AMD: R8xx Speculation

[rjc]

TR added this (emphasis mine):

Those who go shopping for a laptop later this year may see Mobility Radeon HD 5165 and 5145 graphics processors pop up in some spec sheets. No, those aren't typos—but they're not DirectX 11 mobile GPUs, either. Rather, the 5165 and 5145 are faster-clocked versions of 55-nm, DirectX 10.1 Mobility Radeon HD 4800 and 4600 GPUs, respectively. Although we surmised that these products were somehow tied to the 40-nm shortages, AMD said it created them before the yield issues cropped up. The firm's partners simply wanted faster 55-nm parts with better brand names. Go figure.

Still lame.

Looking at the 5165 spec page here and the 5145 page here on the AMD site it looks like the 4670/4550/4350 parts are getting new waistcoats rather than the 4800 and 4600 as TR suggests.

The above has to be cost based, if RV710 is on an older process and smaller die is looking to be around say 30% cheaper to produce than Cedar and say 20-25% cheaper for the RV730 over Redwood will likely be used for cost down products by the OEMs. As anandtech article said on the previous page these two segments are combined 85% of units sold.

Also remember round Computex time AMD seemed to be investigating using UMC for the RV710 product which seemed strange as Park/Cedar production was only 6 months away.

 

[Tchock]

I guess ATI has decided to yield the mobile (programming) field as they will no longer offer any mobile part with DP support, if anyone want DP they instead will have to go green or try and find the older 4800 series.
While many (including ATI/AMD) may be dismissive of the demand for dp in the mobile space, I honestly think this is a key failure for ATI/AMD and one of the long standing reasons why ATI/AMD will always play 2nd fiddle to NV when it comes to what many programmers desire (a greatly expanded LCD- lowest common denominator).

Well considering nV has no DP mobile parts (at all) now and Fermi mobile might see light when Northern Islands desktop come into play... would the LCD matter that much?

Plus, the prospect of a 59XX (Cypress) Mobile is still there, perhaps after stable yields, extensive binning and probably soft caps on what the GPU can do etc.

Renaming is... ugh. Good part is that I wouldn't expect the decision to be based on cost for now/future, this seems to be more of a response to the nVidia chips for Calpella- that AMD can ship "5" series stuff now ("then").

 

[FrameBuffer]

Well considering nV has no DP mobile parts (at all) now and Fermi mobile might see light when Northern Islands desktop come into play... would the LCD matter that much?

Plus, the prospect of a 59XX (Cypress) Mobile is still there, perhaps after stable yields, extensive binning and probably soft caps on what the GPU can do etc.

Renaming is... ugh. Good part is that I wouldn't expect the decision to be based on cost for now/future, this seems to be more of a response to the nVidia chips for Calpella- that AMD can ship "5" series stuff now ("then").

I was under the impression that nV GF200M based mobile parts had double precision ? Damn it so much for getting a mobile for programming (GPGPU/DC/OCL with DP-FP) with decent performance.

 

[Psycho]

Edit: AMD Ships 2 millionth DX11 GPU - these figures better not include sales of the above 2 chips....

Wonder how much of those 1.2 mil. new units since mid december are redwood/cedar being counted now and not then.

 

[mczak]

I was under the impression that nV GF200M based mobile parts had double precision ? Damn it so much for getting a mobile for programming (GPGPU/DC/OCL with DP-FP) with decent performance.

nvidia has afaik exactly one chip which supports DP at all, and that is gt200 (make that two if you count gt200b separately). So no there aren't any mobile nvidia chips supporting DP.
We don't even know yet if Fermi derivatives actually will support DP, though it looks like the assumption is yes.

 

[mczak]

TR added this (emphasis mine):
Still lame.

It also makes for odd numbing scheme, since the 5165 will certainly be a much faster part than the 5400 series. AMD should just say no to renames, I doubt it really helps the OEMs much anyway.

And the very bad decision of a (G)DDR3-option for Mobility HD 5850 is also missing:
So this notebook with probably 1/4 of desktops HD5850 BW was no writing mistake: http://www.comdeal.de/PC-Systeme/NOT...Wn::81274.html

Yeah bad choice (though you should rather say less than 1/2 of bandwidth of desktop HD5750, it doesn't really make sense to compare it to desktop HD 5850).
The specs are a little odd in that sense that some of the models have clearly defined clocks / memory type (even for some low end parts), and others can be anything the OEM likes. I think that slightly varying clocks isn't that bad, but if it has only half of the performance because bandwidth is cut by more than a factor of two, that really shouldn't run under the same name.

 

[FrameBuffer]

nvidia has afaik exactly one chip which supports DP at all, and that is gt200 (make that two if you count gt200b separately). So no there aren't any mobile nvidia chips supporting DP.
We don't even know yet if Fermi derivatives actually will support DP, though it looks like the assumption is yes.

damn.. so looks like I have to either buy old (4870 mobile) or wait longer to fulfill my needs.

