Nvidia's Next Gen Speculation

Quote:

---

Originally Posted by spursindonesia

Agreed, considering that a 192 SP 24 ROP part GF 106 is already 240 mm^2 in diesize, only fanbois would believe that GF 104 is just 322 mm^2 part, while 367 mm^2 of diesize is MUCH more logical and believable.

OTOH, a 576 48 ROP 384 bit part developed from GF 104 basis is also quite possible even using 367 mm^2 as diesize basis, i think the added diesize would be in the neighborhood of 150-160 mm^2. With the added time and experience, pushing that big diesize out of TSMC 40 nm is still doable IMHO, and while the performance might only just been able to match Cayman @380 mm^2, the WOW and pride factor must be considered for them so they won't lose anymore fanbois and lovers around the globe. Remember, they have once designed a 576 mm^2 monster of GT 200, so it has happened in the past and can verywell happen yet again in the future.

---

What NV needs to do is increase their transistor efficiency up and get it size down. 1.95 billion transistors should take up less room up than cypress' 2.15 billion transistor and the fact that it is either matching that size or taking up more room show that NV needs to get this process down better.

I think it is too early to say that a 576 shader NV part would only match cayman not knowing the performance of both. Considering nothing is confirmed at this point, it impossible to say anything about the performance relation of two unreleased part with none of the specs really confirmed.

Quote:

---

Originally Posted by Manicdan

what you want is a laptop that has an external x16 pcie slot on the docking station with its own PSU, so you dock your laptop and it uses whatever video card you want.

is it too much to ask, no
so what hasnt it happened yet?

---

ATi Hydra.

Its in the works but the people behind it are slow like those red-assed monkeys.

Quote:

---

Originally Posted by tajoh111

What NV needs to do is increase their transistor efficiency up and get it size down. 1.95 billion transistors should take up less room up than cypress' 2.15 billion transistor and the fact that it is either matching that size or taking up more room show that NV needs to get this process down better.

I think it is too early to say that a 576 shader NV part would only match cayman not knowing the performance of both. Considering nothing is confirmed at this point, it impossible to say anything about the performance relation of two unreleased part with none of the specs really confirmed.

---

Part of the low power efficiency of GF100 is the unused GPGPU transistors which is axed in GF104 / GF106.

Problem is Nvidia pushes really hard on new parts and usually go too far and really need the process to be able to even pull their goals off. They are threading the needle rather than really refining the architecture over a few generations like ATI.

I think they really need to rethink the way they are doing things as ATI is taking giant bites out of their asses right now.

Quote:

---

Originally Posted by LordEC911

Not going to be over 600mm2, GT200 was right at the limit of what is possible.

---

im not so sure about that. reticles are really big and the die size limit may be imposed by something else entirely. making a chip that big is very challenging so it would be wise for nvidia to stay away from something that large. even at 530mm2 they are having troubles with manufacturability. they certainly dont have the time to make something that large either.

Quote:

---

Originally Posted by tajoh111

What NV needs to do is increase their transistor efficiency up and get it size down. 1.95 billion transistors should take up less room up than cypress' 2.15 billion transistor and the fact that it is either matching that size or taking up more room show that NV needs to get this process down better.

---

what do you mean by transistor efficiency? transistors are not really an accurate measure of area. cells are a little better though.

Quote:

---

I think it is too early to say that a 576 shader NV part would only match cayman not knowing the performance of both. Considering nothing is confirmed at this point, it impossible to say anything about the performance relation of two unreleased part with none of the specs really confirmed.

---

the whole spirit of speculation is to take random guesses and then find out how wrong we were.

Quote:

---

Originally Posted by STEvil

ATi Hydra.

Its in the works but the people behind it are slow like those red-assed monkeys.

---

alot of things are kinda like or tried to already do it. we saw a photo not too long ago of a massive external video card (i think an msi ferrari laptop or something?)

technically a dedicated gpu + integrated one does it already, but the real limit i think has to do with how many PCIe lanes they put onto laptop northbridges, much less than desktops to save some power. we want 20ish lanes to do everything right, but a typical laptop has like 8? i wonder if its possible to have a separate northbridge on the docking station so the laptop dosnt need to waste the energy on things it wont use while just on battery.

