NVIDIA Announces GeForce GTX 480M Mobile GPU

[Sam_oslo]

TDP is not determined by reviewers. And Nvidia proclaims so many things these days...

Of course all stars with some claims, but it is much better to stick to manufacturers claims than the genius who claimed inside info, and was spreading lies about an unreleased product with poor performance and broken etc ..

nViodi's claims is the best source we have for now, but we have to wait for hand on test and review to determine anything for sure about performance, power usage and heat.

[ChinStrap]

Come 485 September we'll find out

word

[Oberon]

How knows, maybe they need time to fix the drivers, BIOS, yield, heat, power usage, or other stuff. But, and it is a big butt . maybe nVidia doesn't feel the need for 512, at least as long as GTX 480 beats the 5870.
In my opinion, nVidia needs a double GPU to compete with 5970, not a 512 single GPU.

Why should they drop the 512 now, to compete with what you think?

Right, NVIDIA didn't release a 512SP GF100 because they "didn't need to". They just spent the time and money to find defective clusters and fuse them off for the hell of it.

[ElSel10]

Right, NVIDIA didn't release a 512SP GF100 because they "didn't need to". They just spent the time and money to find defective clusters and fuse them off for the hell of it.

They don't "search" for defective SMs, they just disable them, as it improves yields.

[zanzabar]

on the gddr, the specs say 256bit buss, 1200mhz and 76.8GB/s. that means that with [(core speed)*(buss width)*(data rate)/8]= total GB/s

so that will give u with some math skills, 38.4GB/s*(data rate)= 76.8GB/s, that clearly leaves a data rate of 2 making it something that operates in ddr and dose 1200mhz so that is gddr3 as if it was gddr2 that would be slower and gddr5 is qdr.


there is also no spec on die size just stats so im assuming that its like the 465 making at a castrated gf100 and not a mini gf100 as the shader speed is 850mhz (aka processor speed, makes sense for the texture fill rate also) and NV rates tdp as the entire card that would not make sense for a smaller part to be rated at 100W with that clock and the 352 shader count. that demonstrates no flexibility and they need to hurry up with a real part that is useful for more than benching. with the limited 225W wall for desktop they need a smaller die that clocks higher as that would bring yields and performance up with cost down and the laptop needs a die even smaller than that that also has higher clocks that will give it more p-states for battery life and more performance.

[Sam_oslo]

Right, NVIDIA didn't release a 512SP GF100 because they "didn't need to". They just spent the time and money to find defective clusters and fuse them off for the hell of it.

They need a double GPU to beat the 5970, not a better single GPU, because GTX 480 beats 5870 with good enough margin already, in my opinion.
As I've said already, who knows why, but you seams to know it all. If so tell me, nVidia should release a 512 to compete with what?

[Florinmocanu]

Sam, there is no question or discussion about this, if Nvidia COULD release a 512 sps (better yields/lower TDP), they would. The fact that they cannot tells a lot about the current state of the GF100 chip.

Any other discussion is pointless.

[Sam_oslo]

Sam, there is no question or discussion about this, if Nvidia COULD release a 512 sps (better yields/lower TDP), they would. The fact that they cannot tells a lot about the current state of the GF100 chip.

Any other discussion is pointless.

No constructive and polite discussion is pointless!, unless one doesn't want/can't catch the point.

As said, how knows why, but the reality and logic says that they don't/didn't need a 512 to beat the 5870. Do you mean nVidia have to release a 512 to compete with it's own GTX 480, or what?

[Chumbucket843]

Sam, there is no question or discussion about this, if Nvidia COULD release a 512 sps (better yields/lower TDP), they would. The fact that they cannot tells a lot about the current state of the GF100 chip.

Any other discussion is pointless.

its not that simple of a problem. gf100 has a lot of leakage power. it might be enough to the point where having less shaders with higher clocks is more power efficient.

[prava]

It doesn't need to be hugely better, a little better would do!

Of course it is based on Fermi-architecture, what else would you expecting? but what would qualify as a a new chip in your book? Do you want a new architecture, gf15000 to call it new chip? This is not a desktop chip, it is a new Mobile chip.

What we are talking about is that this chip will not be a new one and, as such, it will be a castrated GF100. Not all the other mobile chips are castrated, as it doesn't make any kind of sense to use a ~500mm2 to make mobile chips (its a waste). My bet is that this is temporary, until they have GF104 ready for action.