**yet another reason I hate renaminng trying to decifer stupid mobile vs desktop variants.. I don't consider myself and uneducated potential consumer I try to do as much homework as can be (reasonably) expected and even I get confused. I really do feel sorry for the jane and joe consumers out there**

Ughh
GTS 260 Mobile vs GTX 260 Mobile
http://forums.nvidia.com/index.php?showtopic=108560

GTX 260M is compute 1.1.
GTS 260M is compute 1.2.

I would honestly think that even those who has some sort of resemblance of GPU model clamature would expect that GTX to be better and more feature packed the the GTS as they typically are in the desktops.

 

[mczak]

GTX 260M is compute 1.1.
GTS 260M is compute 1.2.

I would honestly think that even those who has some sort of resemblance of GPU model clamature would expect that GTX to be better and more feature packed the the GTS as they typically are in the desktops.

Well it's at least better as in "faster" (though with twice the TDP). But you're certainly not the first to notice the naming scheme (what scheme???) is a mess, and the renaming (sadly AMD does it again too this generation for some mobile chips) doesn't help matters.

 

[Silent_Buddha]

ATI Mobility Radeon™ HD 5165/5145 Graphics... looks like ATI is back to its old renaming antics

Ugh, that's just really really horrible. To say I'm disappointed would be a huge understatement.

Although I'm sure the same people that kept telling me that it was fine for Nvidia to do it will also tell me that it's perfectly awesome that ATI are doing it...

Horrible, absolutely horrible...

Regards,
SB

 

[Squilliam]

Ugh, that's just really really horrible. To say I'm disappointed would be a huge understatement.

Regards,
SB

It seems wierd they would even do it. The name mobile 4xxx makes more sense to these products from my perspective than a lower marketing denominator than their lowest DX11 chips.

HD 4xxx makes sense relative to HD 5xxx from a performance standard and you would expect a high end previous gen product to outclass a mid end current generation product so that makes sense. What doesn't make sense is renaming a high end product below their low end even IGP products.

I think they should find a naming scheme which works and makes sense for everyone involved and then stick with it, no matter what the OEMs are asking. That way someone walking into the store in 3 months time can look at the products on offer and understand WTF the differences between the laptops actually means.

I can't bring myself to condemn them severely as it looks like they are shooting themselves in the foot but I still feel at least a little disapointed.

So what would make sense in terms of a naming convention?

HD first group = relative performance to X gpu and second group = feature set. Lets take the HD 5750 as the benchmark GPU.

HD 5870 = 2111E or 210% relative performance direct X 11 and Eyefinity.

HD 5750 = 1011E or 100% relative performance direct X 11 and Eyefinity

HD 4870 = 1310 or 130% relative performance direct X 11

That should last them longer and it solves the problem of having to rename things as the performance comes first.

Then lets say the 6 series is released with 3D support and more performance.

HD 6870 = 3211E3D or 320% relative performance Direct X 11 3D

HD 6670 = 13011E3D or 130% relative performance Direct X 11 3D.

Once people become familiar with it, they will always know that the new laptop has X% more 3D performance than the one they are replacing.

Edit: In addition to this they can drop the feature monikors once they become standard, so they don't need the E for example next generation.

 

[Silent_Buddha]

So what would make sense in terms of a naming convention?

HD first group = relative performance to X gpu and second group = feature set. Lets take the HD 5750 as the benchmark GPU.

HD 5870 = 2111E or 210% relative performance direct X 11 and Eyefinity.

HD 5750 = 1011E or 100% relative performance direct X 11 and Eyefinity

HD 4870 = 1310 or 130% relative performance direct X 11

That should last them longer and it solves the problem of having to rename things as the performance comes first.

Then lets say the 6 series is released with 3D support and more performance.

HD 6870 = 3211E3D or 320% relative performance Direct X 11 3D

HD 6670 = 13011E3D or 130% relative performance Direct X 11 3D.

Once people become familiar with it, they will always know that the new laptop has X% more 3D performance than the one they are replacing.

Edit: In addition to this they can drop the feature monikors once they become standard, so they don't need the E for example next generation.

That's all a bit too complicated and appears to get even more complicated with each new generation.

I don't see a problem with mobile chips keeping the same naming scheme as desktop parts as long as it is consistent. IE...

[generation][performance category][performance sub category][0 for original/5 for refresh or mobile part]

The only real change would be putting that last number to some use as for the most part it's always zero.

In the case of the 51xx cards, I'm incredibly disappointed as [5] no longer denotes the actual generation of the silicon used and to complicate things [1] might not denote the performance category of this silicon relative to other solutions in the [5] generation/category.

It's like a double whammy of disappointment. I realize there was pressure from OEM's for something along those lines. And with Nvidia also pandering to something along those lines, the pressure on AMD must have been rather large for something similar. Especially with the problems TSMC have had with 40 nm.

However, I just wish AMD would have had the balls to stick up for the consumers with no renaming.

Regards,
SB

 

[Squilliam]

However, I just wish AMD would have had the balls to stick up for the consumers with no renaming.

Regards,
SB

I guess you're right overall, theres a good reason why they don't use the system I proposed. In addition to this, it would have been nice to have a naming convention which simply works whether talking about desktop or mobile parts.

I wonder if instead of renaming all the time they ought to use generational names like they do with CPUs. Like for example Phenom or Turion instead of model names. That way they don't have to worry about one number being smaller than the other.

