NVIDIA’s GeForce GT 220: 40nm and DX10.1 for the Low-End
by Ryan Smith on October 12, 2009 6:00 AM EST- Posted in
- GPUs
There are some things you just don’t talk about among polite company. Politics, money, and apparently OEM-only GPUs. Back in July NVIDIA launched their first 40nm GPUs, and their first GPUs featuring DX10.1 support; these were the GeForce GT 220 and G 210. And if you blinked, you probably missed it. As OEM-only parts, these went into the OEM channel without any fanfare or pageantry.
Today that changes. NVIDIA is moving the GT 220 and G 210 from OEM-only sales to retail, which means NVIDIA’s retail vendors can finally get in on the act and begin selling cards. Today we are looking at one of the first of those cards, the Palit GT 220 Sonic Edition.
| Form Factor | 9600GT | 9600GSO | GT 220 (GDDR3) | 9500GT | G 210 (DDR2) |
| Stream Processors | 64 | 48 | 48 | 32 | 16 |
| Texture Address / Filtering | 32 / 32 | 24 / 24 | 16 / 16 | 16 / 16 | 16 / 16 |
| ROPs | 16 | 16 | 8 | 8 | 8 |
| Core Clock | 650MHz | 600MHz | 625MHz | 550MHz | 675MHz |
| Shader Clock | 1625MHz | 1500MHz | 1360MHz | 1400MHz | 1450MHz |
| Memory Clock | 900MHz | 900MHz | 900MHz | 400MHz |
400MHz |
| Memory Bus Width | 256-bit | 128-bit | 128-bit | 128-bit | 64-bit |
| Frame Buffer | 512MB | 512MB | 512MB | 512MB | 512MB |
| Transistor Count | 505M | 505M | 486M | 314M | 260M |
| Manufacturing Process | TSMC 55nm | TSMC 55nm | TSMC 40nm | TSMC 55nm | TSMC 40nm |
| Price Point | $69-$85 | $40-$60 | $69-$79 | $45-$60 | $40-$50 |
GT 220 and G 210 are based on the GT216 and GT218 cores respectively (anyone confused yet?) which are the first and so far only 40nm members of NVIDIA’s GT200 family. These are specifically designed as low-end cards, with 48 SPs on the GT 220, and 16 SPs on the G 210. The GT 220 is designed to sit between the 9500GT and 9600GT in performance, making its closest competitor the 48SP 9600GSO. Meanwhile the G 210 is the replacement for the 9400GT.
We will see multiple configurations for each card, as NVIDIA handles low-end parts somewhat looser than they do the high-end parts. GT 220 will come with DDR2, DDR3, or GDDR3 memory on a 128bit bus, with 512MB or 1GB of it. So the memory bandwidth of these cards is going to vary wildly; the DDR2 cards will be 400MHz and the GDDR3 cards may go as high as 1012MHz according to NVIDIA’s specs. G 210 meanwhile will be DDR2 and DDR3 only, again going up to 1GB. With a 64bit memory bus, it will have half the memory bandwidth of GT 220.

The GPU configurations will vary too, but not as wildly. NVIDIA’s official specs call for a 625MHz core clock and a 1360MHz shader clock. We’ve seen a number of cards with a higher core clock, and a card with a lower shader clock. Based on the card we have, we’re going to consider 635Mhz/1360Mhz stock for the GPU, and 900MHz stock for GDDR3, and test accordingly.
The transistor count for the GT218 die comes out to 260M, and for GT216 it’s 486M. NVIDIA would not disclose the die size to us, but having disassembled our GT 220, we estimate it to be around 100mm2. One thing we do know for sure is that along with the small die, NVIDIA has managed to knock down power consumption for the GT 220. At load the card will draw 58W, at idle it’s a tiny 7W. We don’t have the power data for the G 210, but it’s undoubtedly lower.

