NVIDIA GeForce 4XX Series Discussion

[Bo_Fox]

First of all, the Tesla is rated below 225w. Now this is not about enthusiasts, Tesla is not for enthusiasts, it's for HPC a very different market where reliability and convenience matters much more than sheer power. In this particular case a TDP lower than (or around) 225w is almost a must, because most of the Tesla cards are probably going to replace 2 CPUs* in old servers, while almost the rest of the rig is going to be kept (especially the PSU and cooling system). By being limited to 225w a single Tesla can replace (the overall power consumption of) 2 typical server CPUs. For example, anyone can replace racks with an 8p server by a 4p + 2 Teslas, 2p +3 Teslas... whatever their requirements are, but they would not have to change anything (except the obvious). If power requirements were higher, they would start having problems finding a way to replace what they had for the cards and probably have to replace the entire powering + cooling system. Some Xeons and Opterons took quite long to be released at the same clocks as their desktop equivalents because of the same reason. Only when a revision made the TDP lower they were released.

You just can't compare the enthusiast market with the enterprise or HPC market (that's why Charlie is such a douche for the article he made), an enthusiast will willingly replace his PSU if it can't power something he really wants, in the HPC market that won't happen, basically because they would have to replace thousands of power supplies AND maybe even completely upgrade the power/cooling capabilities of the building or the floor where their server is, and that's not always posible. In fact, that's what marks the limit in performance for most supercomputers.

* Why 2 and not 1 or 3? Because it would mean the card would draw either 100-125w and would be too slow (or a miracle ) or 300w++ which would be way too much to cool on a single socket.

Nicely said! I couldn't have said it any better than this--not even close!

 

[Bo_Fox]

IF you are talking about the "GTX 390", I guess it really isn't a problem that each GPU is weaker than their top-end single GPU cards.
As long as it is able to take a significant lead, it will be fine.

Also, the 4870X2 does not eat more than 300Ws.
The GTX 295 is more efficient, but both cards typically loads well under 300W and do not exceed their deisgned power.

A single 4870X2 eats up to a whopping 381W.

Anyways, I'd rather see a super-powerful GTX 380 single GPU that eats like 300W (outperforming a 5970), and then a dual-GPU GTX 390 with each core that performs about 65% as good as a GTX 380 (so that it'd still be faster overall), eating around 350-400W.

This is just a personal pipe-dream of mine!

At least NV acknowledges that they have been working long and hard on ironing out the hot-spots of the super-dense 40nm silicon to ensure better reliability and cooling.

 

[DirectorC]

has any prices been released yet?

well if you look at the prices of nvidia's high end launch prices then its somewhat like this in recent time, from what I remember the MSRP's for the cards.

7900GTX=$500
8800GTS640=400(but could be found for 350)
8800GTX=$600
GTX260=$400
GTX280=$650(was supposed to be 600)

do you see the trend

so based on that I would think that they would cost at least 500 dollars, more than likely 650 for the high end version, but guessing them is gonna be hard this time but just remember, nvidia hasn't made a flagship card for under 500 dollars in a while now.

looks like fermi is going to cost gtx395 for 700$!!!
Really stupid....
I hope prices don't go with that trend

Yah well with this debacle they are having now where their wafers are practically useless these GPUs are gonna cost consumers upwards of $1,000 to make up for the loss of being douche bags when it comes to the design and manufacturing process.

 

[KainXS]

Nvidia said they are releasing GTX380 and 360 in Q1 2010 then in Q2010 the rest of the mainstream cards. This means you probably have to wait 6 months from now to get a mainstream Nvidia card.

Q1 2010 is only a rumor last time I checked.

Nvidia will more than likely launch the Telsa's before the Geforces, especially being that most of the Fermi cards they have showed are Telsa's and most of the Fermi info on their site refers to the Telsas, and on Nvidia's own page it says they will choose the first to get them with a deadline of January 31 2010, (Buy Telsa today be the first to upgrade to Fermi) you know a pre-launch for Fermi cards in general.

http://www.nvidia.com/object/mad_science_promo.html

now look at nvidia's own page for the 2 cards and it clearly says the Fermi variants of Telsa, the Telsa C2050 will be released in Q2 2010 and the C2070 will be released in Q3 2010

http://www.nvidia.com/object/product_tesla_C2050_C2070_us.html

they said the fermi telsa's would both be released in Q2 before but now one is not.