 

[Silent_Buddha]

I guess you're right overall, theres a good reason why they don't use the system I proposed. In addition to this, it would have been nice to have a naming convention which simply works whether talking about desktop or mobile parts.

I wonder if instead of renaming all the time they ought to use generational names like they do with CPUs. Like for example Phenom or Turion instead of model names. That way they don't have to worry about one number being smaller than the other.

It's easier for CPU's to have brand names be loosely based on "generation." as CPU life cycles are generally far longer than GPUs.

But even in that aspect, it doesn't quite equate. For example Core i7 contains 2 generations. Core i5 contains 2 generations. As an aside I'm quite disappointed with the Core iX naming schemes. Whether it's i7 or i5 seems a bit arbitrary.

Generational CPU chips also aren't as different between generations as GPUs, they're more like refreshes and then every 5+ years a new architechture is actually released with major "feature" changes.

But even with all that, when looking at CPU brand names, it only seems like they get changed when there starts to be a negative connotation to the name. For example, Pentium being dropped in favor of Core. Athlon being dropped in favor of Phenom.

Regards,
SB

 

[Alexko]

Ugh, that's just really really horrible. To say I'm disappointed would be a huge understatement.

Although I'm sure the same people that kept telling me that it was fine for Nvidia to do it will also tell me that it's perfectly awesome that ATI are doing it...

Horrible, absolutely horrible...

Regards,
SB

It's pretty bad and disappointing, but at least the new names make it very clear that these products are somehow inferior. The HD 5165 looks really crappy even though it's actually faster than the HD 5300... which, granted, adds to the confusion. But it's still better to rename the HD 4670 to HD 5165 than to HD 5670, as Nvidia tends to do lately.

Which, by the way, is probably the reason behind such a move: when your opponent fights dirty, you either turn the other cheek or sink to their level, and it appears ATI's cheeks got a little too red.

That said, I really hope this is just temporary and ATI will discontinue these rebrands as soon as Cedar and Redwood can be made in sufficient quantities.

 

[Mindfury]

HD5570 benchmark：
http://translate.google.com/transla...10/0107/834/000000834487.shtml&sl=zh-CN&tl=en

 

[neliz]

Edit: In addition to this they can drop the feature monikors once they become standard, so they don't need the E for example next generation.

That whole scheme makes SO much sense and is SO sane.. they'll never adopt it.

 

[mczak]

HD5570 benchmark：
http://translate.google.com/transla...10/0107/834/000000834487.shtml&sl=zh-CN&tl=en

Very unsurprisingly, beats the GT220 easily. I wish they'd compare it to a GT240 (ddr3) too, and also HD4670 for better comparison, looks like it might be close (to both).
Comparing to gt216 chip is not very exciting, while it's slightly bigger (110 vs 100 mm²), gt216 is pretty much inferior any way you look at it. But compared to gt215 things get more interesting. gt215 is quite a bit bigger (140 vs 110 mm²), same amount of rops and roughly same memory bandwidth (ok the gt216 has that too), but the gt215 has an advantage in texture fillrate (50%) whereas the rv830 has a large advantage (as usual..) in peak alu throughput (100% or so).

btw I'm actually a bit surprised AMD has gone for length 16 arrays with this chip. I speculated it might make sense to increase array length this generation, but I didn't expect AMD to go all the way from 8 to 16 - not with "only" 400 SPs at least (would have made more sense imho with 480 SPs which would have given 24 texture units, now texture filtering rate is a lot lower than HD4670).
The rv810 is a disappointment though. It's got more features than the predecessor but performance probably just the same, and all that on a slightly larger die (on a smaller process node)? It would have deserved a little more shader power, especially since it now also supports gddr5 (though I don't expect that configuration to be popular nor to make much sense given the pathetic shader clusters).

 

[Silent_Buddha]

If power consumption is down on Rv810, I'm still looking forward to it. Down in that category I'm not really looking for speed anyways. As is it might just be the perfect GPU for a low power silent HTPC.

While I'm not sure how much use I'd make out of bitstreaming, it's one of those things I want to try out.

Either way as long as it's passively cooled, with lower power consumption and more features than my 3450, then I'll be a happy camper.

Regards,
SB

 

[ECH]

Does anyone know what Trillian is?

I think it's safe to assume that AMD is preparing a refresh. We also know that there is suppose to be a 6 input (perhaps CF'd) video card call Trillian. All that's been mentioned is that this is a 6 input HDMI video card unlike Hemlock. However, something is missing. Is this part of AMD's refresh of higher performance video cards? If so, are we looking at more then one SKU? IE:
A. 5890
B. Trillian
C. ??

 

[neliz]

you Mean the HD5870 "Eyefinity6" with 2GB of RAM?
http://benchmarkreviews.com/images/..._radeo_5870_eyefinity6_edition_video_card.jpg

edit: yes Trillian is the "ATI Radeon HD5870 Eyefinity(6) Edition." Max Resolution: 7680x4800.

This isn't part of the magic refresh :/ card should launch somewhere in February, the refreshes weren't slated until March at it's earliest.