The GT300/Fermi Thread - Part 2!

[Chumbucket843]

Its only limitation is probably heat , and we can always increase to 3 slot cooler i would like one .

1GHz 5870 would use ~250 watts if we interpolate what the 4890 was to the 4870. its the same reason netburst didnt work. ramping up clockspeed is terribly inefficient and very few chips can be binned that high with out a silicon revision. 5950 sounds much better imo.

[Chickenfeed]

Hmm if that is truely the memory frequency, that kind of sucks. 160GB/s assuming thats the reference ( stock ) clock. The 5870 does 160GB/s at 1250 on a 256bit bus. I hope it is indeed higher as it would be nice to see the cards extra VRAM running at higher speeds which will be of benifit at higher resolutions no doubt. I really don't see the 470 having worse bandwidth ( worse being the same which generation to genertion is kinda meh ) than the 285 though... Who knows mabey the 480 will only be 1200 (240GB/s)... ( I've estimated 1200-1300ish )

[***Deimos***]

1GHz 5870 would use ~250 watts if we interpolate what the 4890 was to the 4870. its the same reason netburst didnt work. ramping up clockspeed is terribly inefficient and very few chips can be binned that high with out a silicon revision. 5950 sounds much better imo.

Gotta read this - hilarious Intel VP

Yup.

That's why 8800GT (600Mhz) runs at 1.05V and 9800GTX (738Mhz) at 1.15V.
Both are much lower than Intel, and especially AMD CPU (ie 1.35V).

As you said, its combination of good avg clockspeed at min voltage in order to keep power in check. Beyond a certain clockspeed, agressive exponential increase in voltage is needed for stability. As you may imagine, power skyrockets ( ~dynamic powa = (V^2) * C * Mhz ).

Which brings up an interesting point - GTX480/Fermi overclocking... probably very little to no headroom.
1. Would need extra power connectors.
2. Heatsink cooling doesn't scale linearly with size. Once you already have dual slot heatpipe cooler, making it 50% bigger doesn't improve cooling 50% - diminishing returns.
3. Modern high end GPUs are already running 80-90C, with VRM/MOSFET temp of over 100C not uncommon. Fermi overclocking - Do you really want to set fire to fluffy your plush rabbit (sitting on your case)?

[kemo]

http://translate.google.com.br/trans...n&hl=&ie=UTF-8

Those pictures belong to OBR and he is a member here , i wonder if he can leak something

[mindfury]

Only 4 partners will launch Fermi (this might be for Europe only) and XFX won't be one.
Cards won't ship until 1 or 2 weeks after launch. though partner availability in May will be quite bigger.

NV Hopes to get the speeds from the B1 chip that everyone is expecting from the launch chips. So we're a couple of months away from seeing the true GF100 performance.

http://forum.beyond3d.com/showpost.p...postcount=3138

[takamishanoku]

So from everything i've seen; the 480 will be the fastest single card out there. The 470 will be more or less the same as the current 5870. 5970 will still be the fastest dual card. From early in this thread (and part 1) neliz suggested that these cards were meant to come out Christmas time last year and that is the problem. By this stage a refresh (think 1ghz 2gb ram) of the 5870 should make it compete or surpass a 480. But even with a refresh the 480 may have an edge in next gen games (but as not many are available not really useful for most?)

In any case as i've always stated this arc will be KILLER once they can sort all the probs out esp with a refresh/shrink. By that stage the next ati arc will be out. Interesting times.

By years end 5870 performance levels should not only be common but affordable. Not a bad thing i guess.

[spursindonesia]

I think what Neliz meant is, with A3 GF 100, we will see a very limited number GTX 480 which will be clocked high enough to be very competitive (read, kick HD 5870 butt seriously and following closely on HD 5970 back) but not supported with good yield, thus low availability. Then, when B1 arrive, the clock will be the same, but the availability will get much better, and if the rumored 480 SP GTX 480 is true, we could see a new 512 SP "Ultra" part should B1 yield on that spec permitted. Pardon my interpretation if its too far from his actual idea.

[annihilat0r]

600/1200mhz is what I hear for GTX 470.

