G80 De-Mystified; GeForce 8800GTX/GT

It says it's scalable to 1.5Ghz just like how G70 was supposed to scale to 1Ghz. So don't expect anything too high.

For the longest time, they've been saying they won't have unified shader, and it's not an easy change to make it a unified shaders. So my guess to it would be, it's not true unified shaders, but NV will be doing things their own way to satisfy DX10 requirements.

The dual memory standard is a little shocking, perhaps they are offloading the vertex shaders into that memory to improve performance in DX10 since Unified Shaders in DX10 requires pixel, vertex and geometry shaders. So I wouldn't be surprised if the 2nd set of memory is used to process one of the above shaders.

Honestly the market of money determines Nvidia and ATI card makes. I mean if this weren't true they wouldn't be in business.

Through out the DX series up to 9 there really hasn't be a revolution to speak of but small advancments a generation at a time. From pipelines, interface sizes, die sizes, voltages, RAM size/type..... So it'll be interesting to see if the makers take the DX10 to more of a revolution which would be the specs in the original post which are crazy!

But considering the past I am skeptical of such a huge change. Power/heat is going to be the issue with dual core/700M transistors. I'd hope that it would be 80-65nm size then!

December (late) - January is when we're looking forward to the G80 and R600 coming out. I'm not going to purchase either because that means getting Vista also. And SLI 7900GTs vmod to GTX is plenty of power for gaming.

No matter what, we'll sit and wait. :toast:

Quote:

---

Originally Posted by Starscream

could people pls drop the bit about 1.5Ghz core not possible etc.

we dont even know how accurate all these specs are.
And they dont say the core will be clocked at 1.5Ghz but will be scallable upto with this architekture wich doesnt mean that wel actualy see it happen.
And btw most likely there will also be a 65nm refresh of it at one point, so if they go high core clock near 1.5Ghz it will likely be with the 65nm refresh and not with the 80nm core.

It just kinda dumb that people keep repeating things are impossible that they misread from the TS.

ps.
all these new features are likely just to make a bigger gap perf. between the high-end and mid-end.
Till now the high-end always costed alot more then the mid-end but in some games the peformence gap wasnt always big enough to justify the price diffrence.
Wich makes me wonder what performence the mid-end will have asin pipes and bit bus.

---

Yeah, i agree...remember the netburst "scalable up to 10Ghz".....? Maybe its just possible, in theory, 1.5Ghz in 65nm. Imo we will see something like 750Mhz at first

I think that it can be possible, i mean we have cores up an running at almost 1 ghz already. Gradually we will see 1,5ghz, but first i think the cores will be at 800mhz and then later on scale higher and higher just like with the 7800<7900. I prefer use the 6800<7900 cuz to me thats almost indentical core.

Quote:

---

Originally Posted by thephenom

It says it's scalable to 1.5Ghz just like how G70 was supposed to scale to 1Ghz. So don't expect anything too high.

For the longest time, they've been saying they won't have unified shader, and it's not an easy change to make it a unified shaders. So my guess to it would be, it's not true unified shaders, but NV will be doing things their own way to satisfy DX10 requirements.

The dual memory standard is a little shocking, perhaps they are offloading the vertex shaders into that memory to improve performance in DX10 since Unified Shaders in DX10 requires pixel, vertex and geometry shaders. So I wouldn't be surprised if the 2nd set of memory is used to process one of the above shaders.

---

NVidia made ONE statement, saying they'll make a unified part when the time is right, in july 2005. They said there was still more milage in the G70 architecture, and there was. G71 was what they were talking about then! :fact:

The reason they tried to downplay unified shaders was to hype up the battle between the xbox360 and the ps3, as the PS3 doesn't use a unified setup, while the xbox360 does. With NVidia helping on the PS3, of course they're going to try to downtalk the xbox360 gpu, no?

People have since then used this phrase out of context again and again... You guys do know that NVidia has a PATENT for the unified shader/unified shader algorithm, right? If not, stop, and do a quick search here on the word patent, and you'll find where I posted it.

