Radeon HD 7000 Revealed: AMD to Mix GCN with VLIW4 & VLIW5 Architectures

Quote:

---

Originally Posted by Lightman

I suggest you do that! Plus get new GPUClock Tool from AMD and check how well/bad your sample is clocking! New 28nm process so a lot to be learnt again :D

---

Interesting :D


Is 7970 capable of around 14k.P by gpu in 3dM11?

Quote:

---

Originally Posted by LordEC911

Crossfire config is two 7970 ES.
Retail card should be ~210w TDP, 6+1 phase and 8+6pin.

VR-Zone's is a MSI 7950.
384bit and will come in both 1.5gb and 3gb flavors.
TDP is supposedly around 170-180w, 5+1phase and two 6pins.

---

is it me or does that look really good? wasn't 6970 like 250w TDP?

If that is true then it is indeed not bad and it means dual Tahiti XT may be able to draw just about 300W (I believe 210W is what drew the 5870 correct?), a lot better than 6990.

Those TDP numbers could be due to PowerTune's artificial limit but they should be pretty close.

Quote:

---

Originally Posted by Lightman

I suggest you do that! Plus get new GPUClock Tool from AMD and check how well/bad your sample is clocking! New 28nm process so a lot to be learnt again :D

---

There's a new one? where?

the 7950 looks like a sweet card...unlocked to 7970 muahaha lol

Quote:

---

Originally Posted by LordEC911

Crossfire config is two 7970 ES.
Retail card should be ~210w TDP, 6+1 phase and 8+6pin.

VR-Zone's is a MSI 7950.
384bit and will come in both 1.5gb and 3gb flavors.
TDP is supposedly around 170-180w, 5+1phase and two 6pins.

---

Quote:

---

Originally Posted by tbone8ty

is it me or does that look really good? wasn't 6970 like 250w TDP?

---

It does look really good!

I think my GTX470 is rated at TDP of 215W, and if the 7970 is around 210 it would amazing card and replacement.

Specially for people like myself with M-ITX systems that still want some pure grunt.

fellix estimates the chip size around 432,6 mm²

http://tof.canardpc.com/preview2/24d...1434028735.jpg

http://forum.beyond3d.com/showpost.p...postcount=1202

It's going to be very interesting to see how AMD fares with a big chip and a new architecture once again. At least it's not late, and that's a very good sign!

Quote:

---

Originally Posted by AliG

Yeah I agree, ATI might have a better hardware model at the moment, but I'm fed up with the instability of their drivers. If the Nvidia product is at a similar performance level and runs at reasonable temps/power consumption I'm going to go with them next round simply because I trust their software guys a lot more

---

Same here, I've preferred the nVidia drivers for awhile honestly. I'm most excited about HD7000 series just to see it push nVidia's next-gen closer :), though the hardware looks great. My SLI setups have worked essentially flawlessly for me since the GTX 2xx series, and nearly flawlessly in the 8800GT era when I had two in SLI for awhile. I'm happy enough with my performance at the moment with GTX 570 SLI to hold out for the 6xx (or whatever it ends up being called) series. I'll see how things pan out from both sides with 28nm.

Quote:

---

Originally Posted by Olivon

fellix estimates the chip size around 432,6 mm²

http://tof.canardpc.com/preview2/24d...1434028735.jpg

http://forum.beyond3d.com/showpost.p...postcount=1202

---

wow that's a hell of a big chip! Kudos to getting this thing out on schedule with all the layoffs going on, if anything that's at least re-assuring about its performance

Quote:

---

Originally Posted by Olivon

fellix estimates the chip size around 432,6 mm²

---

Still not much compared to GF110's die size of 520 mm².
I really hope these cards are going to bring significant performance improvements for gaming, and not just compute and tessellation. Fermi was a big gaming performance leap for Nvidia; however, I don't see this architecture bringing much more for games than some extra processing units (which don't scale that well past Cayman's numbers in the first place).

Quote:

---

Originally Posted by zalbard

Still not much compared to GF110's die size of 520 mm².
I really hope these cards are going to bring significant performance improvements for gaming, and not just compute and tessellation. Fermi was a big gaming performance leap for Nvidia; however, I don't see this architecture bringing much more for games than some extra processing units (which don't scale that well past Cayman's numbers in the first place).

---

I can be wrong, but i have the feeling the CU arrangement ( 30-32CU made of 4SMID, made of 16 ALU (Vector ) + 1Scalar units ), can bring more efficiency for the processors units. So it's a bit hard to compare with Cayman.. Well we will know it fast enough.

Quote:

---

Originally Posted by zalbard

Still not much compared to GF110's die size of 520 mm².

---

i wonder how that would compare if gf110 was 28nm as well

My rudimentary methods gives me the approximate same die size as Cayman. I'm taking the ratio between the top and bottom corners of the die and the top and bottom line of the square lines on the PCB. The ones that encloses the 4 screw holes. This is of course assuming that the square on both cards are the same which they should be. Cayman's ratio is around 0.485 while Tahiti is around 0.466.

