AMD: R9xx Speculation

[Rangers]

I don't know how much faster the 69x0 series will become. I just have to say that if you look back at benchmarks for oct 2009 when cypress hit , it has improved by leaps and bounds. I would think performance would be up at least 10-30% in every game. I remember a while back when widescreen gaming showed that with the newest driver of the time the 5850 was as powerful as the 5870 was using launch day drivers and hte 5830 was close to hte 5850 in performance with launch day drivers.

I would expect this new gpu to see equal gains across its life

If every GPU sees the same (supposed) gains though it doesn't really help anything in relative terms.

For example I consider my 4890 to be about 20% slower than the 5850. That's still the case if they both improve the same. Same with Nvidia product compared to ATI.

I really always see improvements in specific titles and specific cases. Cannot ever remember a site showing large across the board improvements between specific drivers, quite the opposite. There is a site that benches drivers against each other (new revision against last revision) and it's usually almost no difference in the vast majority of games. Another clue to that would be the driver release notes, which call out performance improvements, and only mention a few specific cases typically.

 

[Pete]

I wish sites would come back and do a retrospective review of old gpus....atm with gpu tech slowing....make even more sense...i no longer see 4870 1GB in so many Cayman reviews.

TechReport.com has included a 4870 or 4890 in their past few reviews. I know there's a site that regularly checked driver updates for performance improvements, though I can't remember it's name. IIRC Anandtech and one other site has tested driver improvements over the lifetime of an ATI and NV card, but from a few card generations ago.

 

[eastmen]

If every GPU sees the same (supposed) gains though it doesn't really help anything in relative terms.

For example I consider my 4890 to be about 20% slower than the 5850. That's still the case if they both improve the same. Same with Nvidia product compared to ATI.

I really always see improvements in specific titles and specific cases. Cannot ever remember a site showing large across the board improvements between specific drivers, quite the opposite. There is a site that benches drivers against each other (new revision against last revision) and it's usually almost no difference in the vast majority of games. Another clue to that would be the driver release notes, which call out performance improvements, and only mention a few specific cases typically.

Does it . The gains normaly come within the first year. When dealing with the 4890 you have to remeber that there was the 4850 and 4870 that shared the same tech. So by the time the 5850 came out the 4890 would have been done seeing its performance increases where as the 5850 would have continued improving for awhile after its release.

http://www.techreport.com/articles.x/19844/9

there is no 4890 tho.



As an aside i'm going to put my 5850 xfx for sale on amazon around christmas. I'm going to get the 6970. I figure this will be the card i play tor with and if its able to play Tor with eyefinity settings i will be very happy

 

[hatter]

TechReport.com has included a 4870 or 4890 in their past few reviews. I know there's a site that regularly checked driver updates for performance improvements, though I can't remember it's name. IIRC Anandtech and one other site has tested driver improvements over the lifetime of an ATI and NV card, but from a few card generations ago.

The website is Tweaktown... whenever there is a new driver, these guys check for performance gains. But only using one or two cards, I think.

 

[Jawed]

Somehow the raw tesselation numbers from hardware.fr http://www.hardware.fr/articles/813-7/dossier-amd-radeon-hd-6970-6950-seules-crossfire-x.html and the PCgameshardware unigineheaven2.1 results http://www.pcgameshardware.de/aid,8...klasse-Grafikkarten/Grafikkarte/Test/?page=12 doesnt look right to me as u increase tesselation setings.
It seems to me that the 460 near half fps of 580 trough the whole normal->extreme tesselation in heaven is actualy quite strange if u compare it with radeon line up. (and mainly 6900 cards which end same than 460 at extreme but have quite a lead in rest. And tesselation rate counts there clerly as the 5870(and even 6870) card is already slover a bit than 460 on normal seting)

I still don't understand how you justify "1/4"

 

[Karoshi]

TechReport.com has included a 4870 or 4890 in their past few reviews. I know there's a site that regularly checked driver updates for performance improvements, though I can't remember it's name. IIRC Anandtech and one other site has tested driver improvements over the lifetime of an ATI and NV card, but from a few card generations ago.

Check http://www.computerbase.de/suche/?bereich=&q=catalyst

For example:
http://www.computerbase.de/artikel/...catalyst-10.9/9/#abschnitt_colin_mcrae_dirt_2

But the last i could find is 10.9, oh well.

