Nvidia Preps OEM GeForce GT 630, GT 645 & Three GT 640s
Nvidia has released its first desktop graphics cards based on its new 28 nm Kepler GK107 silicon for OEMs, along with a few Fermi-based GPUs.
Nvidia has released five new GeForce cards under the 600 series, which are designated for OEMs only. There still is no official word from Nvidia on when the Kepler-based GK107 cards are expected for the desktop (though we should expect them here in the near future).
The top card, the GeForce GT 645, is a re-branding of the recently released Fermi-based GF114 GeForce GTX 560 SE. The other Fermi-based card is one of three GeForce GT 640 cards. It is based on the Fermi GF116, which features 144 CUDA cores, 1.5 GB or 3 GB GDDR3 192-bit memory interface, 720 MHz core clock speed, and 891 MHz memory clock speed.
There are three Kepler-based GK107 graphics cards, which all feature 384 CUDA cores, 24 TMUs and 16 ROPs. The two GeForce GT 640s differ in core clocks, memory, and TDP. At the high-end, the GT 640 has a core clock speed of 950 MHz, with the low-end running at 797 MHz. The memory clocks come in at 2500 MHz (GDDR5) and 891 MHz (GDDR3) respectively. The cards offer TDP of 75 W and 50 W.
One of the more interesting cards of the bunch is the GeForce GT 630. It is a single-slot, low-profile graphics card with 384 CUDA cores, 128-bit wide GDDR3 memory interface, 875 MHz GPU clock speed, 891 MHz memory clock speed and TDP of 50 W. When compared against the low-end GT 640, the GT 630 is actually the better card and in-turn will offer users better performance in their OEM system.
Specifications:
| GeForce GT 630 | GeForce GT 640 | GeForce GT 640 | GeForce GT 640 | GeForce GT 645 | |
|---|---|---|---|---|---|
| Architecture | Kepler GK107 | Kepler GK107 | Fermi GF116 | Kepler GK107 | Fermi GF114 |
| CUDA Cores: | 384 | 384 | 144 | 384 | 288 |
| Core Clock: | 875 MHz | 797 MHz | 720 MHz | 950 MHz | 776 MHz |
| Shaders Clock: | 875 MHz | 797 MHz | 1440 MHz | 950 MHz | 1552 MHz |
| Memory Clock: | 891 MHz | 891 MHz | 891 MHz | 2500 MHz | 1914 MHz |
| Memory Type: | 1 or 2GB GDDR3 128-bit | 1 or 2GB GDDR3 128-bit | 1.5 or 3GB GDDR3 192-bit | 1 or 2GB GDDR5 128-bit | 1GB GDDR5 128-bit |
| Memory Bandwidth | 28.5 GB/s | 28.4 GB/s | 43 GB/s | 80 GB/s | 91.9 GB/s |
| Open GL: | 4.2 | 4.2 | 4.2 | 4.2 | 4.2 |
| PCI-Express: | 3 | 3 | 3 | 3 | 2 |
| DirectX: | 11 | 11 | 11 | 11 | 11 |
| TDP: | 50 W | 50 W | 75 W | 75 W | 140 W |
| GeForce GT 630 | GeForce GT 640 | GeForce GT 645 |
|---|---|---|
| Image Credit: VideoCardz | ||
![]() | ![]() | ![]() |



We all know they are the same old dog in a new bun with some new ketchup on top.
We all know they are the same old dog in a new bun with some new ketchup on top.
Is that even legal? How would even their corporate customers know what they are getting if the model number is all the same? I suppose with cars the same is true, there are various sub-types all with different specs. But it certainly breaks the mould, and the trust that clients have in Nvidia.
The thing is, especially with OEMs, the threshold of entry-level discreet has made quite a leap, and will jump even higher with Trinity. If the reports of enhancements to dual-graphics are accurate it will make even less sense.
AND, if the rumors concerning TrinityII are accurate, HD7750s / 550TIs days are numbered.
If someone checks, I think they will find AMD plans to re-brand low-end 6000 series to 7's so nVidia understandably sees the writing on wall.
