We can't say we're entirely shocked to see 'em, but a new pair of GPUs based on 40 nanometer process technology has surfaced over at
NVIDIA's website. Both of the new devices are expected to be sold exclusively to large OEMs for integration into pre-configured machines, and they'll both support DirectX 10.1, OpenGL 3.0, and CUDA. The lower-end GeForce G210 arrives with a 589MHz core clock speed, 512MB of DDR2 RAM and a 64-bit memory interface; meanwhile, the GeForce GT 220 ups the ante with a 615MHz core clock rate, 1GB of GDDR3 RAM and a 128-bit memory interface. As for outputs, the former packs VGA, DisplayPort and DVI, while the latter sticks with VGA, HDMI and DVI. There's no word on when we'll seen them offered in any entry-level desktop rigs, but surely it won't be long now.
[Via
SlashGear]
Read - GeForce G210
Read - GeForce GT 220
posted Jul 9th 2009 9:22AM
no way I'm getting this- it doesn't have 2 DVi ports so it's not optimal for use with the iz3d monitor- while it can use DVi+ VGA why would I want that haha
Put a DVI adapter on that HDMI port and you've got dual DVI.
good idea thanks!
put that shit in a netbook!
The card shown above was NOT mention for end user purchase, hence whey they say it will only be shipped to OEMs for prebuilt systems. Those customers still have VGA port only monitors,
So of course you don't want this card, not to mention it won't pack the graphical punch hardcore gamers demand. What is important is that we'll be seeing newer versions of the GTX260 and GTX285, probably with new names, providing a bit more speed, less noise less heat and less watts used, which is a great thing.
The low profile and the integrated HDMI port suggests to me that this is targeted at OEM HTPCs.
Awesome, what die-shrunk Geforce 8-series cards are these ones based on?
Judging by their clock speed they're probably a G92 base like the GTS 250, and was first seen in the 8800GT IIRC. All the other 200 series carts are G200-based, however.
It's something new cause they support directx 10.1 where series 8 supported only 10.
It looks a lot bigger than 40nm in that photo.
Sigh....why do people who do not understand GPUs/CPUs even read these stories.
The 40nm is in reference to the Die size on the chip, not the chip itself. If you've seen a CPU or GPU before you will notice a silver looking block in the middle of the chip.
Here is a pic of an AMD CPU, notice that silver block in the middle? That is the die, when building your own computer, this is where you place thermal paste.
The die size can shrink and the over all CPU/GPU size will stay the same. This is done for compatibility reasons. For instances, AMD released new Phenom II's based on the 45nm process instead of 65nm, however for compatibility it keeps the same socket size which is AM2+. Dies size is not proportional to the size of the CPU/GPU itself.
Why is smaller better? The main reason is that the smaller the die, the less likely it is to put out as much heat as an older generation and are usually more energy efficient.
Sorry forgot the pic link:
http://articles.techrepublic.com.com/i/tr/cms/contentPics/ontheCPU.jpg
Oooh, i like the shiny picture. Thanks for making this clear to me - I feel silly now!
But...I dont quite follow your explanation about power. Intuitively, I would have thought that smaller feature sizes would result in higher gate capacitance and hence require a larger vcore (and hence greater power dissipation, contrary to your claim) to achieve equivalent clock rate to a larger design. Why isn't this the case?!
Also I would also have guessed that you'd get greater gate leakage at smaller nodes, what with the dielectrics becoming correspondingly thinner (I assume?!). AFAIK they used the same high-k dielectrics at the 65nm node, so unless they've got some exotic new material why doesn't leakage increase?
Also, does 'nm' mean 'nautical miles' or 'nano metres'?
Please help! I'm so confused!
Dubb, I think you missed the joke.
@j_g_puff
nice
Awesome response.
Another thing that confuses me:
Rather than shrinking dies all the time, wouldn't it be better to have larger die with a lower transistor density? The thermal density would be lower and the larger surface area would improve heat transfer to the heatsink...both of which would reduce thermal fatigue. Of course, you'd have longer tracks in some places which would be harder to drive and would generate more heat, but still. Anyone know why this isn't done?