NVidia said the G80 architecture is what will be used for the next few years, so there's just no way it won't be unified.

For your consideration:
325Mhz GF4
475Mhz FX
400Mhz 6800
500Mhz 6600GT
450Mhz G70 (550 512MB version)
560Mhz 7600GT
650Mhz G71

power consumption creeped up slowly (all standard non OC models):
59W FX 5900U
72W 6800U
80W 7800GTX
84W 7900GTX

1st and foremost, no manufacturer expecting a profit is going to push technology to the breaking point.. it has to be producable in mass quantity and with enough clock speed margins that not just those in Finland but also Singapore can enjoy the hardware.

I will wager my whole Star Wars collection that not only will we not see 1.5Ghz or 750M transistors in 2006, but that it wont happen in 2007 either. Clock speed increases slowly and incrementaly. There has never been such drastic jump, even though often there was new smaller node lithography process. Since G80 will be 80/90nm, it doesn't have advantage of smaller node (ie 65nm), making any such outlandish claims of clockspeed pure fiction. Does it perhaps "scale" to that speed... well, does a 386 scale to 800Mhz? No point in such impractical speculation.

People often cite 200, 250 even 300W for next generation video cards. Do you folks realize how much heat that is?? At only about 90W, the 7900GTX uses a dense fin heatpipe heatsink. To dissipate 300W you would need a heatsink 3 times as large, probably taking up 3-4 slots. Much more likely the number quoted is the max power envelope for a SLI/Crossfire system. And, even then a 125-150W single slot solution is quite risky.

# of transistors: 25M GF2, 57M GF3, 63M GF4, 125M FX, 222M 6800, 300M G70. When new DirectX technology is released, there is appropriate big jump in transistor count. When DirectX refreshes are released (GF4, G70), transistor count increases moderately.

Unified shaders are very easy to make. Afterall, in a sense thats what CPU's did before GF3/GF4 came on the market. GPUs are able to calculate orders of magnitude faster than CPU because of exploiting parallelism and specialization. Customized circuit specialized for specific function faster than general purpose. Unified shader idea goes against all this. I believe what is meant is that unified shader, totally general purpose, cannot be made as fast as separate specialized.

In conclusion, G80 like everything in life, will be limited by physics and compromises. Due to 80/90nm manufacturing, probably within 500Mhz-800Mhz range. If indeed number of processors (shaders) has been increases so dramatically, to stay within reasonable power limits, voltage is probably lowered, and lower clock speed. And although 384 bit memory interface is unlikely but feasable, yields and economics make 2 chip solution impractical.

Most likely around 900 clock area and overclockable to 1ghz

Due to ATI's delay before nVidia the Spec's of the Hardware are most likely Leaked on purpose for other reasons than what the G80 will finally be stocked with.

Quote:

---

Originally Posted by ***Deimos***

For your consideration:
325Mhz GF4
475Mhz FX
400Mhz 6800
500Mhz 6600GT
450Mhz G70 (550 512MB version)
560Mhz 7600GT
650Mhz G71

power consumption creeped up slowly (all standard non OC models):
59W FX 5900U
72W 6800U
80W 7800GTX
84W 7900GTX

---

hehehe, excellent post. Pretty much says it all really, the G80 specs are rubbish.

Is there a ball-park release date for this card yet?
yeah, i bet its been asked a zillion times, but I don't feel like looking.

There is no doubt it will support the unified shading architecture/API.. otherwise it won't be dx10 compliant. But one thing nvidia made certain is that their card will not be fully unified. They have said this many times ("when the time is right", "finding a balance"), and there is a good possibility that only parts of it will be (vertex and geometry shaders could very well be unified).

Why doubt facts.