Quote:

---

Originally Posted by spursindonesia

Agreed, considering that a 192 SP 24 ROP part GF 106 is already 240 mm^2 in diesize, only fanbois would believe that GF 104 is just 322 mm^2 part, while 367 mm^2 of diesize is MUCH more logical and believable.

---

Yeah, 367 has to be minimum. Even with 380mm2 G104 has better density than GF106. Another possibilty is that more is disabled in GF106 than a 64-bit memory controller, but probably not.

Quote:

---

Originally Posted by spursindonesia

Remember, they have once designed a 576 mm^2 monster of GT 200, so it has happened in the past and can verywell happen yet again in the future.

---

I don't think that's likely to happen in the near future though. Even GF106 has disabled parts, and GF100 is already at (or close to) a power wall. I'm sure they can still push the boundaries a little bit with their current tech, but not enough to catch NI unless NI itself is a very minor improvement.

With any luck, Nvidia can pull some kind of RV770 where they shrink the shaders, improve their density, and are able to fit a lot more slightly less powerfull shaders in the same area.

Otherwise, I'm thinking that Nvidia will need to make seperate gaming GPU and GPGPU architectures to be competitive. GF104 vs. GF100 seem to be going this route already in a minimal way.

Quote:

---

Originally Posted by Frontl1ne

Real men use desktops for gaming.

---

lol been there done that. Desktops are a waste of time, cause real men play good games, and good games don't require crazy hardware. Real men don't even care about graphics, and run a game at absolute lowest settings to get max FPS (also been there done that, ie. Q2/3 days).

Basically, if you wanna talk reality, fact is that the laptop is a far superior design to the ancient (and prolly soon to be obsolete) desktop. More and more ppl are buying laptops, while desktops are going the way of the VCR.

Sure, there are tasks still suited to the desktop, just because some ppl (read: NOT gamers) actually need all the power they can get to increase productivity. But do most of us need all that power? Short answer, no. And again, been there, done that. I know all too well the feeling of satisfaction of knowing you've built and OC'ed a nice system, and then ending up surfing the net or playing some 5 year old game.

I'm prolly goin laptop next system...it has the conveniance, it has the right amount of power I need, and I'll have even more satisfaction knowing that I'm gonna be using what I got, and not just trying to boost my e-peen.

Quote:

---

Originally Posted by Baron_Davis

Basically, if you wanna talk reality, fact is that the laptop is a far superior design to the ancient (and prolly soon to be obsolete) desktop. More and more ppl are buying laptops, while desktops are going the way of the VCR.

---

What?

Cool story bro. Keep fingering your pad.

Quote:

---

Originally Posted by Jowy Atreides

What?

Cool story bro. Keep fingering your pad.

---

Awwww is your butt hurt??

Hey, you ever heard of a mouse or did you just now manage to crawl out of the hole you've been living in?

News flash to other members of club "Jowy": the trend towards miniaturization and mobility is here and it's real. Keep worshipping your PC's all you want, I'm just stating the obvious.




Anyway, my initial comment was about making GPU's smaller and more efficient. Does that bother anyone else? Are you guys 480 tri SLI owners as well? Jen-Hsun would be proud...

Quote:

---

Originally Posted by Baron_Davis

Awwww is your butt hurt??

Hey, you ever heard of a mouse or did you just now manage to crawl out of the hole you've been living with?

---

Yeah. Mice are the things best used on a good mouse mat.
Something you don't take with a laptop.

Are you butthurt to try and assert logically backward opinion?

Tablets > Laptops.

Quote:

---

Originally Posted by Jowy Atreides

Yeah. Mice are the things best used on a good mouse mat.
Something you don't take with a laptop.

Are you butthurt to try and assert logically backward opinion?

Tablets > Laptops.