My guess is that manufacturers don't care about thermal fatigue, since power dissipation and clock speed are usually more important to customers (except maybe the military). Thoughts?
Some people obviously can't spot sarcasm when they see it.
"Another thing that confuses me:
Rather than shrinking dies all the time, wouldn't it be better to have larger die with a lower transistor density? The thermal density would be lower and the larger surface area would improve heat transfer to the heatsink...both of which would reduce thermal fatigue. Of course, you'd have longer tracks in some places which would be harder to drive and would generate more heat, but still. Anyone know why this isn't done?
My guess is that manufacturers don't care about thermal fatigue, since power dissipation and clock speed are usually more important to customers (except maybe the military). Thoughts?"
I can't tell if this one is serious or not, but I'll try to answer it at my own risk.
The protruding metal square is not size representative of the actual CPU lying underneath, it is merely a shield from the outside world. In fact, you should only apply a single, thin line either horizontally or vertically, depending on the orientation of the cores (across the two, not between them). Quad-cores call for two pencil-lead thin lines of paste running parallel. That's why the last two revisions of Pentium 4 have the same die size, although the architecture advanced from 90 nm to 65 nm.
Increasing the size of the CPU would help dissipate heat under load, but the associated power cost with increased gate distance would lead to LOWER performance/watt because of greater leakage. Strained silicon allows for faster transistor switching at the cost of usefulness/electron. The difference between Prescott and Cedar Mill was much greater than the difference between AMD's Windsor and Brisbane. Both were 90 nm to 65 nm, but Intel switched from SOI to strained silicon whereas AMD was already on strained Si.
Furthermore, when idle, the high density allows for sleep transistors to act as heatsinks for active transistors. Surface heat retention is almost negligble when dealing with these conditions. I don't have exact figures, but it's something along the lines of halving the surface area/volume ratio decreases heat retention by only a little; while performance, efficiency, and manufacturing costs are all greatly sacrificed.
"Rather than shrinking dies all the time, wouldn't it be better to have larger die with a lower transistor density? The thermal density would be lower and the larger surface area would improve heat transfer to the heatsink...both of which would reduce thermal fatigue"
So you are suggesting we reverse Moore's Law ? lower transistor density means much more expensive chips, high density means higher yield which is better profit, Theral fatigue is caused by heat not density, die shrunk chips fit less electrons down smaller circuits so use less power and generate less heat.
You are suggesting this to reduce heat yet my graphics chip and CPU runs with passive sinks in my HTPC my 7600GS is 3 years old and I have no problems, thermal fatigue is not a certainty, just make sure you use 3rd party copper heatsinks and you have nothing to worry about.
Hung,
It was serious, and your answer was seriously informative. Thanks!
Manufacturing cost per unit is directly proportional to die size.
The cost of a wafer (usually a 300mm disc) is more or less constant for a particular process technology. The factors that determine how much your integrated circuit (e.g. CPU or GPU) cost are:
- How many dies fit on a wafer. Smaller dies mean you get more ICs from a single die, which means lower cost.
- Yield. A significant fraction of the dies don't work because of defects. Techniques like binning, disabling broken functional units, and design changes can help here.
Die shrinks have two benefits:
- Lower cost per unit. Although smaller processes technologies cost more (and may have lower yield), the effect is offset by the increased number of dies per wafer.
- Lower heat / lower power usage / higher clocks. The manufacturer can choose to sell the die-shrunk part at the same clock but with less heat and power consumption. Or they can increase the clock to get better performance (assuming that the design allows it). Or they can do a combination of both.
Note that die shrinks are *not* easy. Timings and electrical characteristics have to be re-optimized, sometimes requiring significant changes. New masks have to be produced, new PCBs may have to be designed, and other factors also change.