[Boissez]

No constructive and polite discussion is pointless!, unless one doesn't want/can't catch the point.

As said, how knows why, but the reality and logic says that they don't/didn't need a 512 to beat the 5870. Do you mean nVidia have to release a 512 to compete with it's own GTX 480, or what?

Nvidia has sunk a lot of money and effort in developping a 512 chip. Not delivering it is a sign of bad planning/execution/design. Believe me, if they could, they would.
Following your logic Nvidia shouldn't have released the G92B 8800GTS back in 2007 (it's not like ATI was a threat at the time) - but they could, so they did.

[Florinmocanu]

its not that simple of a problem. gf100 has a lot of leakage power. it might be enough to the point where having less shaders with higher clocks is more power efficient.

they designed a 512 sp chip, not a 480 sp chip. They cannot deliver that chip under the power limitations we currently have. I wish they could but it's obvious they cannot do it. Not under 40nm or with a massive redesign.

[zanzabar]

they designed a 512 sp chip, not a 480 sp chip. They cannot deliver that chip under the power limitations we currently have. I wish they could but it's obvious they cannot do it. Not under 40nm or with a massive redesign.

if they designed a chip that cant stay in its power envelope then they should have made it smaller and clocked were the bigger one should have been. or they should at least put one out for benching only as thats all that the gf100 is good for right now other than crunching. im not saying that it isnt fun to play games with benching hardware or clocks but its not viable for mass marketing or non custom computers.

[ElSel10]

Not under 40nm or with a massive redesign.

And how would you know that?

[Sam_oslo]

What we are talking about is that this chip will not be a new one and, as such, it will be a castrated GF100. Not all the other mobile chips are castrated, as it doesn't make any kind of sense to use a ~500mm2 to make mobile chips (its a waste). My bet is that this is temporary, until they have GF104 ready for action.

Nobody puts a full blown new desktop GPU in a laptop. But it doesn't mean they are "castrated ", they are Mobile chips that are cut down to save power and heat. They are new chips that are spouse to be like that.

Show me a full blown new desktop GPU that went right into a Mobile GPU, then you are right. Otherwise you don't know what you are talking about,

Nvidia has sunk a lot of money and effort in developping a 512 chip. Not delivering it is a sign of bad planning/execution/design. Believe me, if they could, they would.
Following your logic Nvidia shouldn't have released the G92B 8800GTS back in 2007 (it's not like ATI was a threat at the time) - but they could, so they did.

You are amusing too much. As said, who knows, but maybe they were working on 512 because they thought they would need that much to beat the 5870 at that time. The reality shows that 480 was (and still is) enough for that. Why should they beat 5870 by 30% when 10% gives you the lead? But who knows, maybe they got in trouble too, but that's not important now, what matters now is, the GTX 480 beats the 5870 with good enough margin.

[Florinmocanu]

And how would you know that?

AMD/ATI doubled the vias at some point to ensure better yields. Such a re-design would improve yields on 40nm for GF100.

[Chumbucket843]

Nvidia has sunk a lot of money and effort in developping a 512 chip. Not delivering it is a sign of bad planning/execution/design. Believe me, if they could, they would.
Following your logic Nvidia shouldn't have released the G92B 8800GTS back in 2007 (it's not like ATI was a threat at the time) - but they could, so they did.

sh*t happens. there are some things that are out of nvidia's control.

they designed a 512 sp chip, not a 480 sp chip. They cannot deliver that chip under the power limitations we currently have. I wish they could but it's obvious they cannot do it. Not under 40nm or with a massive redesign.

they could but it would not have very good clockspeeds. im not arguing over what gf100 could have been. my point is that they are doing the most with what they have by disabling an SM and bumping clocks. it could be a better choice than having 512sp's.

[RaZz!]

It was actually pretty fast, after all Fermi is just a few months old. Why should we call it "castrated Fermi"? nobody puts a full blown desktop GPU in a notebook. Maybe a flexible and scalable Fermi-architecture is making this possible so fast?

i hear you, "castrated" was probably the wrong wording. i just meant a trimmed down fermi fitting the mobile needs.

however, even though people make fun of fermi's heat and power draw, these are really things that'll be interesting about 480m and i'm curious whether nvidia is able to reduce power draw and heat of the mobile parts that much or not.