Quote:

---

Originally Posted by zalbard

Still not much compared to GF110's die size of 520 mm².
I really hope these cards are going to bring significant performance improvements for gaming, and not just compute and tessellation. Fermi was a big gaming performance leap for Nvidia; however, I don't see this architecture bringing much more for games than some extra processing units (which don't scale that well past Cayman's numbers in the first place).

---

This is significantly bigger comparatively since Fermi was on 40nm.

Quote:

---

Originally Posted by Cold Fussion

This is significantly bigger comparatively since Fermi was on 40nm.

---

Huum, well if 28nm process have exist when the GTX480 and Cypress/cayman have been released, im not sure the die size will have change so much ( maybe a little less for both brands ) . They will have been packed with more SP at this time.

I doubt you hope see a Fermi2 ( GTX6xx ) who come with 480/512SP ( 32SP was disabled, but the die size between a GTX480/512 was the same )in 28nm this year, just for reduce the die size. But at least with more SP...

Amd just announces the Caymen die is actually 697,000 transistors to many. Lol

Quote:

---

DCenter compiled specifications of "Tahiti", based on bits and pieces of information from various sources. The specs can be listed out as:
http://translate.google.com/translat...s%2F2011-12-11


4.50 billion transistors, die-area of 380 mm², built on TSMC 28 nm process
Advanced GCN 1D architecture
2048 1D processing cores
128 TMUs, 48 ROPs
384-bit wide GDDR5 memory interface, memory clock slightly below 1 GHz, target bandwidth of 240~264 GB/s

---

this has been posted before but jsut wanted to update this thread.

this table is close to what has been rumored minus the XDR2 memory looks to be (384-bit gddr5) and i think ROPs, clusters, and TMUs r off a little 2 but im not sure on the math....streaming processors look to be close
7900
Attachment 122865

7800
Attachment 122866

(tables are from NordicHardware)

thinking about compiling my own table in excel...

http://farm8.staticflickr.com/7163/6...9240e365_b.jpg
HD7000 series rumors table by Tyler Morrison, on Flickr
updated 12-14-2011 @2:10pm

this is a table i compiled from various sources going around the web...if u see a mistake or what i should edit please let me know and ill update it.

i apologize in advance lol

Quote:

---

Originally Posted by tbone8ty

if u see a mistake or what i should edit please let me know and ill update it.

---

384bit = 48ROPs.
7950 will be 384bit so also 48ROPs.

78x0 is Pitcairn not Thames... also Pitcairn is GCN.
7870 should have a TDP ~150w.
7850 should be ~120-130w and probably just a single 6pin.

Memory clock for 186Gbps on 256bit is 5.8ghz. 166Gbps is 5.2ghz.

Transistor count should be a bit higher on Pitcairn than Cayman, so maybe 3.1-3.2b?
Should also end up slightly bigger than Barts, 270-280mm2.

ok thanks i updated the table...i thought 7870 and 7850 were gonna be VLIW4?

prices ok u think?

if there doing nothing by GCN, it better be good and ready

Quote:

---

Originally Posted by tbone8ty

ok thanks i updated the table...i thought 7870 and 7850 were gonna be VLIW4?

prices ok u think?

---

Thames is on 40nm and probably vliw5, a rebranded HD6x00 (can't remember what series off the top of my head).
There is more than one ASIC using GCN.

Prices should be in the neighborhood according to Charlie.

I'm still not 100% sure they will be disabling CUs in less than quads.

http://www.abload.de/img/7970piraf.png

Quote:

---

Originally Posted by OBR

ATTENTION: It seems this slide is legit but little bit outdated, GPU clocks and Load TDP are lower! GPU clocks are under 900 MHz and Load TDP +/- 200W.

Gaming performance is far behind expectations, but i told you so before ... enjoy this first REAL leak about Radeons HD 7900 ... by OBR

PS. Thanks Bruno Mur... for slides! Good boy! I will wait for Donanim or any web who leaks something more first, and will leak next slides. Zero Power Feature, Ocing, Tesselation performance, Gaming performance etc. Stay Tuned!

---

http://www.obr-hardware.com/

Quote:

---

Originally Posted by kaktus1907

http://www.abload.de/img/7970piraf.png



http://www.obr-hardware.com/

---

Quote:

---

Originally Posted by Vapor

Update, July 27, 2011:

Something has been happening with increasing frequency that needs to stop: linking or referring to the work of banned members (whether that's to their posts on another forum or to a site/blog they now run). This is a forum-wide rule but is of particular concern in the News Section where first infraction = temporary forum ban and permanent News Section ban.

Some banned members leak early hardware benchmarks or roadmaps that may seem important, but when somebody has been removed from XS, they no longer have a right to have a voice on XS.

We realize not all banned members are known to all current members so this may not always result in an infraction, it depends on the situation and recent in-thread warnings.