Quote:

---

Originally Posted by ***Deimos***

For your consideration:
325Mhz GF4
475Mhz FX
400Mhz 6800
500Mhz 6600GT
450Mhz G70 (550 512MB version)
560Mhz 7600GT
650Mhz G71

power consumption creeped up slowly (all standard non OC models):
59W FX 5900U
72W 6800U
80W 7800GTX
84W 7900GTX

1st and foremost, no manufacturer expecting a profit is going to push technology to the breaking point.. it has to be producable in mass quantity and with enough clock speed margins that not just those in Finland but also Singapore can enjoy the hardware.

I will wager my whole Star Wars collection that not only will we not see 1.5Ghz or 750M transistors in 2006, but that it wont happen in 2007 either. Clock speed increases slowly and incrementaly. There has never been such drastic jump, even though often there was new smaller node lithography process. Since G80 will be 80/90nm, it doesn't have advantage of smaller node (ie 65nm), making any such outlandish claims of clockspeed pure fiction. Does it perhaps "scale" to that speed... well, does a 386 scale to 800Mhz? No point in such impractical speculation.

People often cite 200, 250 even 300W for next generation video cards. Do you folks realize how much heat that is?? At only about 90W, the 7900GTX uses a dense fin heatpipe heatsink. To dissipate 300W you would need a heatsink 3 times as large, probably taking up 3-4 slots. Much more likely the number quoted is the max power envelope for a SLI/Crossfire system. And, even then a 125-150W single slot solution is quite risky.

# of transistors: 25M GF2, 57M GF3, 63M GF4, 125M FX, 222M 6800, 300M G70. When new DirectX technology is released, there is appropriate big jump in transistor count. When DirectX refreshes are released (GF4, G70), transistor count increases moderately.

Unified shaders are very easy to make. Afterall, in a sense thats what CPU's did before GF3/GF4 came on the market. GPUs are able to calculate orders of magnitude faster than CPU because of exploiting parallelism and specialization. Customized circuit specialized for specific function faster than general purpose. Unified shader idea goes against all this. I believe what is meant is that unified shader, totally general purpose, cannot be made as fast as separate specialized.

In conclusion, G80 like everything in life, will be limited by physics and compromises. Due to 80/90nm manufacturing, probably within 500Mhz-800Mhz range. If indeed number of processors (shaders) has been increases so dramatically, to stay within reasonable power limits, voltage is probably lowered, and lower clock speed. And although 384 bit memory interface is unlikely but feasable, yields and economics make 2 chip solution impractical.

---

Obviously you have forgotten the last of the 478 pins of Netburst technology. :-p.

Quote:

---

Originally Posted by ***Deimos***

80W 7800GTX
84W 7900GTX

---

I was under the impression that the 7900gtx used LESS power:

here is my evidence:
http://techreport.com/reviews/2006q1.../index.x?pg=13

Quote:

---

Originally Posted by ***Deimos***

For your consideration:
325Mhz GF4
475Mhz FX
400Mhz 6800
500Mhz 6600GT
450Mhz G70 (550 512MB version)
560Mhz 7600GT
650Mhz G71

power consumption creeped up slowly (all standard non OC models):
59W FX 5900U
72W 6800U
80W 7800GTX
84W 7900GTX

1st and foremost, no manufacturer expecting a profit is going to push technology to the breaking point.. it has to be producable in mass quantity and with enough clock speed margins that not just those in Finland but also Singapore can enjoy the hardware.

I will wager my whole Star Wars collection that not only will we not see 1.5Ghz or 750M transistors in 2006, but that it wont happen in 2007 either. Clock speed increases slowly and incrementaly. There has never been such drastic jump, even though often there was new smaller node lithography process. Since G80 will be 80/90nm, it doesn't have advantage of smaller node (ie 65nm), making any such outlandish claims of clockspeed pure fiction. Does it perhaps "scale" to that speed... well, does a 386 scale to 800Mhz? No point in such impractical speculation.

