ATI`s R600 News Thread

Code:

---

SPECS
=============================
65nm
64 Shader pipelines (Vec4+Scalar)
32 TMU's
32 ROPs
128 Shader Operations per Cycle
800MHz Core
102.4 billion shader ops/sec
512GFLOPs for the shaders
2 Billion triangles/sec
25.6 Gpixels/Gtexels/sec
256-bit 512MB 1.8GHz GDDR4 Memory
57.6 GB/sec Bandwidth (at 1.8GHz)
WGF2.0 Unified Shader

---

so what are the G80's specs in that kind of format?? so that we can do direct comparison between the 2.

1 problem with GeForce 7900GTX vs X1900XTX was that nvidai and ATI release specs in different format , so there was not direct comparison. Like ATI X1900XTX has pixe 48 shaders, but how many does the GeForce 7900GTX have? by simply comparing:

1. traingles / sec
2. texture fillrate
3. pixel fill
4. mem bandwidth

doesnt seem enough. Its important to know how many :

1. pipes
2. unified shaders
3. ROPS
4. TMUs

for each GPU have to make good comparison. I wouldnt worry too much about memory bandwidth as they r both really high, shouldnt make that much of a difference. I think GPU is more important. Look at Oblivion benchmark, X1900XTX just kills GeForce 7900GTX. While in some games like Battlefield 2, GeForce 7900GTX has the upper hand.

PS: not sure whether pipes and unified shaders can have different number or not. ie another perfect ratio?

suddenly my new x19509xtx seems slooow..

R600 could be the card that runs as a new 9700, exploding into the dx10 era.

No matter what, they need to make a good middle card and then it might be a great start for new vista.

Quote:

---

Originally Posted by nn_step

:D The least expected of the two

---

So I assume its the R580 your referring to since G80 is said to be a power hog.

Quote:

---

Originally Posted by v_rr

No, you are incorrect.

And some people think that G80 have 128 pipes. That is complitly wrong. It´s markting. G80 will have 32pipes that are acomplished with 128 pixel processors. In the case of G71 there are 24 pipes with 48 ALU´s.

Nvidia's G80 has 32 pixel pipes
http://www.theinquirer.net/default.aspx?article=32856

yes there are pipes. And in the case of G80 you don´t know how many they are.
It is said 128 pixel processors/ALU´s, but the G71 has 48 ALU´s and only 24 pipes, so the G80 could have 32 pipes and 128 ALU´s, but the info isn´t out yet.

ATI's R600 has 64 real pipes
http://uk.theinquirer.net/?article=33784

And do you think that because R600 is unified that 64 shaders only? Hitch pipe has can calculate pixel/processor/geometry, but at diferent speed. The pixel part can be 3x faster than the vertex or geometry part.

So you can explayning in a simpler way that R600 has 198 pixel shaders, just because those 64 pipes do the pixel work 3x times faster than the other tipe of work.

And also can be there pixel processors just like the R580. That doesn´t invalidate that de GPU will be unified.

---

Rule number 1 of engagement... Never use the inquirer as your source, especially in multiple cases. EVEN MORE SO WHEN YOU USE AN ARTICLE THAT HAS NOW BEEN PROVEN FALSE THAT CLAIMS THE G80 ISN'T UNIFIED!

Pipes aren't included in a unified architecture. This is a FACT. Don't tell me "but the inquirer told me this, the inquirer told me that", it's bs. End of story.

Second, don't tell me "but they carry 3 - 4 pixel shaders per pipe"...It's bs. Why? There's no longer physical pixel shaders. It's unified shaders. They handle pixel, vertex, and geometry.

This is the make-up of a unified part..
TMU
Shader
ROP

That's it. End of the game.

If it's 64 anything, it's 64 unified shaders.

Quote:

---

Originally Posted by DilTech

Rule number 1 of engagement... Never use the inquirer as your source, especially in multiple cases. EVEN MORE SO WHEN YOU USE AN ARTICLE THAT HAS NOW BEEN PROVEN FALSE THAT CLAIMS THE G80 ISN'T UNIFIED!

Pipes aren't included in a unified architecture. This is a FACT. Don't tell me "but the inquirer told me this, the inquirer told me that", it's bs. End of story.