---

http://www.xtremesystems.org/forums/...=1#post4914806

32 ROPS, yeah, I don't think so.

i just can't take this guy seriously; there isn't any other site out there spreading more hate against single hardware manufacturers than his blog; even apple haters don't go that far :rolleyes:

Quote:

---

Originally Posted by generics_user

i just can't take this guy seriously; there isn't any other site out there spreading more hate against single hardware manufacturers than his blog; even apple haters don't go that far :rolleyes:

---

this

what did AMD do to him crap in his oatmeal
charlie with nv is not this bad

hopefully they will have nice cooling solution under that hood

http://videocardz.com/29546/amd-rade...hamber-cooling

Yargh OBR.

-PB

I am not saying whether the OBR post is right or absolutely wrong but what about it strikes you as being anti-AMD?

what would higher or lower ROPs do? ive seen 64, 48, now 32 floating around.

hypothetically speaking say the SP are 2048 and texture units are 128. can u determine the ROPS from that?

this table seems to be more like the 7970 and 7950 specs mixed together

Quote:

---

Originally Posted by tbone8ty

what would higher or lower ROPs do? ive seen 64, 48, now 32 floating around.

hypothetically speaking say the SP are 2048 and texture units are 128. can u determine the ROPS from that?

this table seems to be more like the 7970 and 7950 specs mixed together

---

Rops are tied to the memory controller, in blocks of 4-8 rops per memory bus channel (usually 8 these days). 32 rops would be either a 256bit bus or a 512bit bus. For 384-bit it'd either be 24 or 48 rops. You can't divide 32 rops evenly to make said 384-bit bus, and as such it's probably fake just based off of that alone.

Quote:

---

Originally Posted by tbone8ty

what would higher or lower ROPs do? ive seen 64, 48, now 32 floating around.

hypothetically speaking say the SP are 2048 and texture units are 128. can u determine the ROPS from that?

this table seems to be more like the 7970 and 7950 specs mixed together

---

The TMU are easy to calculate, there's 4 TMU/CU ..., 32x4. Untill they want double the number and end with 256 ( who will be innefficient anyway )

For the ROPS, well Diltech have said all.

I add just a little quote from Anandtech GCN article:

Quote:

---

After a trip through the CUs, graphics work then hits the pixel pipelines, which are home to the ROPs. As it’s customary to have a number of ROPs, there will be a scalable number of pixel pipelines in GCN; we expect this will be closely coupled with the number of memory controllers to maintain the tight ROP/L2/Memory integration that’s so critical for high ROP performance.

---

and more who have absolutely nothing to do with ROP

Quote:

---

On the other side of the coin we have the graphics hardware. As with Cayman a graphics command processor sits at the top of the stack and is responsible for farming out work to the various components of the graphics subsystem. Below that Cayman’s dual graphics engines have been replaced with multiple primitive pipelines, which will serve the same general purpose of geometry and fixed-function processing. Primative pipelines will be responsible for tessellation, geometry, and high-order surface processing among other things. Whereas Cayman was limited to 2 such units, GCN will be fully scalable, so AMD will be able to handle incredibly large amounts of geometry if necessary.

---

We have all see thoses pictures allready, but this can help for see exactly how it work
http://img35.imageshack.us/img35/1520/amdgcn3th.png

http://img267.imageshack.us/img267/6043/gcncuth.png

Nice to see AMD are taking my primary complaint about their last architecture seriously and giving this one a big push in it's ability to perform tessellation. In a lot of ways Cayman and Cypress were DX10.1 cards with DX11 kind of tacked on competing against cards made for DX11 from the ground up.

It'll be interesting to see AMD with their true next generation architecture, since it'll be their first big change since the HD2900. Going to VLIW4 from VLIW5 was a step in the right direction--but, it wasn't nearly as drastic of a change as a lot of people here made it out to be.

Quote:

---

Originally Posted by DilTech

Rops are tied to the memory controller, in blocks of 4-8 rops per memory bus channel (usually 8 these days). 32 rops would be either a 256bit bus or a 512bit bus. For 384-bit it'd either be 24 or 48 rops. You can't divide 32 rops evenly to make said 384-bit bus, and as such it's probably fake just based off of that alone.

---

Remember, AMD never "scaled" their cache, ROP and memory hierarchy like NVIDIA did. With NVIDIA all three were tied together due to the nature of their core architecture but AMD could always scale them in an asymmetric fashion.

For example, the Barts LE core used on the HD 6790 1GB disabled two of the Render Back Ends of the Barts core to produce a card with a 256-bit memory bus but only 16 ROPs (rather than 32). As such, PAST AMD cores weren't tied down in the same way as NVIDIA's Fermi architecture.

On a related note, there is a way to have uneven/mismatched numbers of rops/etc as well, but I dont think we'll be seeing it this round of GPU's. To put it simply: kal-el.

Quote:

---

Originally Posted by SKYMTL

Remember, AMD never "scaled" their cache, ROP and memory hierarchy like NVIDIA did. With NVIDIA all three were tied together due to the nature of their core architecture but AMD could always scale them in an asymmetric fashion.

