AMD: R9xx Speculation

All the HD5-series chips already have HDMI 1.4a, have had since Catalyst 10.10, it was a mere software update needed for that.

However, the 3D stuff is limited on HD5 so they can't hardware decode the MVC-streams (or whatever they were called) and they have to be run in fullscreen mode - this is now fixed in HD6700 by driving the UVD2 in different manner, just like on the HD6-branded Evergreen mobile parts did.

Now all we need to know if one could flash HD5770 with 6770 bios to get that fixed on HD5770 too.
Yes right I confused the HDMI 1.4a issue (which can be done by driver) with the UVD2-doing-MVC issue (which for some reason requires bios update).
 
20116304412119473263044.jpg


20114770212119473647702.jpg


http://it.oc.com.tw/ocitview.asp?gitid=1554722
 
Or could it be just a case of old GPU-Z reading things wrong, and it being 384 SP instead (4D instead of 4+1D)?

edit: the thread title (4D) suggests it would be 4D too, and 480 SP just doesn't work with 4D really (7.5 SIMDs with 16 VLIW 4D's per SIMD or 15 SIMDs with 8 VLIW 4Ds per SIMD)
 
Last edited by a moderator:
Could explain the performance. I think it's slower than expected in the 06 score compared to redwood and juniper.
 
Or could it be just a case of old GPU-Z reading things wrong, and it being 384 SP instead (4D instead of 4+1D)?

edit: the thread title (4D) suggests it would be 4D too, and 480 SP just doesn't work with 4D really (7.5 SIMDs with 16 VLIW 4D's per SIMD or 15 SIMDs with 8 VLIW 4Ds per SIMD)

Isn't Cayman supposed to be the only 4D chip?
 
Isn't Cayman supposed to be the only 4D chip?

That was only speculation as far as I know, it's clear anyway that Barts "doesn't fit the family" properly anyway as it was planned to be 40nm from the beginning according to Dave Baumann
 
I always thought Turks is going to be marketed as the discrete GPU to crossfire with a LLano APU... So, to run as smoothly as possible one would try to get the GPUs to be as similar as possible, right? And Llano is supposed to have 5D shaders... or not? :D
 
480 SPs at 800MHz? Turks looks like a troublesome little fellow for NVIDIA in this segment.

yes, nvidia can only compete with a cut-down GF116 so here's a fun case of small die strategy versus big die strategy :)
Turks is how much, about half the die size?
 
yes, nvidia can only compete with a cut-down GF116 so here's a fun case of small die strategy versus big die strategy :)
Turks is how much, about half the die size?

I think GF106/116 is around 230mm², which makes it a lot closer to Barts (255mm²). I don't know how big Turks is, but Redwood is ~100mm², if I recall correctly. Turks probably isn't much larger.
 
Seems like Turks is 16 ROPs (and expected 480 SP)
http://we.pcinlife.com/thread-1598136-1-1.html
I don't think so. The 3dmark06 score is "only" about 1000 points higher than HD 5670 (and of course way below HD 5750), which imho fits quite well with the very slightly higher clock and 1 more simd. The 16 ROPs are probably a (bad) guess from gpu-z. With that clock I assume it still comes without pice power connector though.

I think GF106/116 is around 230mm², which makes it a lot closer to Barts (255mm²). I don't know how big Turks is, but Redwood is ~100mm², if I recall correctly. Turks probably isn't much larger.
Redwood is 104 mm². Turks has 14% more transistors, if we assume similar transistor density that would make it ~120mm² - in any case still below GF108 (which couldn't compete with HD5670), and indeed only half the size of GF106 (which of course is faster - disable one of the four SM and it might get close).
 
I always thought Turks is going to be marketed as the discrete GPU to crossfire with a LLano APU... So, to run as smoothly as possible one would try to get the GPUs to be as similar as possible, right? And Llano is supposed to have 5D shaders... or not? :D
5D shaders are confirmed as far as I can tell for Turks (same as Whistler).
So it looks like indeed for now Cayman is the only one with 4D shaders - but future chips surely will adopt that (imho).
 
I wished AMD kept the product stack naming....as of what we have seen, marketing is against them...the swift price drops of Cayman Pro felt a little negative..just read Anand review

http://www.anandtech.com/show/4137/...hd-6950-1gb-xfxs-radeon-hd-6870-black-edition

Sticking with HD6700/6800 and slight downward adjustment of BartsXT and CaymanPro clocks... would give the impression Barts/Caymans line are a step up from the replaced HD5700/5800...like Fermi 2.0...and the window to lavish praise on overclocking..though not as high as Fermi 2.0...its still better than reading words as these from TPU

The overclocks of our card are 860 MHz core (8% overclock) and 1540 MHz Memory (23% overclock). While the core overclock is a bit disappointing, memory seems to be able to reach extremely high levels. Based on the 860 MHz number, my guess is that the GPU chip on our sample did not make the clock qualification for the HD 6970 (880 MHz), so it ended up on a HD 6950.

I think AMD went full guns blazing for HD6000...with the stagnated fab process and it hurts them in their lineup (renamed Juniper..gargh)..... Cant say I'm unhappy with ma CaymanProXT. I love ma unlocking! But as of now i say...the view towards AMD performance GPU is not good...and there too exits a gap between Cayman XT < GTX 580 < Antilles, unable to exploit in sales or marketing .....6980 seems pushing it..

I think a 850/1000 BartsXT HD5790 would compete with GF114 salvage parts (SRP $229~199)
I think a 780/1200 CaymanPro HD6850 1GB would compete with GF114 ($259)..shorter PCB included...
I think a 750/1200 AntillesPro HD6950 (dual HD6850) 4GB @ $529 would sit nicely for power users! ($519)
I think a 2GB HD6850 820/1250.... would come at a later date ....if IHV starts to threaten with 950~1000mhz OC GF114....

Both would now seem like a marked improvement over HD5000 series for series.
Both would get more say in OC opportunity in IHV boards and reviews. The value proposition just like the short termed unlocking craze..

Hey..i still love ma CaymanProXT. Thanks for going all out AMD. :)
 
Last edited by a moderator:
Oh I don't know. I think AMD has every segment pretty much covered. They have several variant products for each GPU. Lots of options at many price points. Splitting hairs over $30 intervals is a waste of time. They may have more choices today than ever before.

You are also making calls based on knowledge AMD didn't have about NVIDIA's products when they released various cards. If AMD was omniscient I'm sure NVIDIA (and Intel) would be dead, and AMD would probably have taken over the world by now. :)
 
Back
Top