Exactly, noobs that don't do their research will suffer from this.
Way to to confusing your customers nVidia.
Nvidia doesn't force PhysX on anyone (in fact, most games don't even support it and the Kepler cards aren't even as good at it as the Fermi cards, so it seems that Nvidia is stepping back on PhysX). Furthermore, PhysX does not hamper game evolution. Also, AMD re-badges and reuses older cards and architectures too.
Radeon 5750 was re-badged to Radeon 6750 and 5770 to 6770.
All Radeon 6000 cards that weren't re-badges or weren't Radeon 6900 cards used the same VLIW5 architecture as the Radeon 5000 cards even if they weren't exactly re-badges.
Many of the "new" Radeon 7000 cards will reuse the same VLIW5 architecture that has been in use fore two generations prior to the Radeon 7000 cards. For example, all mobile and desktop Radeon 7600 and below cards will use the VLIW5 architecture, so although they might not be re-badges, they will at the very least be using a several years old architecture.
So, both Nvidia and AMD will be reusing their first DX11 architectures, Fermi and VLIW5, for their low end cards two generations later. Both Nvidia and AMD have been doing this for a while now. It is nothing new for either company.
I completely agree with the first paragraph, Nvidia pulled a lot of the muscle out of the new cards to favor FPS over compute power (which PhysX is one of the victims as you named). However, I have no problem if the same architecture is used for a new generation of cards, just don't make it the same exact part with no modifications or tweaks and call it new and improved with no change except for the name.
GeForce GT 630
Architecture Kepler GK107
Cuda Cores 384
TDP 50W
Low profile.
Means just one - if released to custom PC market this is the fastest low profile GPU with lowest TDP. I know what I am getting for my Low Profile PC.
No, if you read the review, you'd know that the GTX 680 is the "gamer focused" card that purposefully sacrifices compute, and that the other high end kepler cards will focus more on it. Also, rebranding is one thing; calling three different cards the same thing is just taking the mickey.
... The Radeon 7000 cards are all good for compute, even the Pitcairn and Cape Verde cards that aren't even compute focused like Tahiti is. All of the Fermi cards are decent for compute. The GTX 680 is the top Kepler gaming card out right now and we don't know for sure if that will change (685, 680 TI, etc, is all possible, but we don't know for sure). So no, alpha wasn't wrong there. The Pitcairn and Cape Verde didn't sacrifice nearly as much compute performance to get comparable gaming performance per watt and per mm2 of die to Kepler.
The 7970 and other cards are for gamers just as much as they are for professionals and the same is true for the Fermi cards such as the 480 and 580. Reading the review or not, you should know that within the tech industry, nothing is for granted until it's out and even then, it might not be. For example, we don't know if there even will be a GK100 cards besides Teslas or Quadros.
Also, here's Nvidia's take on re-badges:
Then, we then got a faster version: the 8800 GTS 512MB. It was more expensive, but we were still happy.
And then it got complicated.
The original 8800 GT, well, it became the 9800 GT. Then they overclocked the 8800 GTS and it turned into the 9800 GTX. Now this made sense, but only if you ignored the whole this was an 8800 GT to begin with thing.
The trip gets a little more trippy when you look at what happened on the eve of the Radeon HD 4850 launch. NVIDIA introduced a slightly faster version of the 9800 GTX called the 9800 GTX+. Note that this was the smallest name change in the timeline up to this point, but it was the biggest design change; this mild overclock was enabled by a die shrink to 55nm.
All of that brings us to today where NVIDIA is taking the 9800 GTX+ and calling it a GeForce GTS 250.
The only changes here were increasing clock frequencies very slightly and a die shrink of the same GPU later on. Re-badges at it's finest. That 8800 GT really got around, didn't it? Like alpha said, both companies do this.
"Today, Nvidia has announced their new GPU line, the GeForce GT Z660, which is a rebranding of the Z540, which is a rebranding of the Z420, which is a rebranding of the Z310, which is a rebranding of the Z295, which is a rebranding of the Z1, which is a rebranding of the GT 940, which....is a rebranding of the GT 230M"
It'll happen if the executives get their way. With both companies.