For example, the Barts LE core used on the HD 6790 1GB disabled two of the Render Back Ends of the Barts core to produce a card with a 256-bit memory bus but only 16 ROPs (rather than 32). As such, PAST AMD cores weren't tied down in the same way as NVIDIA's Fermi architecture.

---

I agree, but let#s not forget we dont speak about a low end card who was a cutted Barts core. But well, its not really important anyway.

Quote:

---

Originally Posted by STEvil

On a related note, there is a way to have uneven/mismatched numbers of rops/etc as well, but I dont think we'll be seeing it this round of GPU's. To put it simply: kal-el.

---

Are they using similar memory controllers?

Quote:

---

Originally Posted by SKYMTL

Remember, AMD never "scaled" their cache, ROP and memory hierarchy like NVIDIA did. With NVIDIA all three were tied together due to the nature of their core architecture but AMD could always scale them in an asymmetric fashion.

For example, the Barts LE core used on the HD 6790 1GB disabled two of the Render Back Ends of the Barts core to produce a card with a 256-bit memory bus but only 16 ROPs (rather than 32). As such, PAST AMD cores weren't tied down in the same way as NVIDIA's Fermi architecture.

---

That's going with the 4 split rather than the 8 split, still matches the mathematics we've always seen. They didn't disable RBE's, they disabled half of the Rops per RBE.

I seriously can't even remember the last time we saw a gpu that wasn't symmetrical rop to memory controller. In fact, I'm now curious... When IS the last time AMD/ATi released a chip that wasn't? (serious question) AFAIK both brands ARE actually tied together in this aspect presently.

Everyone keeps saying 32CUs and 2048 ALUs.... but that's not counting the scalar units. So why not advertise as an 2080 ALU part? Or are the scalar ALUs only there for GPGPU?

We want benches already.

Quote:

---

Originally Posted by hurleybird

Everyone keeps saying 32CUs and 2048 ALUs.... but that's not counting the scalar units. So why not advertise as an 2080 ALU part? Or are the scalar ALUs only there for GPGPU?

---

I think it's a little bit complicate to include it.. basically, the Scalar unit is there for take some operation for dont charge the SMID with them, (" specifically, simple integer operations to control flow operations like conditional branches (if/else) and jumps, and in certain cases read-only memory operations from a dedicated scalar L1 cache. Overall the scalar unit can execute one instruction per cycle, which means it can complete 4 instructions over the period of time it takes for one wavefront to be completed on a SIMD. (copied from anandtech article) "on/off mathematicals op" .

Let say it is not responsible of the same part of the vertex Alu, or SP ... Dunno if they count it or not. It's a little bit like the tesselation engine, hard to count them in the SP. GPU become more and more complex, and i think the SP numbers is just purely indicative for get an idea of the spec now.

Will the 7000 series be dx 11.1?

Quote:

---

Originally Posted by Pokerface

Will the 7000 series be dx 11.1?

---

it should as its a minor update to accommodate the win8 driver model and lower quality color support (the end of the 24bit minimum), i would assume that all the amd and NV dx11 cards will work with dx11.1 with driver updates.

Thank you zanzabar.

AMD: We Are Shipping Our 28nm GPUs Now - XBitLabs

Quote:

---

"We are ramping 28nm [products] with TSMC in Taiwan and shipping the products here and now. We are very excited about the products," said Rory Read, chief executive officer of AMD, during IT Supply Chain conference organized by Raymond James.

---

I just read that the idle power is advertised at 3W for the HD 7970. Very nice especially for those running multiple cards.

Fuad talks about 365mm² regarding chip size :

Quote:

---

The chip is 365 square millimeters in size and performance wise it should end up faster than Geforce GTX 580.

---

http://www.fudzilla.com/graphics/ite...ent-in-january

"It should end faster of the 580" . I dont understand why he base his article on that. new architecture, 28nm, new series ...

365mm2 definitely leaves them to release an update later in 28nm

huum, Cypress was 334mm2... but 365mm2, is a bit too much far of the estimation made..

slide says "embago" lol

Do we have any knowledge of how the GCN shaders will compared to those from the R600 architecture? IE how do 2048 GCN cores compare to 2048 VLIW4 cores (suped up 6970 effectively)

Quote:

---

Originally Posted by AliG

Do we have any knowledge of how the GCN shaders will compared to those from the R600 architecture? IE how do 2048 GCN cores compare to 2048 VLIW4 cores (suped up 6970 effectively)

---

GCN will have higher efficiency and run-time utilization than VLIW. It will have some sort of limited dynamic scheduling, but nothing exotic like in Fermi. GCN will rely on the sheer amount of in-flight threads/wavefronts to choose from and hide memory access latency and other pipeline stalls, and much less on the compiler.
The VLIW architrecture required some quite laborious compiler work to group instructions into so called "clauses" and offered zero hardware support for any run-time optimizations (instruction scheduling/reordering).

Quote:

---

Originally Posted by cx-ray

I just read that the idle power is advertised at 3W for the HD 7970. Very nice especially for those running multiple cards.