470's clocks have been lowered for 480 to better distinguish itself from its little brother. And more OC headroom.

[saaya]

Yet the high end cards continue to boost graphics, and without them, laptops wouldn't have their derivatives. So it doesn't really matter if not tens of millions are sold. It's continuously pushing the industry. Without it, you'd still have Geforce 2 MX onboard in your laptop.

Ps. how offtopic can a post go lol. Plus this is XS, laptop talk is prohibited. j/k

well, the thing is that the two are spreading apart...
the difference between laptop graphics and desktop graphics becomes bigger and bigger... and the difference between standard graphics in laptops and real game graphics in laptops is growing bigger and bigger as well...

game devs focus on the lower end of the pyramid while the top of it is growing into a needle thin tower... i think the top of the pyramid will grow less and less and become thinner and thinner, while the base of the pyramid will grow wider and higher slowly. like with everything going mainstream, you start with a pyramid and end up with a huge fat base and a tiny thin highend needle top sticking out. right now the pyramid is collapsing into itself, thats why from our (highend) perspective things arent moving at all, and from an overall perspective things are moving slowly.

http://translate.google.com.br/trans...n&hl=&ie=UTF-8

why only 4ghz effective? 5870 is 1200... and 470 has less width than the 480 so why lower the clocks as well? hmmm

600/1200mhz is what I hear for GTX 470.

470's clocks have been lowered for 480 to better distinguish itself from its little brother. And more OC headroom.

that doesnt really make sense... they are behind the 5870, why reduce clocks? i think they did it to improve yields or reduce heat... not to make their 480 look better...

[[XC] Oj101]

This is the funniest Hitler reaction video I have EVER seen. Not necessarily because it is true but because it is very well pulled off

http://www.youtube.com/watch?v=VpZXhR1ibj8

Does anyone know when the Fermi NDA expires? I need figures so I can decide which boat to jump on.

[KingOfsorroW]

This is the funniest Hitler reaction video I have EVER seen. Not necessarily because it is true but because it is very well pulled off

http://www.youtube.com/watch?v=VpZXhR1ibj8

Does anyone know when the Fermi NDA expires? I need figures so I can decide which boat to jump on.

oh man, looks like you didn't bother to read anything... The video was posted before and the 26 of march is the release date.

[[XC] Oj101]

No, I didn't. I have avoided this thread as I'm waiting for Part 2 with actual numbers instead of rumour after rumour. I'm sorry I didn't read through 94 pages and follow every Youtube link before making my post here. It was a really funny video and I wanted to share it. Shall I delete my post?

[japamd]

I think what Neliz meant is, with A3 GF 100, we will see a very limited number GTX 480 which will be clocked high enough to be very competitive (read, kick HD 5870 butt seriously and following closely on HD 5970 back) but not supported with good yield, thus low availability. Then, when B1 arrive, the clock will be the same, but the availability will get much better, and if the rumored 480 SP GTX 480 is true, we could see a new 512 SP "Ultra" part should B1 yield on that spec permitted. Pardon my interpretation if its too far from his actual idea.

It does make sense to me, but I think a rumored 512 SP part should have higher clocks to distinguish it better from 480 SP card, in both price and performance.

[Behemot]

Those pictures belong to OBR and he is a member here , i wonder if he can leak something

Yeah this guy every time gets 5 cards somewhere and than sells them 2 weeks after release

[Dimitriman]

1GHz 5870 would use ~250 watts if we interpolate what the 4890 was to the 4870. its the same reason netburst didnt work. ramping up clockspeed is terribly inefficient and very few chips can be binned that high with out a silicon revision. 5950 sounds much better imo.

You're assuming 0 improvements in manufacturing process from launch to date.

[trinibwoy]

why only 4ghz effective? 5870 is 1200... and 470 has less width than the 480 so why lower the clocks as well? hmmm

Do you know which clocks the 470 needs in order to compete with the 5870?

[LiquidReactor]

New architecture = all bets are off

what efficiency are you talking about there are graphs from amd showing dramatic increase with 2 gb ram

I think he's referring to a test someone posted here in the 4800 vs GTX 200 series days showing that in the same games with the same settings ATI used less video memory.