[Sam_oslo]

i hear you, "castrated" was probably the wrong wording. i just meant a trimmed down fermi fitting the mobile needs.

however, even though people make fun of fermi's heat and power draw, these are really things that'll be interesting about 480m and i'm curious whether nvidia is able to reduce power draw and heat of the mobile parts that much or not.

Yeah and I agree. It will be really interesting to see some tests and reviews on the real performance, power usage and heat. Because it will show how scalable and flexible this new architecture really is.

[Boissez]

You are amusing too much. As said, who knows, but maybe they were working on 512 because they thought they would need that much to beat the 5870 at that time. The reality shows that 480 was (and still is) enough for that. Why should they beat 5870 by 30% when 10% gives you the lead? But who knows, maybe they got in trouble too, but that's not important now, what matters now is, the GTX 480 beats the 5870 with good enough margin.

You keep mentioning the GTX480 10% lead like it's the only performance metric that matters. I do not think so. So therefore I am not wooed by GTX295-like performance with 50% higher power consumption.

And from the looks of it this also applies to the GTX480M

[Sam_oslo]

You keep mentioning the GTX480 10% lead like it's the only performance metric that matters. I do not think so. So therefore I am not wooed by GTX295-like performance with 50% higher power consumption.

And this also applies for the GTX480M...

I personally mean GTX 480 is performing exactly based on expectations, with good enough margin above the competing GPU (the 5870). It is exactly where any of my sane meanings expected it to be.

But , yes, I mean the power usage and price should drop. This power usage may get better by time with maturing BIOS, drivers and such, and we have already seen some improvements in a relatively short time. Hopefully can get even better, and it should. The price will get better when ATi can get it's acts together and put up a good fight in single-GPU war.

[mindfury]

on the gddr, the specs say 256bit buss, 1200mhz and 76.8GB/s. that means that with [(core speed)*(buss width)*(data rate)/8]= total GB/s

so that will give u with some math skills, 38.4GB/s*(data rate)= 76.8GB/s, that clearly leaves a data rate of 2 making it something that operates in ddr and dose 1200mhz so that is gddr3 as if it was gddr2 that would be slower and gddr5 is qdr.

GTX480 spec:Memory Clock (MHz) 1848
http://www.nvidia.com/object/product...tx_480_us.html

So GTX480m mem clock=1200MhzX2

[Solus Corvus]

I personally mean GTX 480 is performing exactly based on expectations, with good enough margin above the competing GPU (the 5870). It is exactly where any of my sane meanings expected it to be.

That's good because your insane meanings put it at 50-60%.

http://www.xtremesystems.org/forums/...postcount=1526
http://www.xtremesystems.org/forums/...postcount=1528
http://www.xtremesystems.org/forums/...postcount=1530

Not to pick on you though - those old fermi threads are full of great tidbits. Ahh, the memories, lol.

[zanzabar]

GTX480 spec:Memory Clock (MHz) 1848
http://www.nvidia.com/object/product...tx_480_us.html

So GTX480m mem clock=1200MhzX2

i guess that u are right, so NV just cannot give out specs now, but why would they use 600mhz gddr5 i did not know that it came that slow or could be undervolted.

they should be using the real frequency or the effective frequency they should not be allowed to use a magical frequency that they make up. i see their vendors using only the effective clock though.


lets just look at the part, its a full gf100 die, the entire die gets powered so on a laptop u are stilling the full desktop part, then its clocked to 425mhz. what dose that leave for a laptop i dont know how low the 480 can clock but im guessing that it wont work with optimus so that would have a card thats got a 100W gpu running at something like 30W+the card all the time as i dont see it clocking below 150mhz. that would make a laptop that would have like 30min on battery when idling making it not much of a laptop. what they need is a smaller die with the same shader count maybe less but with much higher clocks as the architecture is supposed to scale in size and scale with higher clocks why is it taking so long to get something out.

[Caparroz]

(..) But it doesn't mean they are "castrated ", they are Mobile chips that are cut down (...)

Wow, that much logic blew me away...

(...)Why should they beat 5870 by 30% when 10% gives you the lead? (..)

I'm willing to bet that you don't work at the private sector.

You do know that gf100 was designed with working 512 cuda cores in mind, right?