NVIDIA Launches Fermi Based GeForce GT 610, GT 620, GT 630 Into Retail
by Ryan Smith on May 19, 2012 8:00 PM ESTWhile we were off at NVIDIA’s GTC 2012 conference seeing NVIDIA’s latest professional products, NVIDIA’s GeForce group was busy with some launches of their own. The company has quietly launched the GeForce GT 610, GT 620, and GT 630 into the retail market. Unfortunately these are not the Kepler GeForce cards you were probably looking for.
| GT 630 GDDR5 | GT 630 DDR3 | GT 620 | GT 610 | |
| Previous Model Number | GT 440 GDDR5 | GT 440 DDR3 | N/A | GT 520 |
| Stream Processors | 96 | 96 | 96 | 48 |
| Texture Units | 16 | 16 | 16 | 8 |
| ROPs | 4 | 4 | 4 | 4 |
| Core Clock | 810MHz | 810MHz | 700MHz | 810MHz |
| Shader Clock | 1620MHz | 1620MHz | 1400MHz | 1620MHz |
| Memory Clock | 3.2GHz GDDR5 | 1.8GHz DDR3 | 1.8GHz DDR3 | 1.8GHz DDR3 |
| Memory Bus Width | 128-bit | 128-bit | 64-bit | 64-bit |
| Frame Buffer | 1GB | 1GB | 1GB | 1GB |
| GPU | GF108 | GF108 | GF108/GF117? | GF119 |
| TDP | 65W | 65W | 49W | 29W |
| Manufacturing Process | TSMC 40nm | TSMC 40nm | TSMC 40nm | TSMC 40nm |
As NVIDIA was already reusing Fermi GPUs for GeForce 600 series parts for the OEM laptop and desktop market, it was only a matter of time until this came over to the retail market, and that’s exactly what has happened. The GT 610, GT 620, and GT 630 are all based on Fermi GPUs, and in fact 2 of them are straight-up rebadges of existing GeForce 400 and 500 series cards. Worse, they’re not even consistent with their OEM counterparts – the OEM GT 620 and GT 630 are based off of different chips and specs entirely.
At the bottom of the 600 series retail stack is the GeForce GT 610, which is a rebadge of the GT 520. This means it’s either a GF119 GPU or cut-down GF108 GPU featuring a meager 48 CUDA Cores and a 64bit memory bus, albeit with a low 29W TDP as a result. This is truly a rock bottom card meant to be a cheap as possible upgrade for older computers, as even an Ivy Bridge HD4000 iGPU should be able to handily surpass it.
The second card is the GT 620, which is a variant of the OEM-only GT 530. With 96 CUDA cores we’re not 100% sure that this is GF108 as opposed to the 28nm GK117, but as NVIDIA currently has a 28nm capacity bottleneck we can’t see them placing valuable 28nm chips in low-end retail cards. Furthermore the 49W TDP perfectly matches the GF108 based GT 530. Compared to the OEM GT 620 the retail model has twice as many CUDA cores, so it has twice as much shader performance on paper, but because of the 64bit memory bus it’s going to be significantly memory bandwidth starved.
The final new 600 series card is the GT 630, which is a rebadge of the GT 440. Like the GT 440 this card comes in two variants, a model with DDR3 and a model with GDDR5. Both models are based on GF108 and have all 96 CUDA cores enabled, and have the same core clock of 810MHz. At the same time this is going to be the card that deviates from its OEM counterpart the most. The OEM GT 630 was a Kepler GK107 card, so this rules out getting a Kepler based GT 630 retail card any time in the near future.
As always, rebadging doesn’t suddenly make a good card bad – or vice versa – but it’s disappointing to once again see this mess transition over to the retail market. We hold to our belief that previous generation products are perfectly acceptable as they were, and that the desire to have yearly product numbers in an industry that is approaching 2 year product cycles is silly at its best, and confusing at its worst.