---

That isn't idle board power.

7900 Series The World's Most Powerful and Advanced Graphics

http://www.abload.de/thumb/amdradeon...pb_8_vjrxu.jpg http://www.abload.de/thumb/amdradeon...pb_1_8qogt.jpg http://www.abload.de/thumb/amdradeon...pb_2_slpnh.jpg http://www.abload.de/thumb/amdradeon...pb_7_oorme.jpg http://www.abload.de/thumb/amdradeon...pb_3_eaptk.jpg http://www.abload.de/thumb/amdradeon...pb_5_kop8z.jpg http://www.abload.de/thumb/amdradeon...pb_6_0prx3.jpg http://www.abload.de/thumb/amdradeon...pb_4_rnrhk.jpg http://www.abload.de/thumb/slide30fnp54.jpg

Hah!

Totally faked OBR slides... the one just above is real.
OBR's... wrong font, old visual style and faked specbars (intentionally prolonged to give the impression that AMD were giving super unrealistic numbers)
What a c*nt.

hmm 1 week ?
http://vr-zone.com/articles/amd-brin...011/14255.html

Looks interesting I guess. TBH I'm more intrigued at the rumors of GK100 having 1024 CUDA Cores; GCN would have to be at least 70% more efficient for AMD to keep up in single gpu setups from my math (I assumed linear performance scaling, and gtx 580 being 10% faster than 6970). I guess that means they always have the 7990 to go against NVidia's high end, but power consumption would be astronomical as would price

I will wait to see Nvidia vs ATI before making any decisions. There are so many architecture changes happening with this new generation of videocards (from both companies) that I just want to see why those changes are being made. Overall I'm just excited for the 28nm revolution.

Agreed on the 28nm. I've held off on a gpu upgrade for a long time now purely because I didn't want a 40nm product. With that being said, doing the math you can see that GCN would have to be drastically more efficient than VLIW to compete with GK100. From everything we've heard, it seems that the majority of changes were done with the workstation market in mind (impyling that I don't think we'll see this round go to AMD).

Quote:

---

Originally Posted by kaktus1907

7900 Series The World's Most Powerful and Advanced Graphics

http://www.abload.de/thumb/amdradeon...pb_8_vjrxu.jpg http://www.abload.de/thumb/amdradeon...pb_1_8qogt.jpg http://www.abload.de/thumb/amdradeon...pb_2_slpnh.jpg http://www.abload.de/thumb/amdradeon...pb_7_oorme.jpg http://www.abload.de/thumb/amdradeon...pb_3_eaptk.jpg http://www.abload.de/thumb/amdradeon...pb_5_kop8z.jpg http://www.abload.de/thumb/amdradeon...pb_6_0prx3.jpg http://www.abload.de/thumb/amdradeon...pb_4_rnrhk.jpg http://www.abload.de/thumb/slide30fnp54.jpg

---

I like big bars and I cannot lie ...

I'm interested. Probably will build a new sys soon :)

Quote:

---

Originally Posted by AliG

Agreed on the 28nm. I've held off on a gpu upgrade for a long time now purely because I didn't want a 40nm product. With that being said, doing the math you can see that GCN would have to be drastically more efficient than VLIW to compete with GK100. From everything we've heard, it seems that the majority of changes were done with the workstation market in mind (impyling that I don't think we'll see this round go to AMD).

---

IF GK100 will have 1024 CUDA cores. It's all just rumors and we don't know how fast these will be yet.

can someone measure the length of the 7970 for us based on the image showing the side of the card and PCIe slot, thanks!

Still using a vapour chamber? I thought they were switching to those liquid chambers, or is 6th generation vapour chamber the same thing?

These slides look a bit more legit that the OBR ones, but still I wouldn't look too deep into performance numbers. Nice to see DX11.1 and PCI-e 3.0 supported though. :)

I sure hope those slides are not another round of spin.

I semi lost faith in AMD after all that has gone on with BD.

-PB

Quote:

---

Originally Posted by FischOderAal

IF GK100 will have 1024 CUDA cores. It's all just rumors and we don't know how fast these will be yet.

---

The trend makes sense though, think about it:

G80 => GT200 => GF110/100 => GK100

128 => 256 (240 usable) => 512 => 1024???

Considering the pattern NV has been following for quite a while now, it wouldn't surprise me if it does have 1024 CUDA cores. Though I do wonder how they could possibly cool something of that monstrosity. Even at 28nm that chip would have to be over 500 mm^2 not to mention consume a butt load of power

What makes me nervous is that all those games on the last slide use tessellation, If I were marketing something I'd show off what I improved... a lot.

I want something faster than my 480 and I refuse to go dual cards ever again, hope this sucker will be epic. Can't wait for the reviews.

Quote:

---

Originally Posted by nascasho

What makes me nervous is that all those games on the last slide use tessellation, If I were marketing something I'd show off what I improved... a lot.

I want something faster than my 480 and I refuse to go dual cards ever again, hope this sucker will be epic. Can't wait for the reviews.