AKA more efficient.

Whether that's still the case today or not I don't know.

Yup, the test was run with Far Cry 2. ATI was very close to using 2x less memory that GTX 200 cards used in that game. So I seriously doubt ATI cards somehow became less efficient with vram usage

[Dimitriman]

At the same time I remember similar tests showing that Doom 3 consumed fairly easily 500MB of video memory when run at 1600x1200 resolution.

So again, all depends.

[zalbard]

Hmm if that is truely the memory frequency, that kind of sucks. 160GB/s assuming thats the reference ( stock ) clock. The 5870 does 160GB/s at 1250 on a 256bit bus. I hope it is indeed higher as it would be nice to see the cards extra VRAM running at higher speeds which will be of benifit at higher resolutions no doubt.

It gotta be higher, I don't see them using really slow GDDR5 for some reason... Doesn't make sense.

[skycrane]

oh man, looks like you didn't bother to read anything... The video was posted before and the 26 of march is the release date.

your kidding right ?????? there are 93 pages of BS, 1 page of actual information. and you want me to read through all of that NVIDA sucks! ati sucks!

this is the second thread dedicated to this, they had to close the first one because of all the BS getting posted. and im a bit suprised this thread has not ben closed yet.

[-Sweeper_]

[Behemot]

This was already posted. Oh, the Catalytic driver looks nice

[prava]

what efficiency are you talking about there are graphs from amd showing dramatic increase with 2 gb ram

Do you mean this one? If it is, please look at it again and tell me where are the improvements when using 2GB vs 1GB you are talking about.

That's why 8800GT (600Mhz) runs at 1.05V and 9800GTX (738Mhz) at 1.15V.
Both are much lower than Intel, and especially AMD CPU (ie 1.35V).

As you said, its combination of good avg clockspeed at min voltage in order to keep power in check. Beyond a certain clockspeed, agressive exponential increase in voltage is needed for stability. As you may imagine, power skyrockets ( ~dynamic powa = (V^2) * C * Mhz ).

Which brings up an interesting point - GTX480/Fermi overclocking... probably very little to no headroom.
1. Would need extra power connectors.
2. Heatsink cooling doesn't scale linearly with size. Once you already have dual slot heatpipe cooler, making it 50% bigger doesn't improve cooling 50% - diminishing returns.
3. Modern high end GPUs are already running 80-90C, with VRM/MOSFET temp of over 100C not uncommon. Fermi overclocking - Do you really want to set fire to fluffy your plush rabbit (sitting on your case)?

You are mixing things, mate. Voltage alone has nothing to do with power consumption nor heat nor anything. Just because something works at 100v doesn't mean its warmer than another thing that works at 20v

Then, answering to your points...
1) No, you don't need any extra power connectors. The max amperage rating per connector is just a standard stuff, not that they will cut off when you are hitting that limit. If it was like you say nobody would be able to overclock a GTX295 (6+8), 5870 (6+8) and so on...

2) HS in gpu aren't chosen because of they efficiency only, but because their efficiency + cost + availability, we are talking business here, $5 saved on each cooler means a lot of money when you sell millions of them.

3) Again, you are misunderstanding many, many things. First, the bigger the chip is, the easier to cool it, which means that for cooling matter a big ass GF100 is good. Then, the VRM and mosfet section cooling depends entirely on the design, you don't know exactly how the new cooler works so we may have awesome mosfet and VRM temperatures. About overclocking, well, the same, GTX295 has a big ass TDP and hasn't have any problem about being overclocked, why? Because it has a very good HS design...

[shoopdawoopa]

I hate being red green color blind! Why would they use two shades of one color for a graph. Which is which?

[Carfax]

I hate being red green color blind! Why would they use two shades of one color for a graph. Which is which?

The dull red shade is for 2GB and the bright red shade is for 1GB.

The chart is misleading all the same, because it's using 3 1920x1080 monitors in Eyefinity.

At normal resolutions, the 2GB part won't offer nowhere near as much of a performance increase..