Yah well with this debacle they are having now where their wafers are practically useless these GPUs are gonna cost consumers upwards of $1,000 to make up for the loss of being douche bags when it comes to the design and manufacturing process.

you can only hope they won't make the same mistake that they have been doing.

 

[CDdude55]

Nvidia delays Fermi to March 2010; AMD to launch new GPU in January-February

http://www.digitimes.com/news/a20091228PD207.html

 

[Bo_Fox]

Awww... more bad news! (At least good news about ATI, but hope that's not just the 2GB version of the 5870).

 

[phanbuey]

geez... 3 months before they even launch, and due to supply issues it will probably be sold out/impossible to find for another 3 months after that. :/ TSMC fail....

 

[bobzilla2009]

It is sucky indeed, they really need to get those cards out so they can work on their midrange, otherwise amd have won this generation easily. Also, nvidia have lost integrated graphics as a source of income now, so discreete gpu's, hpc's and tegra are all they have.

 

[kid41212003]

And I'm getting sick of this thread, haven't read any useful posts that actually related to the Fermi for awhile now.

 

[qubit]

Another 2900XT?

I wonder if Fermi is going to be another lemon like the 2900XT: big, hot & slow?

 

[erocker]

I wonder if Fermi is going to be another lemon like the 2900XT: big, hot & slow?

How would we know? It's possible. It's possible it won't either. There isn't much about Fermi to talk about that hasn't been covered. Maybe we'll know more in a couple months.

 

[qubit]

How would we know? It's possible. It's possible it won't either. There isn't much about Fermi to talk about that hasn't been covered. Maybe we'll know more in a couple months.

Indeed, I'm just musing. However, the longer this thing is delayed, the more likely that it will fall into this trap.

Qubit returns to daydreaming... Heck, I wonder if it will implement the audio part of HDMI properly, like ATI and won't need that little cable?

 

[erocker]

No, it will use a new adapter called "Nvidiout" as a part of their "The way it's meant to be hooked up to a display" campaign. The Nvidiout cable will cost a thousand dollars and be available sometime in 2012. It will make bacon.

*Meh, bored... Back to work.

 

[qubit]

No, it will use a new adapter called "Nvidiout" as a part of their "The way it's meant to be hooked up to a display" campaign. The Nvidiout cable will cost a thousand dollars and be available sometime in 2012. It will make bacon.

*Meh, bored... Back to work.

Nice.

 

[dir_d]

What we need to know is if in March the Fermi launch will be a "paper launch" or will it be actually cards being sold in march

 

[erocker]

Nvidia plans to launch a 40nm GDDR5 memory-based Fermi-GF100 GPU in March, and will launch a GF104 version in the second quarter to target the high-end market with its GeForce GTX295/285/275/260

So their first card coming in March will not be a "high end" card. 2nd quarter is when exactly?

For performance level markets, Nvidia will allow its GeForce GTS250, GT240/220 and 9800GT/9500GT defend against AMD's Radeon HD 5770/5750, 4870/4850 and 4670/4650.

Nothing new, same old hardware. Also 4xxx series will be dumped right after CES and be replaced with new DX11 cards. Nvidia doesn't have much of a "defence." Perhaps another rename is in order?

For the mainstream market, Nvidia will mainly push its GeForce 210.

Low performing DX10.1 for the future? Sounds bad.