---

agreed, if tessellation games increased 60-250% then games that were not using much tessellation might be only 40-60% faster

Quote:

---

Originally Posted by nascasho

What makes me nervous is that all those games on the last slide use tessellation, If I were marketing something I'd show off what I improved... a lot.

I want something faster than my 480 and I refuse to go dual cards ever again, hope this sucker will be epic. Can't wait for the reviews.

---

Thoses slides is not complete it miss a lot of pages... this page is for show how the tesselation have increased

Quote:

---

Originally Posted by AliG

The trend makes sense though, think about it:

G80 => GT200 => GF110/100 => GK100

128 => 256 (240 usable) => 512 => 1024???

Considering the pattern NV has been following for quite a while now, it wouldn't surprise me if it does have 1024 CUDA cores. Though I do wonder how they could possibly cool something of that monstrosity. Even at 28nm that chip would have to be over 500 mm^2 not to mention consume a butt load of power

---

Yes, it makes sense and I wouldn't be surprised either. However seeing the issues NVIDIA had in the past products due to the big DIEs I kinda expect them to opt for different approach this time... But you never know :)

Quote:

---

Originally Posted by AliG

The trend makes sense though, think about it:

G80 => GT200 => GF110/100 => GK100

128 => 256 (240 usable) => 512 => 1024???

Considering the pattern NV has been following for quite a while now, it wouldn't surprise me if it does have 1024 CUDA cores. Though I do wonder how they could possibly cool something of that monstrosity. Even at 28nm that chip would have to be over 500 mm^2 not to mention consume a butt load of power

---

The GT200 actually had 240 CCs on the chip… I think.

One thing to note is the GT200 had a shrink from 65 nm to 55 nm, so from G80 to GF100 there have been 2.5 node shrinks.

G80 » GT200 : 90 nm » 65 nm, 484 mm^2 » 576 mm^2 (1.19x), 128 » 240 (1.88x) : 1.58x CCs/mm^2
GT200b » GF100 : 55 nm » 40 nm, 470 mm^2 » 530 mm^2 (1.13x), 240 » 512 (2.13x) : 1.89x CCs/mm^2

And that's not to mention TDPs. So if GF110 » Kepler is anything like the above in terms of CCs/mm^2 increase, I think there would be fewer than 1024 CCs in the top-end Kepler chip unless they pull some magic with the SPs or other stuff or they conjure up a chip close to 600 mm^2. My guess is 640 < CCs ≤ 960.

Quote:

---

Originally Posted by Lanek

Thoses slides is not complete it miss a lot of pages... this page is for show how the tesselation have increased

---

Thanks for clarifying, I was wondering what that chart meant.

Also, you're right on the slides bunch more to the looks of it on neogaf, but they're too small to read.

Yes + there's some faked pages who turn around ... modified one from 6970 launch ,so take all this with a lot of precaution ( the last slide look correct )..

http://www.abload.de/image.php?img=slides0vka9.png (note they are completely unreadable )



Is the launch really advanced to the 22?

AMD Brings the Launch of Radeon HD 7900 Series Forward: December 22nd, 2011

Read more: http://vr-zone.com/articles/amd-brin...#ixzz1ge0ZjbQj

It better score at least 10K in 3dmark2011 graphics total score, but according to ORB leak it will be only ~7000?

I hope hes faking it lol, that's what i get now when i OC my 570gtx to 950mhz:D

Quote:

---

Originally Posted by Manicdan

agreed, if tessellation games increased 60-250% then games that were not using much tessellation might be only 40-60% faster

---

Only 40-60% faster? That would be huge performance jump. I would be happy with that type of average. 40-60 would be the type of performance improvement we saw with the 5870 and 4890. I think with a brand new architecture and immature drivers we will be lucky to see that. But if you look at the past, these marketing slides tends to exaggerate quite a bit and are atleast 10% off(i.e from marketing slides 6970 was supposed to be 15% faster than a gtx 480 which would put it at gtx 580 speeds).

Quote:

---

Originally Posted by The|Hunter

It better score at least 10K in 3dmark2011 graphics total score, but according to ORB leak it will be only ~7000?

I hope hes faking it lol, that's what i get now when i OC my 570gtx to 950mhz:D

---

Dunno I have 10K ( graphic score ), with 2x 5870@1040ghz. + 2600K@5.2ghz. ( no mipmap etc )

Quote:

---

Originally Posted by DilTech

That's going with the 4 split rather than the 8 split, still matches the mathematics we've always seen. They didn't disable RBE's, they disabled half of the Rops per RBE.

I seriously can't even remember the last time we saw a gpu that wasn't symmetrical rop to memory controller. In fact, I'm now curious... When IS the last time AMD/ATi released a chip that wasn't? (serious question) AFAIK both brands ARE actually tied together in this aspect presently.

