Thursday, June 24th 2010
NVIDIA GeForce GTX 460 Reference Design Pictured
Here are the first pictures of NVIDIA's GeForce GTX 460 graphics card, an upper-mainstream model based on the company's new GF104 GPU which was pictured earlier. The pictures reveal the reference design card to be shorter than any of the GF100-based graphics cards (such as GTX 470, GTX 480), and compacted in many ways. The cooler is dual-slot, and instead of an air-channel that's draws air from the interior and blows it out from the rear, the cooler has a centrally-located fan right over the GPU. As expected from the older article, the GPU package indeed is rectangular in shape rather than square.
The PCB is black, though a green PCB cannot be written off given the product's positioning. There are traces for eight memory chips on the card (looking at the components on the reverse-side of the PCB), confirming a 256-bit wide memory interface, though six chips are occupied (indicating that for this SKU only, a 192-bit wide memory interface is used. There is only one SLI finger showing that it only supports 2-way SLI multi-GPU standard. Connectivity on the rear panel is consists of the usual 2x DVI-D and mini-HDMI. Power is drawn in by two 6-pin PCI-E power inputs. Other specifications include DirectX 11 compliance, 336 CUDA cores, 768 MB of 192-bit GDDR5 memory (or another SKU with 1 GB of 256-bit GDDR5 memory), and clock speeds of 675 MHz core, 1350 MHz shader (CUDA cores), and 900 MHz (or 3600 MHz effective) memory. The GTX 460 768 MB is expected to launch next month at a price of US $230.
Source:
PCinLife
The PCB is black, though a green PCB cannot be written off given the product's positioning. There are traces for eight memory chips on the card (looking at the components on the reverse-side of the PCB), confirming a 256-bit wide memory interface, though six chips are occupied (indicating that for this SKU only, a 192-bit wide memory interface is used. There is only one SLI finger showing that it only supports 2-way SLI multi-GPU standard. Connectivity on the rear panel is consists of the usual 2x DVI-D and mini-HDMI. Power is drawn in by two 6-pin PCI-E power inputs. Other specifications include DirectX 11 compliance, 336 CUDA cores, 768 MB of 192-bit GDDR5 memory (or another SKU with 1 GB of 256-bit GDDR5 memory), and clock speeds of 675 MHz core, 1350 MHz shader (CUDA cores), and 900 MHz (or 3600 MHz effective) memory. The GTX 460 768 MB is expected to launch next month at a price of US $230.
63 Comments on NVIDIA GeForce GTX 460 Reference Design Pictured
GTX480 replaces GTX295(single or dual don't matter, power consumption matters)
GTX470 replaces GTX285
GTX265 replaces GTX275
GTX460 replaces GTX260
in each case the cards are faster
Anyway, at 675Mhz it won't matter too much if it has 4 ROPs less (24 vs 28). 28 is 16% higher than 24 and 675 Mhz is 17% higher clock. The two differences cancel each other. Same raster power and if anything the advantage goes for the 460 because of the higher clock (higher availability).
And memory bandwidth... except for extreme cases memory bandwidth has been irrelevant since 2007. And by that I mean that extreme memory overclocks (25%+ OC) barely offer a 2-3% advantage in real performance. A GTX260 on DX10 and a GTS250 on DX9 perform exactly the same and both look the same. Make the trade, the 8800GT will be a better PhysX card...:D
now... a 260 retailed for about $240 Nov-Dec of 2008 ... which is why this card is a fail. I mean srsly... same price, same power consumption, same performance... its basically a really power-hungry 5770 with less ram.
Right now, as many of us know who have been PC gamers for awhile, most PC games are being dictated by these two video cards:
Xenos GPU specs
* Custom ATI Graphics Processor
* 500 MHz
* 48 Dynamically Scheduled Shader Pipelines
* Unified Shader Architecture
* 500 Million Triangles per Second
* 48 Billion Shader Operations per Second
* 16 Gigasamples per Second Fill Rate (4x MSAA)
* 10MB Embedded DRAM with 256 GB/s Memory Bandwidth
en.wikipedia.org/wiki/Xenos_%28graphics_chip%29
NVIDIA Reality Synthesizer (RSX) GPU
* 500 MHz
* 24 Pixel Shader Pipelines
* 8 Vertex Shader Pipelines
* 8 Pixel Rendering Pipelines (Raster Units)
* 1.1 Billion Vertices per Second
* 250 Million Triangles per Second
* 12 Gigatexels per Second
* 8 Gigasamples per Second Fill Rate (2x MSAA)
en.wikipedia.org/wiki/RSX_%27Reality_Synthesizer%27
In both cases these little gems are only capable of running DirectX 9.0c level graphics. Now, obviously, there are some exceptions to this but that is all they are really. Until the consoles are able to implement DX10/DX11 level graphics this is pretty much going to be the situation we are in where we get a handful of titles that have these features while the majority of them are nothing more that ports of a title released on the console.
Myself, for various different reasons, I'm slowly moving towards a console for my gaming needs since I'm having a more and more diffcult time justifying spending the levels of money I am on a system that I get like maybe 3 games a year if I'm lucky that make it worth it. When I consider all of the other firearm related goodies I could be getting with that money it's becoming more and more attractive as I consider my next upgrade and downsizing a bit on it. :D
Consoles technically have beaten PC gaming not by being better than PCs, rather by being too crap to keep up with PCs, making PC games no hard enough of modern systems. Crysis 2 will fail. Hard.