The prices on GT 220 cards are expected to range between $69 and $79, with the cards at the top end being those with the best RAM. This puts GT 220 in competition with AMD’s Radeon HD 4600 series, and NVIDIA’s own 9600GT. The G 210 will have an approximate price of $45, putting it in range of the Radeon HD 4300/4500 series, and NVIDIA’s 9500GT. The GT 220 in particular is in an odd spot: it’s supposed to underperform the equally priced (if not slightly cheaper) 9600GT. Meanwhile the G210 is supposed to underperform the 9500GT, which is also available for $45. So NVIDIA is already starting off on the wrong foot here.
Finally, availability should not be an issue. These cards have been shipping for months to OEMs, so the only thing that has really changed is that now some of them are going into the retail pool. It’s a hard launch and then some. Not that you’ll see NVIDIA celebrating; while the OEM-only launch was no-key, this launch is only low-key at best. NVIDIA didn’t send out any samples, and it wasn’t until a few days ago that we had the full technical data on these new cards.
We would like to thank Palit for providing us with a GT 220 card for today’s launch, supplying us with their GT 220 Sonic Edition.

82 Comments
View All Comments
brybir - Monday, October 12, 2009 - link
Your statement is only partially true.There are several games that have some DirectX 11 features out right now. Perhaps the more accurate thing to say is that DirectX 11 will not see feature set adoption en mass until sometime in 2011.
I think ATI somewhat admits this as they spent a good deal of time tweaking some of its driver and hardware features to boost the performance of directX 9 engine games. There was something about that on Anand a few weeks ago about that. Reply
BenSkywalker - Monday, October 12, 2009 - link
You show the 9600GSO winning the majority of the benches you decided to allow it to take part in, it is cheaper then the 4670, and the 4670 is the clear winner?Why do you bother quoting the price of the 9600GT when you refused to show benchmarks for it?
Right now on the Egg you can get a 9600GSO for $40 AR, $60 before rebate. The article may be right in terms of the parts that are launching being a bad value, but more then anything that is because of how soundly they are bested by nV's existing parts- which are already faster and cheaper then the 4670. Reply
Ryan Smith - Monday, October 12, 2009 - link
It's one of those 96SP GSOs based on G92. We include for reference only; you can't buy them any more (and the 96SP model listed on Newegg is wrong, it's a 48SP model). Replychizow - Monday, October 12, 2009 - link
Is this the one you're referring to Ryan?http://www.newegg.com/Product/Product.aspx?Item=N8...">http://www.newegg.com/Product/Product.a...E1681412...
From all reports and feedback I've seen, that Asus part is 96SP for $40 AR (was actually $35 at one point!) just as Ben stated.
I don't really care about this segment of parts for actual 3D so you may be right. I was only interested in that Asus 96SP because it was identified by many users as "the card" to get for dedicated PhysX on a budget. I was hoping to find a low-power consumption candidate on 40nm for dedicated PhysX, but it seems these parts have too few SP to be viable. Reply
BenSkywalker - Monday, October 12, 2009 - link
Why not include benches of the nV part that you quoted the price for then(btw- NewEgg is also listing one of the 384MB 96SP 9600GSOs for $56.49)?It is rather misleading to put it mildly that you decide to include benches for one part that is quite a bit cheaper, and quite a bit slower, and then quote the price on the next rung up on the performance ladder. It would be akin to showing benches for the GTX275 and lamenting it because the GTX285 was so expensive, it doesn't make any sense. Reply
Ryan Smith - Monday, October 12, 2009 - link
Which part? The 9600GT, or the 48SP GSO?What you see are all of the low-end cards we were able to get our hands on for this article. Reply
jordanclock - Monday, October 12, 2009 - link
The graphs on the "Temperature and Noise" page are high-to-low. Usually they're the opposite, following the idea that lower numbers are better. I know, I know, it's a lame nitpick.As for the card: If they added this checkbox feature for OEMs (which I believe) then why bother allowing for retail distribution? This card really doesn't bring anything that isn't already done with cards they already have available. Seems like a rather silly release. Reply
Ryan Smith - Monday, October 12, 2009 - link
Fixed. Thanks. Replyjonup - Monday, October 12, 2009 - link
Ryan, the Palit table shows their 1GB card's frame buffer as 512MB. Replystrikeback03 - Monday, October 12, 2009 - link
Seems recently there have been a lot of low-end cards with 1GB of memory onboard. Is this just a marketing ploy, or is there actually some scenario where it would be useful to have lots of memory but not much processing power? Reply