The GeForce GTX 275’s Inner Workings
As mentioned on the page prior, Nvidia’s GeForce GTX 275 is most easily thought of as half of a GeForce GTX 295. It employs the same 55 nm version of the GT200 graphics processor and includes 240 shader cores and 80 texture address/filtering units, just like the higher-end GeForce GTX 285.
Nvidia differentiates the two single-GPU solutions by disabling one of the GTX 275’s ROP/frame buffer partitions, yielding a total of 28 ROPs and an aggregate 448-bit pathway to 896 MB of GDDR3 memory. Given the timing and power improvements that we discussed with Nvidia when it launched GTX 295, we’d expect this new board to consequently use less power than the GTX 285. But as we’ll see in the power consumption chart, that’s that not necessarily an absolute. One more hint that this card centers on the 55 nm chip, though: whereas the 65 nm GeForce GTX 280 required one eight-pin and one six-pin power connector, the GTX 275 (like the 285) gets by with two six-pin PCIe connectors instead.
Clock speeds on the GeForce GTX 275 are down just slightly from what you’d find on a GTX 285, but higher than the same components on a GTX 295 board, presumably as a result of only needing to cool one GPU and 896 MB of memory instead of two chips plus 1,792 MB. The GeForce GTX 275’s core runs at 633 MHz (versus GTX 285’s 648 MHz). Its shader clock is set to 1,404 MHz (versus 1,476 MHz). And its GDDR3 memory modules run at 1,134 MHz (compared to 1,242 MHz on a stock GTX 285).
| GeForce GTX 285 | GeForce GTX 275 | GeForce GTX 260 Core 216 | Radeon HD 4890 | Radeon HD 4870 | |
|---|---|---|---|---|---|
| Manufacturing Process | 55 nm TSMC | 55 nm TSMC | 55 / 65 nm TSMC | 55 nm TSMC | 55 nm TSMC |
| SPs | 240 | 240 | 216 | 800 | 800 |
| Core Clock | 648 MHz | 633 MHz | 576 MHz | 850 MHz | 750 MHz |
| Shader Clock | 1,476 MHz | 1,404 MHz | 1,242 MHz | 850 MHz | 750 MHz |
| Memory Clock | 1,242 MHz GDDR3 | 1,134 MHz GDDR3 | 999 MHz GDDR3 | 975 MHz GDDR5 | 900 MHz GDDR5 |
| Frame Buffer | 1 GB | 896 MB | 896 MB | 1 GB | 1 GB / 512 MB |
| Memory Bus Width | 512-bit | 448-bit | 448-bit | 256-bit | 256-bit |
| ROPs | 32 | 28 | 28 | 16 | 16 |
| Price | $340 | ~$249 | $180 | ~$249 | $180 |
A Familiar Form Factor
While a GeForce GTX 285 gives you two dual-link DVI and one component output, The GeForce GTX 275 is limited to the two digital connectors—fine by us really, given the growing prevalence of HDMI.
The card itself should look very familiar. It’s based on the same 10.5” PCB design used by Nvidia’s other dual-slot boards, like the GTX 260 Core 216 and GTX 285. While each of the three cards employs slight layout variations, system builders must certainly appreciate the somewhat-standardized dimensions.
It’s also worth noting that, even under load, the GTX 275’s blower is remarkably quiet—quieter on our test bench than the 120 mm fan on our Thermalright Ultra 120 Extreme CPU cooler, in fact.
I expected the 4890 to be faster
I'd take the benchmarks with a pinch of salt...kinda get the feeling they're pro nVidia here!
http://www.theinquirer.net/inquire [...] cal-gt275s
The main question is if the boards the reviewers get are going to be different to the ones that appear on sale. Or is the Inq just having a go at nVidia because they don't like them at the moment.
yeah im gona wait a couple of weeks after the wide release before making my mind up. by then should have more of an idea on price and performance, and hopefully the 185 drivers will be fully released.
then i shall decide on which card to get
Anandtech has the 4890 beating the 275 in 7/8 benchmarks. It's just getting to be farcical now tbh.
Can't we go back to the good old days of naming conventions on nVidia cards (such as the Geforce 2) where the low end part was the MX, then the Geforce 2, then the GTS & then the Ultra? Now there are so many model names, memory configurations & die sizes, I'm suprised even they know which card is best any more!
http://www.theinquirer.net/inquire [...] cal-gt275sThe main question is if the boards the reviewers get are going to be different to the ones that appear on sale. Or is the Inq just having a go at nVidia because they don't like them at the moment.
Charlie is definitely an ATI fan, have a read of more of his articles and you'll see that. It's a shame that there are not many sites that are neutral when it comes to hardware reviews.
www.novatech.co.uk already have loads of both cards on sale. They were available from midnight. How come all the review sites apart rom this one show the radeon 4890 to be faster in most benchmarks, but this is the opposite way around?
Can't we go back to the good old days of naming conventions on nVidia cards (such as the Geforce 2) where the low end part was the MX, then the Geforce 2, then the GTS & then the Ultra? Now there are so many model names, memory configurations & die sizes, I'm surprised even they know which card is best any more!
I agree, David. They should offer degree courses in this stuff. The honours would go to the those who summise that just because a card has a higher model number doesn't necessarily mean it's better. Unfortunately by the time those people are ready to collect their degrees they'll be old hat since by the time the ink's dried on the certificate there'll be yet another set of cards to decipher
ATI needs to push some of the developers with games like Nvidia are doing to get the scene moving.
each to their own with games, but i still get 90+ fps with 1x gtx 260 on cod 4 & 5 @ 1920, so what's the point in buying a newer card? if you wanna keep up with the "Jones's" then you are a fool with $$ or ££'s to waste.
We need some new games...games that are gonna push graphics to the limit...make us want, sorry, NEED to spend money on new hardware.
Bottom line is NVidia can't produce these and make a profit so the halo effect of a higher end card will not do them any good when they can't produce them in volume (won't in fact either as they lose money on them)and ATI will just drop the price on their gpu anyway.
NVidia are stuck with a die too big to OC, and too expensive to produce in volume unless they are massively faster ... which they are not.
Goodbye NVidia ... IBM or Intel are about to buy you.
I like how they stress the importance that the drivers are still in beta here, but on the 4890 review they don't mention it at all.
and to quote:
"Update: In fact, it looks like ATI is aiming for $220 with mail-in-rebates, which gets us a little closer), we’d be much more likely to step up from the Radeon HD 4870 1 GB to a retail HD 4890 and try our hand overclocking the 4890 even farther."
However for this review they didn't overclock it at all. They overclocked it a titchy wee bit in its own review.
Here though theis implies that they are running an overclocked card against a stock Nvidia card which is simply not the truth.
'I expected the 4890 to be faster'
Hahaha...confused? That is right, m8. This review is a wrong call.
The article is comparing a ATI's 4890 to Nvidia's 260 and 285 by using a driver for the new 4890 that actually does not support fully this card and on the other side a well updated and well tested driver for the Nvidias' counterparts.
That is why the conclusions about the cards' performance are ill founded.
Wait until the release of the new catalyst 9.4 and you will see that 4890 does not only kill NV 280, but also NV285 at this price range.
I'm sorry, but I think this is something you should look into:
DirectX 10.1 in Stalker - you say, that "all are running DX10 to keep it consistent", but which player would play a game at dx10, if there's an option to improve performance with dx10.1? therefore the whole comparison is pointless for me.
Same point with Physx - you should put the results in the same table as the rest of the results when a game supports physx, so people get a better overview, not just some artificial results that don't tell the whole story.