People often cite 200, 250 even 300W for next generation video cards. Do you folks realize how much heat that is?? At only about 90W, the 7900GTX uses a dense fin heatpipe heatsink. To dissipate 300W you would need a heatsink 3 times as large, probably taking up 3-4 slots. Much more likely the number quoted is the max power envelope for a SLI/Crossfire system. And, even then a 125-150W single slot solution is quite risky.

# of transistors: 25M GF2, 57M GF3, 63M GF4, 125M FX, 222M 6800, 300M G70. When new DirectX technology is released, there is appropriate big jump in transistor count. When DirectX refreshes are released (GF4, G70), transistor count increases moderately.

Unified shaders are very easy to make. Afterall, in a sense thats what CPU's did before GF3/GF4 came on the market. GPUs are able to calculate orders of magnitude faster than CPU because of exploiting parallelism and specialization. Customized circuit specialized for specific function faster than general purpose. Unified shader idea goes against all this. I believe what is meant is that unified shader, totally general purpose, cannot be made as fast as separate specialized.

In conclusion, G80 like everything in life, will be limited by physics and compromises. Due to 80/90nm manufacturing, probably within 500Mhz-800Mhz range. If indeed number of processors (shaders) has been increases so dramatically, to stay within reasonable power limits, voltage is probably lowered, and lower clock speed. And although 384 bit memory interface is unlikely but feasable, yields and economics make 2 chip solution impractical.

---

Nice, well said.

Does it bother anyone that they pulled the link down and erased the last thread? Hope so.

I still think that its not impossible for the 8800 core to be scaleble up to 1500mhz. If ya have noticed, many overclockers have gotten the g71 up to almost 1ghz with fancy cooling, and if the 8800 is a entirely new core i dont doubt at all that we will in time se cards running @ stock way over 1Ghz.
Sure they will have to cool the card down somehow but air is not the only way.
I dont understand why you think in such negative way. Hope people, hope. Its what i wanna say.

Quote:

---

Originally Posted by Hicks

hehehe, excellent post. Pretty much says it all really, the G80 specs are rubbish.

---

Actually...

Quote:

---

25M GF2, 57M GF3, 63M GF4, 125M FX, 222M 6800, 300M

---

Notice, transistor count doubled pretty much for each big leap in DX.

DX 7 to 8 doubled.
DX 8 to 9 doubled.

DX9 to 10 doubling isn't out of the question.

Also, every source has been saying to expect these heat/power issues, so that's 100% to be expected.

As for MHZ, they say the architecture is scalable to that point, they don't say actual clockspeeds for the different parts. Companies make the "scalable" statement all the time, generally it's much higher than they'll ever actually take it.

Pentium 4 is scalable to 10 Ghz in 2009....oh wait.

Quote:

---

Originally Posted by Sanborn

Pentium 4 is scalable to 10 Ghz in 2009....oh wait.

---

Maybe on 0nm process.

I kid, I kid.

Quote:

---

Originally Posted by LOE

funny thou, AMD and INTEL manage to stay in reasonable thermal envelopes while increasing processing power

how come ATI and NVIDIA fail to?

---

AMD/Intel have their own fabs for one, for two transistor count(conroe is 291 million transistors), for three die size...

GPU's have gotten quite a bit more complex than CPU's.

DilTech, the 750M transistors is really really pushing credibility. Unless much is dedicated to cache, looking at least 400-450mm^2. That is HUGE, and very very expensive, and the larger the lower the yields. A chip of that enormous size would have huge power and communication requirements (lots of relays). If nVidia simply extended existing G71 design, doubling of the pixel shaders (for example), would NOT double the transistor count/core size. In fact, it would be quite a modest increase, because majority of transistors are control logic, 2D, purevideo, as well as other 3D stages etc..
I think 400-450M transistors is a very reasonable estimate.

Quote:

---

Originally Posted by Pillo-kun

I still think that its not impossible for the 8800 core to be scaleble up to 1500mhz. If ya have noticed, many overclockers have gotten the g71 up to almost 1ghz with fancy cooling, and if the 8800 is a entirely new core i dont doubt at all that we will in time se cards running @ stock way over 1Ghz.
Sure they will have to cool the card down somehow but air is not the only way.
I dont understand why you think in such negative way. Hope people, hope. Its what i wanna say.