Also note that 40nm refers to the size of individual features on the IC, not to the size of the IC itself. 40nm is absolutely tiny - far smaller than anything that you would be able to see without a microscope. The GPU itself is rectangular and usually around 100-1000mm^2 (high-end GPUs are usually larger).
Also important is that "die size" refers to the actual size of the die, not to the size of the heat spreader. All modern desktop CPUs and many GPUs have a heatspreader (often soldered to the die) that has two main functions:
- Protects the die from the force of heat sink installation
- Provides greater surface area to transfer heat to the heat sink, compensating for inferior contact between the heatsink and the die/heatspreader
Note that the proper method of installing thermal compound is to use a thin, even layer. You want to maximize the metal-to-metal contact area, but still prevent air pockets from forming. Thermal compound is actually very poor at conducting heat compared with aluminum or copper, but it's much better than air.
@Dubb
"Sigh....why do people who do not understand GPUs/CPUs even read these stories.
The 40nm is in reference to the Die size on the chip, not the chip itself. If you've seen a CPU or GPU before you will notice a silver looking block in the middle of the chip. ...Here is a pic of an AMD CPU, notice that silver block in the middle? That is the die, when building your own computer, this is where you place thermal paste."
You shouldn't bitch about the technically ignorant if you yourself are clueless... If the GPU's silicon die size was only 40nm, They'd be able to fit billions of GPUs on a wafer! 40nm is the minimum feature size, aka the transistor gate length.
@Dubb
WRONG!
prolly the 8400 and 8600.. I mean the 9400 and 9600..
1GB of VRAM on a 128-bit interface? sounds like a terrible idea.
At least they're nice and small.
how much $$$$?
nvidia ion, anyone?
nope, ion 2
http://www.engadget.com/2009/07/01/nvidia-said-to-be-prepping-ion-2-for-late-2009/
since the original ion uses the 9400, i'm gonna assume the ion 2 will be geforce 2 series based.
Its nice to see the low end 2 series finally comes out but they should have just focussed on getting DX11 cards out the door, even though it doesnt matter so much for the low end stuff. GPGPU would have been nice for these in Win7. I guess AMD will be the first to market since the 48xx series has been out since June 2008 and they already previewed their new silicon secretly to a few tech sites.
Competition FTW
seriously, this is an improvement, the 265GX that i put in my machine the other day is the size of a coffee table book.
Hmm.... performance-wise, how would these compare to my aging GeForce 8800GTS 512 meg? Faster? Quieter? Lower power/heat?
Much, much slower. Probably lower power, but that kind of comes with this level of performance.
Even though this is a better chip, I have to agree, it's probably slower. You'll want to wait until they re-release the GTX260 or GTX285 as 40nm before you replace your card. Most likely Nvidia will rename the cards, not upgrade them to 40nm silently.
And considering that Nvidia loves pushing their model numbers up whenenever they can, I suspect it'll be called a GT300 something, not a GT265 / GT290.
I've been out of the loop for a while. Anyone know when their new flagship gaming card is coming out and what the name of it is?
Thanks
Off the top of my head I'd say october for new flagship cards. Most likely they'll be the GTX300 series.
Ray - that is helpful thanks. I'm not an nVidia or ATI snob - whoever happens to have the best product at a reasonable price and the quietest cooling system are what grab my attention. We'll have to see later this year who has that as I'm ready to upgrade again. Crysis is just too demanding :)
Yes, the new flagships will be the 300 series, but Nvidia hasn't fully confirmed dates nor model names yet. So far we just have rumors and unconfirmed leaks, imho the card posted here is a handicapped preview of what's to come. Normally Nvidia releases new chips as the highend card not the cheap low end, so I have no idea what they're up too, perhaps they just want to tap the cheap desktop market since the Aero desktop for Vista and Win7 is best on an efficient GPU (as in low watt, low heat) which this new chipset will deliver.
This makes me suspect (as we all knew anyways) that the GTX300 series won't arrive until on or after Win7, which is August for pre-built and October for retail purchase.