 

[gongo]

I would guess the opposite myself.

Aaand with those great words...i took a leap of faith and got me a Cayman Pro! Could have gotten XT but i thought over it that 28nm and PCIE3.0 is a year off, and Cayman wont face the shortages of 40nm...so doubt XT will retain the type of price/perf of Cypress. 2GB and familiar drivers made me pass over Fermi. I also hold a small hope that with dual bios, similar pcb/chips..someone will figure out to unlock AMD's GPU this time! Matured VLIW4 drivers on you Dave and AMD, please dont let us down! No..not expecting miracles, but continued support and communications with game developers to unleash Cayman powers.

First impression is...does CCC runs on no's of SP or how do you explain switching from HD4800 to HD6900 is like going from mech HDD to SSD....CCC is blazing fast with Cayman Pro!(i hope its not a support issue) I also like now GDDR5 no longer flickers (LTTP Cypress) when downclocked...and with autotune -20%, i can switch to low-power profiles seamlessly...no annoying Flash flickers too! I noticed each profiles save your settings individually unlike HD4800 universal settings apart from Overdrive..a little getting used to.

Build quality is actually good, the fan is silent, ramps up smoothly, and way quieter than HD4800 and i like the back plate! It gets hot though, very hot, so reviewers who said it does not transfer heat is incorrect....but it made me question AMD decision to stick with half slot cooling(boo! Eyefinity) because as i said, the backplate gets hotter than HD4800 especially near the vent ends..and i cannot feel as much exhaust as full slot cooling.

Overclocking is not as bad as made out, i drag Overdrive sliders to the max of 840/1325 and completed 3DM11 for scores of around P5000....ran FF14 benchmark at high for an hour, no lock ups, no "drivers has stopped responding", no image corruption. At auto-fan, temps rose to about 85...(still unsure about half slot cooling)...Caymans should be able to run 900-950 with stock volts, question is when will AMD allow a higher limit with Overdrive? I counting on soon(after launch reviews/sales)...and not until eternity(which would be LAME).

Games i have only tested Dawn of War 2..and the minimal fps when from 24 to 35fps, average when from 45fps to 75fps, and that is with added AA...speaking of AA..why is the default AAMode : Performance/MSAA...the image of the fence is terrible...had to switch to Adaptive...much better...but no idea how much fps i will lose...

Summing up, i think going from 4870 1GB to 6950 2GB is well worth it, Cayman presents me a feeling that its engineering quality(hw & sw) is a step up from RV770. I disagree with AMD marketing Barts as RV770 replacements and Cayman as GTX200 replacements...that is aiming low, i know i wont be satisfied paying for Barts if i had RV770. AMD needs to aim higher next round...could have avoided the lingering negativity surrounding Cayman...

I need to do more game tests...but i am not expecting too much, just a 50% improvement all around....until 28nm/PCIE3.0 then!

 

[gongo]

Dave, AMD, anyone...why does my CCC shows Catalyst Ver 10.11 and Packing Version 8.79? I am using the 10.12a hotfix as per linked..?

 

[Arnold Beckenbauer]

Dave, AMD, anyone...why does my CCC shows Catalyst Ver 10.11 and Packing Version 8.79? I am using the 10.12a hotfix as per linked..?

That's why:

[ati2mtag_SoftwareDeviceSettings]
HKR,, DDC2Disabled, %REG_DWORD%, 0
HKR,, DisableBlockWrite, %REG_DWORD%, 1
HKR,, DisableDMACopy, %REG_DWORD%, 0
....
HKR,, Catalyst_Version,,"10.11"

This driver is 8.79 as 10.11, but it's the newest driver (cat. 10.12 is older).

 

[CarstenS]

BTW - the driver has been updated and should bring improvements in a few of the DX SDK tests. Download it again and give it a go.

Up to 75% improvement for certain tessellation factors - very nice!
http://www.pcgameshardware.de/aid,8...klasse-Grafikkarten/Grafikkarte/Test/?page=12 (scroll all the way down)
Now the SubD11 results are very much inline with what you'd expect when referencing Unigine Heaven 2.1 for example.

Judging from the file name, the driver on the press FTP is the same as the 10.12a Hotfix from Dec. 16th:
http://support.amd.com/us/kbarticles/Pages/AMDCatalyst1012ahotfix.aspx

 

[fellix]

Meh... rather moderate.