 

[a_ump]

supposedly Fermi/GTX3XX is going to dominate every segment in the market, or so says nordichhardware
http://www.nordichardware.com/news,10469.html
n just a lil bit more confirming what most of us have been assuming but more recent conformation
http://www.i4u.com/article29486.html
(fermi not till march 2010, ATI right on schedule)

 

[AsRock]

Sorry not kept up with this thread but is this old news or ?. Either way a interesting read .
http://www.nvidia.com/docs/IO/43395/BD-04983-001_v01.pdf

 

[a_ump]

Sorry not kept up with this thread but is this old news or ?. Either way a interesting read .
http://www.nvidia.com/docs/IO/43395/BD-04983-001_v01.pdf

if its bout fermi n its specs n whatnot then yea lol like 10 pages old news XD. not to be harsh but yea lol i wish people would read all the stuff, i know i always do n then u get a better understanding on what n where the thread is. Plus u learn as much as anyone knows on the subject

 

[qubit]

nvidia PR comes out all guns blazing

Fermi is gonna have serious 0wnage in every single performance segment! Yeah baby!

Allegedly.

 

[Zubasa]

http://tpucdn.com/reviews/ASUS/EAH5970/images/power_maximum.gif

A single 4870X2 eats up to a whopping 381W.

Anyways, I'd rather see a super-powerful GTX 380 single GPU that eats like 300W (outperforming a 5970), and then a dual-GPU GTX 390 with each core that performs about 65% as good as a GTX 380 (so that it'd still be faster overall), eating around 350-400W.

This is just a personal pipe-dream of mine!

At least NV acknowledges that they have been working long and hard on ironing out the hot-spots of the super-dense 40nm silicon to ensure better reliability and cooling.

That is where the melting VRMs come from isn't it?
Normally it is impossible to load a 4870X2 to 381W anyways.

 

[yogurt_21]

I wonder if Fermi is going to be another lemon like the 2900XT: big, hot & slow?

I doubt it, it's not like nvidia has to rush something out because their current cards are nowhere near competitive like ati did with the 2900xt. the gtx295, gtx285, gtx275 are still very competitive.

so if fermi wasn't up to snuff they could simply hold off until they had something that was.

as it stands I seriously doubt it's that femri isn't comptetitive. It's that they simply can't get enough good chips per wafer; which is somehting that always improves with time. And that probabally fuels the midrange-lowend with tons of chips that can be salvaged by disabling parts of them. That would explain why they're being launched first. Especially when nvidia likes to claim the highend crown first to fuel mid and lowend sales normally.

 

[qubit]

I doubt it, it's not like nvidia has to rush something out because their current cards are nowhere near competitive like ati did with the 2900xt. the gtx295, gtx285, gtx275 are still very competitive.

so if fermi wasn't up to snuff they could simply hold off until they had something that was.

as it stands I seriously doubt it's that femri isn't comptetitive. It's that they simply can't get enough good chips per wafer; which is somehting that always improves with time. And that probabally fuels the midrange-lowend with tons of chips that can be salvaged by disabling parts of them. That would explain why they're being launched first. Especially when nvidia likes to claim the highend crown first to fuel mid and lowend sales normally.

Yes, that all makes a lot of sense. Let's hope you're right.

 

[phanbuey]

LoL at the PR... its like "yes very impressive with our internal numbers" well F4*K, make those numbers external - the worst it could do is cause some people to hold off buying the AMD dx11 solution for a while. Which is good for NV.

Real PR uses real numbers - this is the type of PR AMD was using for the R600 (and the original phenom). "Yes its a monster, yes yes, 512bit bus, 320SP's are internal numbers are AWESOME, way more tflops than G80" and then

 

[Bo_Fox]

That is where the melting VRMs come from isn't it?
Normally it is impossible to load a 4870X2 to 381W anyways.

Ha ha.. only when running the special OCCT GPU test that a few creative people wanted to call a "power virus".

When TPU tested for the max power consumption, Furmark was used at default settings. Furmark could be adjusted to reach even higher consumption, especially if the .exe file is renamed. With OCCT, most of the 48xx cards will crash within a minute.

Think about it, a 4870X2 is actually a little more power-efficient than having two 4870's in crossfire when under full load.

I know this is OT now, but I did not want to just ignore your post without replying to it.

Well, to be back on topic, once again I do not care if NV releases a single-GPU card that eats up to 300W, becuase we have dual-GPU video cards that can eat more than that.