On top of that, I still have the impression that GF104 could be seen more like a GT200 with 50% more shaders than a crippled GF100. The rumor mill says that Nvidia calls GF104 the next 8800GT and that partners are eagerly awaiting it as the messiah. I don't think the chip that is being picturedin this thread would arise those feelings. Don't let the GTX460 name fool any of you, names can change almost 1 day before launch. And if you had (conspiracy theory ahead :laugh:):
1- a chip that went wrong like GF100 and that was delayed
2- enough time to make a redesign. remember that the 8800gt-like redesign has always been the plan. it was with G92, it was with the cancelled GT200 refresh and it's always been with GF104. the work on the refresh starts months if not years prior to the launch of the original chip.
3- a competitor that seems to be one step ahead of you.
wouldn't you:
1- work on a slightly different architecture: we at least know (rumors, granted) that GF104 is based on 24 SP clusters and has more TMUs per cluster.
2- hide it from your competitor
3- play with the names: pretty much like... how do we name the second Fermi SKU? 460 as always. wait no name it 470! 460 is now 470? yes but there's another card and it's 465? 465!!?? why not 460? because there's a 460, in fact, this one based on another chip and it's 192bit or 256bit, so what does everybody think? clearly inferior...
But is it so clearly inferior? Well at this point it's pretty clear it will have 336 (8800GT sucessor) and 384 SPs (8800GTS sucessor) so it is going to be inferior to the full Fermi aka 480, but it doesn't have to be necessarily much slower, just like G92 was not clearly inferior to G80. That it's going to be 192 and 256 bit, we do know, but so is Cypress. GF100 based cards use very slow GDDR5 memory especially the lower SKUs. At 4800 Mhz it has enough bandwidth to be as fast as the HD5870 and if shaders work as well as in GT200 (remember that SP number per cluster went back to 24), it will also have the shader power: 50% increase on shaders over GT200 minus some inneficiencies = enough to reach the the level of Cypress* which is 25% faster than GTX285. It will mainly depend on texture units tbh. If they have been increased back to the same amount found on GT200, we might have a winner.
I know it's pure speculation and maybe of the bad one, but everybody expects the worst and I always love showing the other angle of things. Just some food for thought and discussion material.
* I said Cypress because I wanted to talk about the gap between 470 and 480, like 8800GT was between 8800GTS 640 and the GTX. I find that saying Cypress level is easier than "gap between 470 and 480", not because I want to imply it will be faster than Cypress, althoug I'm aware it's the same thing. <- I swear I understand what I mean with this.
Following Nvidia's tendency of filling gaps left by ATi, the 460 will probably perform between 5770 and 5830, being closer to the latter. Reading some specs, the 460 would have 56 TMU's, the same 470 has but certainly the ROP count (more than memory) will cripple it somehow.
Coming from Nvidia, I think the price is OK, of course it could've been cheaper but you must pay for DX11 and Nvidia's exclusives.
Well, that's my bet :cool:
so if Nvdia can't come with better performance (at least on par or better than HD 5830), then it will be hard for this card to compete with HD 5770, because of it inefficient design (more expensive, higher power consumption).
just for the simple reason that console gaming is much more cheaper and reliable than PC.
as long as Nvidia and ATI release cards with expensive prices, consoles will continue to be better and cheaper than PC.
I think that by the release of PS4 and Xbox720, PC gaming will just die.
And consoles aren't really cheaper or more reliable. You can build a PC for the same price as say a PS3 and play the games the PS3 does, and you can listen to music while playing and not need to play only select games, you have internet, Linux, anything you want and it is all for the same price as the PS3 and best of all, Sony hasn't made it how they want it. Consoles will never be better than PCs. They are killing PC gaming only by being too crap compared to PCs.
PC gaming will never die. Unless console users want to put up with TDPs of over 500W, the console will never have the performance of a desktop computer, and there are plenty of gamers out there who will agree with me that they won't sacrifice functionality and performance just for something cheaper (which will undergo 5 revisions before it's deemed to be a final product).
seems i be right
I mean if im gonna blow $230, i will front the extra $50 and go with a 5850 which is much, much faster. Unless of course Benetanegia is right... we will see.
BENCHES!
u guys think i should still buy a 5850?
i want a midrange card. a 5850 costs $500 in here whereas a 5830 costs $400 in here.
If you're talking about US dollars, both cards are pretty expensive. I live in a third world country where a GTX285 still costs around US$475 so I understand you. Luckly you can buy on the internet, sometimes even with duty taxes added you can get reasonable prices.
But if your only options are those you give then I'd say you should think about your gaming needs first. If you play or intend to play games in DX10 mode mostly, you could get a 4890 or a GTX275, probably if you find them, they could be cheaper than current generation and they'll deliver a very good performance in DX10. If you game using low resolutions, let's say 1280x800 for example, you could buy a 5770 and surely you'll be satisfied with that. But if you play beyond 1680x1050, you want to give DX11 a try and 5850/5830 are your options, you should take into consideration this:
Normally, a 5850 is 25% faster than 5830 then if we begin with 5830 price at $400, the logic price result for a 5850 would be $500 (400*1.25). In your specific case, both cards are adequately positioned in terms of price/performance. Therefore the choice is indistinct, you should buy a 5850 if you have the money and want to get more power.
how about a heavily overclocked 5830? i heared that some versions of 5830 can overclock to 1ghz core and can get very close to a stock 5850.
if that would be the case then what???
and i want to game at 1600*1200 res with 4AA and no AF.
I think 5830 is still expensive and 5850 is too expensive. at least in some countries other than USA. i dont know why.
But if ATI and Nvidia dont address their high and bad pricing issues, they will face financian disaster.
Not everyone can afford to buy a $500 card. only a very few people who dont value the money.
its possible to OC 5830 over 1GHZ on core.