---

Long forgot R600 a.k.a. HD2900 had decoupled ROPs from memory channels. There were 16ROP - 512bit cards based on R600 as well as 16ROPs - 256bit cards (HD2900GT) based on the same R600 chip. This was possible due to massive 1024bit ring-bus which was dropped in later chips (starting from RV770) due to power, die area and efficiency reasons. You also could disable only one quad-ROP if needed and they were also decoupled from texture units. R600 was very nice on paper, we all know how it fared in real world :D

Quote:

---

Originally Posted by Lanek

Dunno I have 10K ( graphic score ), with 2x 5870@1040ghz. + 2600K@5.2ghz. ( no mipmap etc )

---

Gpu score or total score?


At first i was hopping for ~ 14K by gpu P score, lol now im at 10K.. Imo anything lower is really a let down.


Those 64 tessellator engines should do some magic.. and its just a 1280x720 reso for Christ sake lol :D

Quote:

---

Originally Posted by The|Hunter

It better score at least 10K in 3dmark2011 graphics total score, but according to ORB leak it will be only ~7000?

I hope hes faking it lol, that's what i get now when i OC my 570gtx to 950mhz:D

---

I think it was Vapor who said it, but I don't think we're even allowed to talk about OBR/his site anymore.

Just trying to keep you outta trouble brotha:up:

Quote:

---

Originally Posted by The|Hunter

Gpu score or total score?


At first i was hopping for ~ 14K by gpu P score, lol now im at 10K.. Imo anything lower is really a let down.




Those 64 tessellator engines should do some magic.. and its just a 1280x720 reso for Christ sake lol :D

---

graphic score only (not Pscore ).. If we speak about Pscore: GTX590 stock = 8800-9000 and 6990 score in the 9300 ....even a GTX580SLI pull just 10K score. ( stock speed ofc ). (i have take thoses number in the review of thoses cards, so drivers have maybe increase a little bit the scores then .) 14K for a single GPU is a little bit unrealistic ( i think scores dont scale really well, due to the numbers of test, their different nature, and the way they are calculate).

(for add to the example 2x 590 SLI ( quad SLI ) score 13425 and 2x 6990 ( quadcfx ) score 13734 )

Quote:

---

Originally Posted by Lanek

graphic score only (not Pscore ).. If we speak about Pscore: GTX590 stock = 8800-9000 and 6990 score in the 9300 ....even a GTX580SLI pull just 10K score. ( stock speed ofc ). (i have take thoses number in the review of thoses cards, so drivers have maybe increase a little bit the scores then .) 14K for a single GPU is a little bit unrealistic ( i think scores dont scale really well, due to the numbers of test, their different nature, and the way they are calculate).

(for add to the example 2x 590 SLI ( quad SLI ) score 13425 and 2x 6990 ( quadcfx ) score 13734 )

---

Ok tnx, i guess 10K P gpu score isn't so bad, that's if it scores 10K at all hehe :)

I will not too much wait about score in 3Dmark suite at launch, ( first cause what i want to know it is how they OC, and what the score OC ),
but remember whatever it is Nvidia or AMD, in general in 3Dmark, we see a big boost 1-2 drivers then when this is a new architecture ( 3Dmark06 + cypress, a 2000pnts boost on the second official driver, same in vantage. and for Fermi it was the exact same thing with the GTX480. )

For thoses type of bench, new arch need allways some drivers tweaking for perform at their full potential.

Specially 3Dmark11 is a lot more complex in what it test compared to old benchmark ( who was mostly graphic based ), Physic, OpenCL, computing, + shader performance... a driver for a new architecture with new or enhanced performance in 3Dmark11 will surely need a lot of tweaking.

Well it looks like it can do at least 1075mhz,

http://www.abload.de/img/amdradeonhd7970_pb_7_oorme.jpg

I will not be surprised we will see factory overclocked card at 1ghz+ ... specially AMD have open the possibility for AIB to present a lot of different cards at launch for the 7950. ( hence why the launch is maybe still fixed at January for them ). It should come fast too then for the 7970.

For be honest i really dont know what can be the real performance of the card. Too much change. I will not speculate too much, and just wait the surprise ( good or bad ).

Quote:

---

Originally Posted by The|Hunter

Ok tnx, i guess 10K P gpu score isn't so bad, that's if it scores 10K at all hehe :)

---

Last time I checked, the HD6970 scored something like... P5500 or P6000 (correct me if I'm wrong). I think you shouldn't expect more than P7500~P8000. It has 33% more shaders, and a new architecture, why would you expect it to run almost twice as the last gen?

My guess is somewhere between ~7000 and ~7500, and it would be very nice. If it hits 8000 overclocked I could swap my HD5850 Crossfire for one HD7970. I hit something like 8800, but when performance is almost identical, 1 card >>> 2 cards.