With all of the variation of chipsets, clock speeds, memory types and die sizes; how does the "average" computer shopper know how to compare the specs from card to card? I have to admit I was a huge hardware fan back in the day (10 years ago) and I have moved on to software and frankly can't keep up. If I were to build a computer now it would require lots of research just to learn the different chipsets and features available and I would have not point of comparison between video cards except on a hardware review site that compared the two.
The same thing is happening with CPU's now, what reference do we possible have for actual processing power and throughput? Maybe I am just ranting before my second cup of coffee this morning but I strongly feel that some sort of "consumer reference" for benchmarking CPU's, GPU's and even TV's these days is absolutely neccessary.
I know there has been lots of blowback with consumers after they buy a netbook with "Intel Inside" only to find out that it can barely run WinXP on it.
Woohoo...looks like they finally figured out how to put an HDMI on without the fugly dvi to hdmi adapter that sticks about a foot away from the computer and breaks off easily.
Yay low profile! Might have to grab one for my media centre, the 7300GS doesn't seem capable of 1080 content unless anyone knows of a better low profile video card?
If you can't run 1080p it's probably because of your CPU, video cards only accelerate some video files. A 7300GS isn't very fast, in fact it's technically handicapped, so perhaps your problem really is the video card acceleration.
But does it run Crysis?!?
Yes, but you'll need 3 of them to run tri-SLI :P
I find what is most interesting is that the card is said to support DX 10.1 in this article. I was under the impression that all G92/GT200 arch-based cards are not completely compliant with DX 10.1 specs. So that would mean that this would be a debut of some new/tweaked architecture. And if that new-ish architecture is not based on the arch coming with DX 11 cards could that mean we will see DX 10.1, 40 nm updated GT200 series cards debuting before DX 11 cards in September?
I'm just curious if that is correct or if it was just a mistake in the article?
yeah it's correct (nvidia website). it was the first thing i noticed as well. i didn't think they would bother given the release of dx11. they carried on like they were not interested when ATI went 10.1
"Another thing that confuses me:
Rather than shrinking dies all the time, wouldn't it be better to have larger die with a lower transistor density? The thermal density would be lower and the larger surface area would improve heat transfer to the heatsink...both of which would reduce thermal fatigue. Of course, you'd have longer tracks in some places which would be harder to drive and would generate more heat, but still. Anyone know why this isn't done?
My guess is that manufacturers don't care about thermal fatigue, since power dissipation and clock speed are usually more important to customers (except maybe the military). Thoughts?"
They do care (a bit) about "thermal fatigue", which I think is your reference to what's known as electromigration. More importantly, the thermal density IS something GPU makers have to worry about, because if your heatsink can't transfer the heat from the tiny little chip into the big heatsink and fan that's attached, then the GPU will overheat and then melt. But those aren't too big concerns to GPU makers RELATIVE to the cost of the GPU. The main determining factor for the cost of making a GPU, once you've finished designing the chip, is the die size. The die size is the physical square area of the chip (usually measured in square millimetres, or mm^2). That's why going to a smaller manufacturing process such as 40nm (which more or less indicates how small they can make the transistors or switches which make up the GPU) saves you money - it lets you pack in the same number of transistors into a smaller area (or die size).
Here's the main point, though - WHY does a smaller die size mean a lower manufacturing cost? Because this is how CPU's or GPU's are made: they are produced on silicon wafers. Usually the silicon wafer is 300mm in diameter, and it's basically a thin disc of silicon. How you make the CPU's/GPU's is you "pattern" (through a process called photolithography) the tiny transistors which will make up the circuits of the CPU/GPU. It's a combination of an optical projector-like process and a chemical process, done over and over again.
So if your CPU/GPU design takes up lots of space on the wafer, then you will be able to fit much fewer chips onto your silicon wafer. The bulk of the cost is in processing each wafer and how many optical/chemical steps your chip requires to be made, NOT how many chips are actually being fit onto the wafer. So if your chip is half the size and let's say you can fit twice as many on the wafer as before, then all of a sudden your chip costs 50% less to make.