---

NV40 (6800) was successor of FX series. Yet clockspeeds went down from 475Mhz to 400Mhz. G70 had considerable redesign of 6800, yet clockspeeds stayed about the same (400Mhz vs 430Mhz). Why do you ASSUME clockspeeds will increase because its a new design? Especially, when they are still using the same 80/90nm manufacturing process as current generation. The circuitry still has to perform the same calculations as before (ex HDR), and is made on same/similar scale. The only factors I can imagine that would enable any drastic clock speed increase would be use of exotic materials like Silicon Germanium with dual stress liners and virtually every possible performance enhancing technique available, and/or cryogenically cooled using phase-change/Ln2.

Remember, above all else, nVidia is a publically traded company. Their interest is NOT to produce a few research samples that do the barely imaginable. They MASS PRODUCE product for CONSUMERS and expect to maintain PROFIT MARGINS. So the design MUST be practical and economical to manufacture.

My best estimate is the clockspeed will be 500-600Mhz (using 1.2V or lower), in order to curb power/heat. Perhaps when 65nm is available we will see something akin to G70->G71 with nice ~30-40% increase in clockspeed and die shrink to economical <300mm^2 range.

I too think 700 million transistors is pushing it...but...

As you just said, 400 to 450 sounds reasonable for the extra power apparently added, correct?

Now, anyone remember the statement that 1/3rd of the die will be dedicated to DX10 functionality? As in, it's going to take 1/3rd of the die to incorporate the new features of DX10....

So, say it'd be 400-450 with just the added shaders and without this added DX 10 functionality(As 450 WITH that added bit would result in a card equal to the G71), now add the second 128bit memory controller(putting us up to about 500 million), now add on an additional third of the amount of transistors and we're now at 665 million transistors which could be stated as ~700 million.

It's all possible, unlikely, but the numbers DO make sense.

Only reason I'm trying to make sense of all this is simple.... When's the last time VR-Zone was forced to pull down a story that wasn't true? I know NVidia made em pull down the story about the G71 being 90nm they posted a few days after the G70 came out, and ATi made em pull the early X1950XTX review, both were legit. Never though have I seen em have to pull a fake article.

Yeah thats the only way I could have really seen them incorporate those extra transistors.

It may be far-fetched, but it would make sense.

Quote:

---

Originally Posted by DilTech

I too think 700 million transistors is pushing it...but...

As you just said, 400 to 450 sounds reasonable for the extra power apparently added, correct?

Now, anyone remember the statement that 1/3rd of the die will be dedicated to DX10 functionality? As in, it's going to take 1/3rd of the die to incorporate the new features of DX10....

So, say it'd be 400-450 with just the added shaders and without this added DX 10 functionality(As 450 WITH that added bit would result in a card equal to the G71), now add the second 128bit memory controller(putting us up to about 500 million), now add on an additional third of the amount of transistors and we're now at 665 million transistors which could be stated as ~700 million.

It's all possible, unlikely, but the numbers DO make sense.

Only reason I'm trying to make sense of all this is simple.... When's the last time VR-Zone was forced to pull down a story that wasn't true? I know NVidia made em pull down the story about the G71 being 90nm they posted a few days after the G70 came out, and ATi made em pull the early X1950XTX review, both were legit. Never though have I seen em have to pull a fake article.

---

:yepp: :thumbsup:

http://theinquirer.net/default.aspx?article=34495

OMG, if Fuad doubts the info, then it's mos def BS.

Quote:

---

Originally Posted by TheINQ

INQ GRAPHICS Ed, Fuad Abazovic, writes that he doubts the data, given the die size and process technology involved.

---

Come on guys, it was a bad leak with inflated specs. Stop spinnin the issue.