Edit: Oh, my mistake. Sorry, I thought you were talking about overall score. I don't pay much attention to the GPU score, so I don't really know.

it seems like we have a pretty good idea and semi official specs for the 7970. what i dont know is where people are getting their information about the high end keplers. i personally have seen nothing but wild speculation on the specs of the high end keplers and yet people are claiming its going to wipe the floor with the 7970? i would wait until we have more concrete info on kepler. just remember, it was reported that the high end 7000 series was going to get xdr2 memory and we see how that worked out :rolleyes:

Quote:

---

Originally Posted by Andi64

Last time I checked, the HD6970 scored something like... P5500 or P6000 (correct me if I'm wrong). I think you shouldn't expect more than P7500~P8000. It has 33% more shaders, and a new architecture, why would you expect it to run almost twice as the last gen?

My guess is somewhere between ~7000 and ~7500, and it would be very nice. If it hits 8000 overclocked I could swap my HD5850 Crossfire for one HD7970. I hit something like 8800, but when performance is almost identical, 1 card >>> 2 cards.


Edit: Oh, my mistake. Sorry, I thought you were talking about overall score. I don't pay much attention to the GPU score, so I don't really know.

---

It's really not easy to compare the actual SP on Cayman, and the GCN architecture, it will be too long to explain it here but Anandtech have make a good and understandable preview in June http://www.anandtech.com/show/4455/a...-for-compute/3

Again, i dont know how it will perform, I just explain it is hard to compare SP numbers. It will be possible if it was a die shrink of the Cayman VLIW4 port, But hardly with this architecture.

Quote:

---

Originally Posted by phantomferrari

it seems like we have a pretty good idea and semi official specs for the 7970. what i dont know is where people are getting their information about the high end keplers. i personally have seen nothing but wild speculation on the specs of the high end keplers and yet people are claiming its going to wipe the floor with the 7970? i would wait until we have more concrete info on kepler. just remember, it was reported that the high end 7000 series was going to get xdr2 memory and we see how that worked out :rolleyes:

---

It's mainly just because Kepler is looking to be double the 580 (slides have popped up showing around double 580 performance), which unless 7970's new design has some amazingly improved performance per shader ratio then it's heavily doubtful that it'll be capable of putting up those kind of numbers. We're expecting a 30-40% improvement on the AMD side of things, NVidia are aiming for doubling up there end it would seem. Even the GK104 (nvidia's mid-range part) should match the 7970 if the specs are true.

That's why.

All that said, AMD doesn't aim at NVidia's top-end parts in the first place with single gpu's. So 780 (if that's it's name, that's what the slide claims) will be likely up against a 7990 instead anyway.

Quote:

---

Originally Posted by Manicdan

can someone measure the length of the 7970 for us based on the image showing the side of the card and PCIe slot, thanks!

---

Working on it.

Edit- The sideways card, bad slide to use, I got ~11.25inches or 28.6cms for the shroud. Used 7.1cm for the pci-e slot.
ReEdit- Using the slide below, I got 27.89cms or ~11inches which I think is more accurate.
http://www.abload.de/img/amdradeonhd7970_pb_2_slpnh.jpg

Quote:

---

Originally Posted by Pantsu

Still using a vapour chamber? I thought they were switching to those liquid chambers, or is 6th generation vapour chamber the same thing?

These slides look a bit more legit that the OBR ones, but still I wouldn't look too deep into performance numbers. Nice to see DX11.1 and PCI-e 3.0 supported though. :)

---

Liquid chambers for slim professional cards. Takes awhile for them to ramp up mass production.

NEW Launch date is Dec 22nd!


http://www.techpowerup.com/156876/AM...cember-22.html

Allready posted one page before, m8 .... but no worry.

Yes, it seems confirmed by some other site too. ( VRzone ).

http://www.obr-hardware.com/2011/12/...ance-info.html

Quote:

---

Okay, i have NOT a cards (only few ppl have it now) but i have some performance numbers. I have real results from Live presentation in 3D Mark 11, Unigine Heaven, Battlefield 3, Dirt 3 and Crysis 2.

Have numbers from slides too (unreal PR stunts here) from slide above. I can say how HD 7970 perform, i know it from these few real tests ... stay tuned, all will be here soon.

If i am telling, performance is only few percents above HD 6970 (in Games not 3D11 and Unigine) it is NOT speculation, but info from real GAME results ...

---

It's really hard to enforce this no OBR rule when it's the only source of "info".

Quote:

---

Originally Posted by spicypixel

It's really hard to enforce this no OBR rule when it's the only source of "info".

---

Really??? There is info all over the place if you know where to look or who to listen to.
Why would you even listen to him for GPUs is beyond me since he has been giving out misinformation for the last couple of launches.

Quote:

---

Originally Posted by spicypixel

It's really hard to enforce this no OBR rule when it's the only source of "info".

---

don't forget donanimhaber :)

So can we stop quoting OBR then? =/

-PB

yes, please. let him sink his own boat.

Quote:

---

Originally Posted by HelixPC

NEW Launch date is Dec 22nd!


http://www.techpowerup.com/156876/AM...cember-22.html

---

SkyMTL was once again spot on when he said weeks ago that if the launch date ends up in 2011 that it will be a paper launch. :clap:

Either way, glad to know we're almost ready for the next battle of the titans.