---

LOL why do you assume I wanna get a laptop just so I can "work" away from home? I want a laptop cause it's small, simple, and I can use it in different places to do different things. Basically, it's powerful and conveniant. Why don't you ask your friends and family what they value more in a product? Power or conveniance. I bet even YOU are smart enough to know the answer. Guess what happens when you offer someone both? Guess what's gonna inevitably happen in the very near future due to hardware miniaturization? That's right...desktop power will meet laptop and smaller sizes. I predict smartphones and tablets will be as powerful as today's laptops in the next few years. Guess who's gonna want and more importantly NEED a desktop in a decade?

0.0000001% of people

And with cloud computing emerging as a mainstream technology? LOL GGPO PC

I ain't no Nostradamus, but I don't need to be a prophet to see the future of computing...it's pretty damn obvious. Just pick up a Bestbuy flyer and look at what's being advertised...simple way of seeing the direction technology is headed.

how about both of you stop it?

Quote:

---

Originally Posted by Baron_Davis

Awwww is your butt hurt??

Hey, you ever heard of a mouse or did you just now manage to crawl out of the hole you've been living in?

News flash to other members of club "Jowy": the trend towards miniaturization and mobility is here and it's real. Keep worshipping your PC's all you want, I'm just stating the obvious.




Anyway, my initial comment was about making GPU's smaller and more efficient. Does that bother anyone else? Are you guys 480 tri SLI owners as well? Jen-Hsun would be proud...

---

More or less what Fusion Llano is. That's what you're looking.

Quote:

---

Originally Posted by Baron_Davis

.....

---

Reason why my C&C 3 lags is because of freak people that are still using laptops for gaming. EA server always take the slowest system to synchronize.

I can't see myself without a monitor to play my games, minimum 24 inch resolution 1920x1200.

Laptops are mainly for working because they are too weak for gaming and the resolution does not attract me. If you have the money to buy a good gaming laptop and does not care about resolution then is clearly the best choice.

Quote:

---

Originally Posted by Nintendork

More or less what Fusion Llano is. That's what you're looking.

---

I need the power, too, though. I don't think AMD's integrated GPU will be good enough for what I want. I'm waitin for mobile HD6000 series for that.

i7 + HD6000 mobile + Intel G3 SSD's in RAID0 + lotta RAM = perfect laptop

for me at least...

Quote:

---

Originally Posted by Baron_Davis

I need the power, too, though. I don't think AMD's integrated GPU will be good enough for what I want. I'm waitin for mobile HD6000 series for that.

i7 + HD6000 mobile + Intel G3 SSD's in RAID0 + lotta RAM = perfect laptop

for me at least...

---

Dude it's not going to happen. They are still on 40nm, the HD6k's won't suddenly use less power than the HD5k's and be faster. If you don't mind using it plugged in the 480M is fast, but if you want mobile parts that can compare to desktop parts it won't happen as long as the chip makers have a 300watt window of opportunity.

desktop will never die

Quote:

---

Originally Posted by Iconyu

Dude it's not going to happen. They are still on 40nm, the HD6k's won't suddenly use less power than the HD5k's and be faster.

---

It's theoretically possible on the same node. Just look at HD5850 vs. GTX 260 (faster and uses less power on the same node). However, I think what we can look forward to is higher performance AND higher power consumption (albiet with higher performance/watt)

there is also the issue of diminishing returns, you can't keep making improvements based on the same magnitude time and time again. this is a law that applies to all.

Quote:

---

Originally Posted by qcmadness

Part of the low power efficiency of GF100 is the unused GPGPU transistors which is axed in GF104 / GF106.

---

What? What are you talking about... you mean just because it doesn't have ECC, all of a sudden it's way more powerful? Sorry, not accurate. The things the smaller chips don't have do not affect power much.

Recent nvidia products are not energy efficient(since G92 vs R670, G80 & R600 were both hungry monsters). HD5000 is the best perf/watt we can have now on the 40nm process.

I see mobility HD6000 just as a plain rebrand. That or just release them with 28nm shrink because mobility HD5000 is more unaffected by competion than their desktop bros.

