brands may go away, or at least some of them.

As for GTX/GTS, all you have to do is tell the customer GTX is for performance parts with excellent gaming experience (and huge power draw), GTS is for casual market.

And without those prefixes, you could just say: "460 and over is high-performance, 440/450 is mainstream". Or better yet, and since those distinctions are pretty arbitrary anyway, you could just tell them "bigger is better" and provide additional information about specific SKUs upon request.

With the current AMD numbering how do you explain it, especially when there's multiple generations on the shelves?
If a 5850 is a performance part, is 5830/4850/5770 a performance part or was it once a performance part just reached EOL ?-- which means higher power draw, fewer features for the same price. It's hard to tell for the untrained eyes if there's only numbers, isn't it?

Again: bigger is better, knowing that the first digit is for the generation; which, by the way, is true for NVIDIA as well. And prefixes don't change that: what is better, 9800 GTX, or GTS 450?

Why do you praise AMD's Phenom/Athlon approach while condemning Intel/NVIDIA's?

I don't praise the Phenom/Athlon approach, just the X2/3/4/6 part. Plus AMD CPUs seldom have disabled features, so that helps.
 
The main difference could be VRAM, since 4GB will be quite expensive there will still be a large market who only need 1GB effective VRAM while 2GB will be useful for eyefinity/high res/SSAA/newer games.

I speculate the same VRAM differentiation with Cayman Pro and XT. Barts XT and PRO will more than likely both be 1GB.

Why do people assume frame buffer is only for high resolution, Eyefinity, SSA and newer games?

I just need to have a handful images open in Photoshop before maxing out a 1GB frame buffer and things crawl to a halt. Of course, I could work on less images at the time.
 
Device id should be accessible at very low hardware level (it is probably located in the content of a physical register), at least for PCI cards it is. I am almost certain that this is not the case for the card's series.
 
Why is everyone so desperate to downgrade Cayman to a 68x0 part? Do you guys want it to suck that badly?



'Cos it will leave the impression that the performance jump between 5870 and 6870 is quite noticeable. It's not "downgrade". :mad: And, by the way, is Barts the successor of Cypress? But if 6870 (Barts) sucks and is slower than 5870, then AMD will have a hard time to explain why and they will lose a lot of sales. Some of them will go to NVidia. 'Cos they have always had better reputation. :oops:
 
Last edited by a moderator:
Back
Top