Looks like an attempt to optimize the access pattern to the DS (off-chip) buffer in the global memory.
I guess this could be profiled per application base, as each one potentially generates an unique I/O behavior, depending on tessellation impl.

 

[CarstenS]

That would mean a lot of work for Terry and his successor in the months to come.

 

[leoneazzurro]

Up to 75% improvement for certain tessellation factors - very nice!
http://www.pcgameshardware.de/aid,8...klasse-Grafikkarten/Grafikkarte/Test/?page=12 (scroll all the way down)
Now the SubD11 results are very much inline with what you'd expect when referencing Unigine Heaven 2.1 for example.

Judging from the file name, the driver on the press FTP is the same as the 10.12a Hotfix from Dec. 16th:
http://support.amd.com/us/kbarticles/Pages/AMDCatalyst1012ahotfix.aspx

Did you test if with the new drivers there are some improvements also in other applications, or there are none at this point?

 

[CarstenS]

Since I have been working overtime even without additional drivers, I could only re-test SubD11 and give TessMark a go. No other tests as of yet, I'm sorry.

 

[Dave Baumann]

This driver is 8.79 as 10.11, but it's the newest driver (cat. 10.12 is older).

I've not actually downloaded the drivers myself, yet. 8.79.6.2 is what you want.

Judging from the file name, the driver on the press FTP is the same as the 10.12a Hotfix from Dec. 16th:
http://support.amd.com/us/kbarticles/Pages/AMDCatalyst1012ahotfix.aspx

Yes, Andrew tells me the Cayman Hotfix was refreshed.

 

[fellix]

I'm curious how the tessellation performance (probably with higher factors) would scale with the memory clock?!

 

[swaaye]

There have been cards that have slowed down considerably with newer drivers. GeForce FX comes to mind. When they removed some of the horrific hacks of earlier drivers and/or stopped writing shader replacements for new games. Oblivion isn't really playable at 640x480 on a 5900 Ultra whereas a 9700 Pro is not too bad at 1280x800.

I remember seeing Radeon 8500 compared over years of driver releases and it didn't change a lot. They did fix lots of bugs though. That board was a total wreck for the first year.

The ATI cards with the ringbus design were supposedly in need of a lot of tweaking so looking for change over time with X1800 and X1900 would probably show something. R600 had an even more tweakable ring bus. I read this in one of those Eric Demers interviews.

I wouldn't mind seeing G80 compared from original drivers up to today's 260.xx pack. That would be interesting I think. Probably because I'm still using one.

 

[Arty]

6970 CF outperforming GTX580 SLI here. They gave a terrible rating to the 6970/6950 but its the opposite for the CrossFireX review. Good job on ATI on this front I guess.

 

[Erinyes]

My initial reaction was very "Meh" on reading the reviews. Sure there are some games where 6970 performs 30-40% better than a 5870 but for the most part it seems to be just 10-15% better(or as the general consensus is, pretty much on par with GTX 570). This makes it ~20-25% faster than a 6870. The die size is 50% bigger though, so its not very efficient to say the least.

I mean the VLIW4 arch is supposed to be 10% more area efficient. Given that number of shaders is less than Cypress, ROP's are the same, and TMU's have increased from 80 to 96, what the hell did they need an additional 60mm2 of die space for?

Im sure the performance will increases as the driver team is able to learn and optimise for the new architecture. Heck im willing to bet that we will see an incease of ~10% by the time Antilles launches(of course that now means that driver optimisation for VLIW5 will be a much lower priority now as VLIW4 is the basis for future products). But Cayman launch performance was definitely underwhelming. They could have made it look better by a little better pricing. The 6970 should have been $349, and the 6950 should have launched with a 1 GB variant at $279 as well as the 2GB variant. That would have made a much stronger case for Cayman.

Also nobody has mentioned the reason that the launch was delayed by three weeks. I know we had tons of speculation that it was clocks/bios/drivers,etc etc but no info on the actual reason for the delay

 

[swaaye]

When I read that Cayman was a new architecture designed by Demers' team, I couldn't help but wonder if it would end up like R600, with a much more efficient version coming later. I'm sure that there's zero chance of it being proved as inefficient as R600 was though. R600 was just crazy nuts.

How many teams are there working in parallel over there? Could Cayman have benefited from anything learned from Cypress or is that too recent?