Second, don't tell me "but they carry 3 - 4 pixel shaders per pipe"...It's bs. Why? There's no longer physical pixel shaders. It's unified shaders. They handle pixel, vertex, and geometry.

This is the make-up of a unified part..
TMU
Shader
ROP

That's it. End of the game.

If it's 64 anything, it's 64 unified shaders.

---

Just so I understand what your saying .

You are saying that

G80 has
64tmu's
64 Rop's
And can perform 256 shader opp's per cycle.

lol...i'm not brave enough to start arguing with Diltech on Nvidia. :p:

The numbers look promising...but its vaporware until I see some real physical shots.

Plus, I hope everyone knows that G81 is the H1 2007 timeframe...so "Anything ATI does, Nvidia can do better" ;)

Perkam

I think R600 will be 80nm
If ATI use 65nm,maybe the yield not good
because R600 will have plenty of transistor and die size ..

Yeah, the die on that thing is going to be SO frickin massive, like 1" by 1" lol.

Quote:

---

Originally Posted by perkam

lol...i'm not brave enough to start arguing with Diltech on Nvidia. :p:

The numbers look promising...but its vaporware until I see some real physical shots.

Plus, I hope everyone knows that G81 is the H1 2007 timeframe...so "Anything ATI does, Nvidia can do better" ;)

Perkam

---

Eh, it's release time again, you know how it goes when it comes to videocards. In all actuality, I'd been betting on the R600 to win, and part of me still thinks it will, which is why I think the current "rumored specs" is complete BS. :fact: I really don't see them doing a 512bit memory bus though, not with only a 500 million transistor die-size atleast... If it uses a 512bit, then I'd say there's a reallllllllllllly good chance that it will only be 64 unified shaders due to die size though.

As for the G81, I'm not even going to go there, for tax reasons I have to buy a card this year so next year I don't even want to bring into my mind on these cards. I'll be using a G80 for awhile.

Quote:

---

Originally Posted by Turtle 1

Just so I understand what your saying .

You are saying that

G80 has
64tmu's
64 Rop's
And can perform 256 shader opp's per cycle.

---

No, I'm saying that if the number 64 holds true, it's 64 unified shaders.

We know the G80 is 128 unified shaders, that much there's no point in questioning at this time. As for how many operations it can perform per cycle, no clue, we'll find that out soon enough on how many threads each shader can perform per cycle.

We have no word on the amount of TMU's, and no clue on the amount of Rops, but I highly doubt it's 64. Neither of these cards would really make use of 64 rops. To give you an idea, the 6800Ultra had 16 rops, the 7800GTX had 16 rops, the 7900GTX had 16 rops, the x1800xt had 16 rops, x1900xt/x, and x1950xtx also had 16 rops. As clockspeed raises, these become less of an issue, especially for them to jump the rops up 4 times. This amount hasn't changed in the past 2 generations(2 full years). 24 to 32 I can see, but 64 would just be a waste as it'd be more efficient and a bigger gain to use that space to add more unified shaders.

Quote:

---

Originally Posted by DilTech

Rule number 1 of engagement... Never use the inquirer as your source, especially in multiple cases. EVEN MORE SO WHEN YOU USE AN ARTICLE THAT HAS NOW BEEN PROVEN FALSE THAT CLAIMS THE G80 ISN'T UNIFIED!

Pipes aren't included in a unified architecture. This is a FACT. Don't tell me "but the inquirer told me this, the inquirer told me that", it's bs. End of story.

Second, don't tell me "but they carry 3 - 4 pixel shaders per pipe"...It's bs. Why? There's no longer physical pixel shaders. It's unified shaders. They handle pixel, vertex, and geometry.

This is the make-up of a unified part..
TMU
Shader
ROP

That's it. End of the game.

If it's 64 anything, it's 64 unified shaders.