Example:
HD5730 (5650 + GDDR5 instead of GDDR3 and core clocked higher) --> HD6650

Quote:

---

Originally Posted by tajoh111

What NV needs to do is increase their transistor efficiency up and get it size down. 1.95 billion transistors should take up less room up than cypress' 2.15 billion transistor and the fact that it is either matching that size or taking up more room show that NV needs to get this process down better.

I think it is too early to say that a 576 shader NV part would only match cayman not knowing the performance of both. Considering nothing is confirmed at this point, it impossible to say anything about the performance relation of two unreleased part with none of the specs really confirmed.

---

Yeah i concede that my post is mostly based on speculations, and Cayman prediction might be a bit further on the optimist side, but i digress. Atleast with Cayman we will see it in reality soon enough while everything on the green side is based on total speculation, so i don't think i'm going overboard with my previous post. Nothing is really concrete at the moment so everything is a guesswork. But my diesize comment stands on its own right.

Regarding design efficiency, GF 104 is already stripped of many GF 100 GPGPU emphasis, so i don't think nVidia has got anymore room to maneouvre in this current process node, they have to wait until 28 nm process round arrives. To make this new giant chip possible & feasible (ready in Q1 2011), they have to make it ASAP, without much redesigning & developing, just blow GF 104 specs 50% up and diesize 40% up, bingo !

Just make sure you can manufacture the full chip right from the start at decent yield, otherwise ..... I want good competition from both side of fences and if it takes a ~520 mm^2 250 w chip from the greenside to match/beat Cayman, so be it.

Quote:

---

Originally Posted by hurleybird

Yeah, 367 has to be minimum. Even with 380mm2 G104 has better density than GF106. Another possibilty is that more is disabled in GF106 than a 64-bit memory controller, but probably not.

I don't think that's likely to happen in the near future though. Even GF106 has disabled parts, and GF100 is already at (or close to) a power wall. I'm sure they can still push the boundaries a little bit with their current tech, but not enough to catch NI unless NI itself is a very minor improvement.

With any luck, Nvidia can pull some kind of RV770 where they shrink the shaders, improve their density, and are able to fit a lot more slightly less powerfull shaders in the same area.

Otherwise, I'm thinking that Nvidia will need to make seperate gaming GPU and GPGPU architectures to be competitive. GF104 vs. GF100 seem to be going this route already in a minimal way.

---

Bro, while nVidia management currently acts like idiots, i still believe there are plenty of good engineers in that house, so while RV 770 green edition is quite unlikely in this same 40 nm node, a decent yielding 150% version of GF 104 is IMHO still within reach for these guys to develop, sooner than later. Yeah, the chip might be GF 100 sized, but efficiency (perf/watt & perf/area) IMHO will increase somewhat, and the chip will be manufacturable.

nvidia's engineers are good .. no doubt about that .. but its the idiots or should i say nv'idiots in marketing and the likes that screw it up ... and force the engineer to take a route that will likely cause them trouble .. just for the sake of the pr effect ....

so wait, this thread is based on... nothing?
this is not how it works, you dont start a topic about an unreleased part without any infos out whatsoever...
there arent even rumors or speculations about it...

zedX, right, there are 490 cards boxed up and ready to ship that will now get burned... puh-lease! :rolleyes:
and those rumors on mauro paketti dont make much sense either...

where does nvidia go from here?
all they CAN do on 40nm is make a bigger gf10x like part or drop prices and flood the market, focusing on quantity over asp.

a bigger part would only work if yields and capacity of tsmc wasnt a problem, and there are rumors that both of them still ARE a problem.
that doesnt mean they wont do it, but i dont think we will see it soon... my guess is roughly gf100 size but based on gf10x, so itll have more sps
if we take gf10x and scale it to gf100 size we get around 600sps and tdp wouldnt be a problem either... the only thing that limits nvidia on 40nm with gf10x is die size, no longer tdp...
but a 600sp gf10x part would only be around 15% faster than a gtx480... probably enough to beat a 6870 but, but then there will be a 6970... so nvidia really really needs to get their perf per transistor up...

and about the second option, well guess what, thats exactly what nvidia is doing right now... :D