44 Comments
View All Comments
Orwell - Sunday, May 20, 2012 - link
Heck, even an old 9600GT has 68GiB/s (256/8 * 1090 * 2) of bandwidth.At least pick a decent old card to compare with then! ;)
Oh, and yes, partially thanks to consoles this card can still play nearly any game at high detail, given you disable AA and stick with 1680x1050 or so. Also, they can be had for as little as €40 so its probably cheaper that these three new chips. Moreover, they don't need to be oven baked. Reply
Stuka87 - Monday, May 21, 2012 - link
Well not any game, but a good deal yes.These cards are not going to play something like BF3 at high detail at 1680x1050. They would struggle at low-medium detail settings at that resolution. Reply
CeriseCogburn - Saturday, May 26, 2012 - link
Maybe the amd fans would agree that the Dx11 upgrade cheapos surpass the sad integrated pre AM3 and socket 775 loser chips, and give the cu$tomer a gigantic bang for their buck ?I mean you should see how happy and proud they are they fled the amd X200, X1250, X2100 integrated and cranked their aging system to the DX11 sky for a cheap $35 bucks and a tech notch on their belt.
Of course all the brilliant would be honest underdog CEO's and BSA Market Analysts here have a different story to tell about how stupid the purchasing public is - not necessarily a PR win if their words got out.
Maybe the rage3d fans should think about all the crappy Intel integrated boards these low cost DX11 cards beat by 500% ?
Maybe those buying them aren't so dumb after all - and INSTEAD the usual cheap moaning penny pinching I can't afford a three cent snot bubble upgraders of amd fan persuasion we have here are the clueless and completely thoughtless going against their usual pennies from the peanut butter jar price/perf scrapingly poor "enthusiast" menschen upgrade mind spew. Reply
yannigr - Sunday, May 20, 2012 - link
For $40 / 30€ they are not bad. For old PCs without graphics I mean. For new PCs with ultra cheap graphics the right way is Llano/Trinity/...Intel. ReplyScali - Sunday, May 20, 2012 - link
Which may be exactly why they're not bothering to introduce new GPUs at the low-end?I guess this level of performance will be phased out as it is now being surpassed by APUs. So why develop shiny new 28 nm GPUs for that? They probably wouldn't sell enough to make it worthwhile. Reply
Wwhat - Sunday, May 20, 2012 - link
They should make mixed mode wafers where the central part has the high end thus large GPU's and then the edge has the small low end ones.I wonder if they ever did the calculation on that, what with the lousy yield of 28nm that might be a real good idea. Reply
Scali - Sunday, May 20, 2012 - link
Not sure about that...Firstly it would require the extra effort to design and test these low-end parts before it would work.
Secondly, as far as I know it is not possible, or at least not cost-effective to build multiple chips on a single wafer.
Namely you normally re-use a single mask for all chips on a wafer. When you mix them, the mask has to be replaced halfway during lithography.
Lastly, the wafer has the least defects in the center, and the most defects at the edges. The yields may be too low for it to be worthwhile anyway. So it's just as well when they go to waste. Reply
Lonyo - Sunday, May 20, 2012 - link
Except they already have low end GPUs which are being used in the OEM versions of these cards, so they have already developed shiny new 28nm GPUs, they just aren't using them in retail. ReplyCeriseCogburn - Saturday, May 26, 2012 - link
Why thank you yannigr for the 1st bit of sense so far.(although you could be sporting a racist name, so anything you say is not worthy, and should be ignored, and you lost all credibility, and if you had a nicer name maybe someone would listen to you or think you could possibly be telling the plain and obvious truth.)
rolls eyes
Oh sorry, I started fitting in and acting like the children here who taught me all too well. Reply
vision33r - Sunday, May 20, 2012 - link
It's not the enthusiat market that keeps Nvidia afloat it's these low end solutions. I order them by the hundreds for businesses. Maybe a few more years Intel and AMD's internal GPU will be powerful enough to replace them. Reply