---

thats what i was trying to say! thank you for pointing that out more claerly!!! and yes the g80 can do physics by itself without another gpu!!!! it uses some of the shaders or pipes or whatever to do physics and i have seen the demo and it looks very good! i will wait to see who wins this gun battle and choose what i get.......the r600 or g81(by then)

it seems to me that even with unified shaders , u can still have different number of pipelines. didnt some site said that nVidia hasnt release the number of pipelines the G80 has yet?

also nvidia hasnt said how many ROP and TMU it has. or did it?

given ATI's reputation, remember R520's legacy? it was only marginally faster than GeForce 7800GTX 256, then about 1 month later nVidia launched GeForce 7800GTX 512 and just killed the X1800XT...

i seriously doubt the r600 will have 512bit bus. seriously the pcb would be huge and need alot of memory chips.

i bet its 512bit ringbuss just like the x1900 cards have. cmon you know its too expensive to put that many damn wires lol. cheaper to just put faster ram.

Yeah, I think we may see another back and forth round like the G7x vs R5xx. I don't think either camp will be able to improve much though without first shrinking to 65nm.

The real war is going to be the mid-range for both camps, maybe not in our eyes, but in their profits eyes it's the midrange that matters more than anything else here.

Quote:

---

Originally Posted by brandinb

i seriously doubt the r600 will have 512bit bus. seriously the pcb would be huge and need alot of memory chips.

i bet its 512bit ringbuss just like the x1900 cards have. cmon you know its too expensive to put that many damn wires lol. cheaper to just put faster ram.

---

Something I've learned about ATi... Never put anything past them.

Quote:

---

Originally Posted by DilTech

Yeah, I think we may see another back and forth round like the G7x vs R5xx. I don't think either camp will be able to improve much though without first shrinking to 65nm.

The real war is going to be the mid-range for both camps, maybe not in our eyes, but in their profits eyes it's the midrange that matters more than anything else here.



Something I've learned about ATi... Never put anything past them.

---

QFT. The performance will most likely be somewhat even between the two like it has been two past generations.

ATI clearly knows when to step up to the plate, an IQ advantage and software voltages help, but CCC kills it :(

I just hope ATi drivers improve, because the hardware is wonderful.
(the drivers just suck)

yeah CCC blows ass for lack of better words.

I can't even adjust my AA/AF settings with full control.

3.7 cats ruled it all. lol.

but i cant change vGPU or vGmem wiht any software , however i can do that with ATI tools (for ATI X1800 and X1900 )... or there is a way to softmod ur GeForce's vGPU ??

Quote:

---

Originally Posted by theteamaqua

but i cant change vGPU or vGmem wiht any software , however i can do that with ATI tools (for ATI X1800 and X1900 )... or there is a way to softmod ur GeForce's vGPU ??

---

depends on what card, but most likely no...

dont worry though, the soft voltage increases on the ati's isnt very good imo, need to hardmod anyways for real clockers.

Quote:

---

Originally Posted by Revv23

depends on what card, but most likely no...

dont worry though, the soft voltage increases on the ati's isnt very good imo, need to hardmod anyways for real clockers.

---

Except that hardmodding cards that easily die isnt much fun in the end:fact:

How much will a FlagShip Card Cost ????

Quote:

---

Originally Posted by ANP !!!

How much will a FlagShip Card Cost ????

---

probably $600+ when first released because of high demand.

Quote:

---

Originally Posted by brandinb

i seriously doubt the r600 will have 512bit bus. seriously the pcb would be huge and need alot of memory chips.

i bet its 512bit ringbuss just like the x1900 cards have. cmon you know its too expensive to put that many damn wires lol. cheaper to just put faster ram.

---

The memory ring bus used in R520, R580 is already 512bit internatly, but communicates with the ram at 256bit externally. So internally its already designed to operate at 512bit, shouldn't be a problem implementing that with R600

Even so, if the numbers are doubled with g80 and a little more with r600 from the previous gen, its just a sign where heat and additions of transistors becomes a jumble until they go 65nm.

I just hope they are effective running todays games with full AA and AF without a hitch blowing away the state of the art gen today.
12k or whatever in 3dmark06 is impressive if the numbers are correct but still gameframes is what counts. (at least for me) Not that extreme ;)

64 unified shaders seem to be too few, remember xbox xenos already has 48 unified shaders, there has to be something else (maybe they can perform 3 shader operations/clock), no one (except ati+nvidia) really knows how many shader-operations both chips can perform, maybe one ati unified shader has 3 alus and the nvidian 2, there is no point to discuss about this without more detailed information...