GTX460 "review"

[thilanliyan]

http://www.xtremesystems.org/forums/showthread.php?t=255160

Wasn't this card supposed to be faster than what's shown there?

EDIT: I must have been thinking of the 1GB version.

 

[Zap]

Just more reason to keep using my 200 series cards. That, and being unemployed.

 

[GaiaHunter]

I will wait for official reviews, but if these numbers are true than it would be consistent with the loss of stuff like ROPs or texture units.

That was what makes the 5830 such a disappointing card - half rops compared to the 5850 and less 16 texture units - I remember reading something that the GTX 460 could have the same problem.

And the data is for the 768MB version so the 1GB should be faster - additionally most of those games are benched at 8xAA.

So only a few days to go though.

 

[GodisanAtheist]

Give it a driver revision before you really judge it too harshly. Slower than a 5830 is kind of an accomplishment in and of itself, to be honest...

 

[Unkle_Tar]

http://www.xtremesystems.org/forums/showthread.php?t=255160

Wasn't this card supposed to be faster than what's shown there?

Just behind the 465 if im reading the graphs right.

 

[railven]

Give it a driver revision before you really judge it too harshly. Slower than a 5830 is kind of an accomplishment in and of itself, to be honest...

What would a driver revision do for it that the current drivers haven't done?

Is it because it is designed differently versus the other GTXs? And if so, does that mean these will use a different driver version than the other GTXs?

Curious, this might be the nVidia I get for my GF as she is hardcore Green. I hope the official reviews are better. I also wish this leak had TPW and Heat benches.

 

[EarthwormJim]

Give it a driver revision before you really judge it too harshly. Slower than a 5830 is kind of an accomplishment in and of itself, to be honest...

It's faster than the 5830 in the majority of the tests. It's really just Unigine that it is really slow. More than likely a driver bug, it should not be half the speed of a GTX465.

What would a driver revision do for it that the current drivers haven't done?

Is it because it is designed differently versus the other GTXs? And if so, does that mean these will use a different driver version than the other GTXs?

Curious, this might be the nVidia I get for my GF as she is hardcore Green. I hope the official reviews are better. I also wish this leak had TPW and Heat benches.

I'm sure there's optimizations specific to every single Nvidia GPU. We don't get to see exactly what is inside each core, just block diagrams pretty much. So while this core seems to have the same building blocks as the other GTX4xx series, it might not be as simple as being cut down pieces.

It's also possible that optimizations that apply to the GTX4xx series are not being enabled with this new card and it's new GPU yet.


I want to see how well this card overclocks with it's new core, and how well it scales despite it's constraint on memory bandwidth.

 

[dookulooku]

For a card that supposedly has a significant bandwidth disadvantage against every other card it's benchmarked against there, those are pretty good results, especially when you factor in the bandwidth-hungry 8xAA that's being used in a number of those tests.

 

[RussianSensation]

http://www.xtremesystems.org/forums/showthread.php?t=255160

Wasn't this card supposed to be faster than what's shown there?

I think everyone was talking about the 1GB, 256-bit version of the card The one tested is 768mb + 192-bit.

I am excited about overclocking results for the 1GB version @ $230 USD.

 

[dookulooku]

It's faster than the 5830 in the majority of the tests. It's really just Unigine that it is really slow. More than likely a driver bug, it should not be half the speed of a GTX465.

8xAA/16xAF is pretty memory intensive, so it could be 768 MB not being enough.

If the prices are accurate, then it looks like we might finally have a price war. All the GF100 variants were placed above the Cypress variants to avoid direct competition, probably due to GF100's high cost. GTX 460 looks to match/exceed 5830 performance, while being priced below it.

 

[EarthwormJim]

8xAA/16xAF is pretty memory intensive, so it could be 768 MB not being enough.

If the prices are accurate, then it looks like we might finally have a price war. All the GF100 variants were placed above the Cypress variants to avoid direct competition, probably due to GF100's high cost. GTX 460 looks to match/exceed 5830 performance, while being priced below it.

The 768 version is also on a 192-bit bus vs the 1gb version having a 256-bit bus. That might make a big difference at those AA/AF levels.

It looks like these cards overclock really well too.

http://translate.google.com.br/tran...en-der-GeForce-GTX-460-enthuellt-1026888.html

"without increasing the core voltage with 830 MHz still stable, with a slight voltage boost 900 MHz were even possible."

 

[dookulooku]

I will wait for official reviews, but if these numbers are true than it would be consistent with the loss of stuff like ROPs or texture units.

That was what makes the 5830 such a disappointing card - half rops compared to the 5850 and less 16 texture units - I remember reading something that the GTX 460 could have the same problem.

The total number of texture units is not as important as the number of shader units, with the RATIO of texture to shader units taken into account. A GPU with a ton of texture units but few shader units is going to be dog slow. A GPU with a ton of shader units will be bottlenecked by few texture units, but will be better than the first GPU.

5830 loses 16 texture units compared to 5850, but the ratio doesn't change, and the 5830 is clocked higher. The performance loss from 16 fewer texture units and 320 fewer shaders is only about 10%.

I think everyone's opinions will differ, but I would rather cut ROPs than shader and texture units. Of course the 5830 loses all of those compared to the 5850/5870, but it takes a disproportionately large reduction in ROPs.

Losing ROPs just means performance in old games at high resolution and high AA settings won't be that great. The affect on newer games will be smaller since shader and texture limitations would come into play before ROPs. I think that's a fair sacrifice.

 

[GaiaHunter]

I think everyone's opinions will differ, but I would rather cut ROPs than shader and texture units. Of course the 5830 loses all of those compared to the 5850/5870, but it takes a disproportionately large reduction in ROPs.

Losing ROPs just means performance in old games at high resolution and high AA settings won't be that great. The affect on newer games will be smaller since shader and texture limitations would come into play before ROPs. I think that's a fair sacrifice.

I'm not sure if it works exactly that way with Cypress.

Look at 5850 - OC it to the 5870 clock speeds and you lose like 3%, and you have the same rops but only 90% of the texture units and 90% of the shader power.

Look at 5970 vs 5850 CF - same story, 90% of the shader power, 90% of the texture units but pretty much the same performance.

Look at 5830 vs 4890/4870/5770 - more texture units (40% more), more shaders (42.5% more), more bandwidth same rops and the performance isn't that much different (I know that many people think that the 5000 series shaders are weaker vs 4000 series shaders, but please bear in mind that 5000 series has better IQ quality and that probably costs that performance amount - read BFG10K articles on AF/AA of 5000 series).

I think less ROPs is what kills 5830 performance.

 

[ArchAngel777]

Give it a driver revision before you really judge it too harshly. Slower than a 5830 is kind of an accomplishment in and of itself, to be honest...

I am getting tired of hearing this statement. It isn't like the 460 is a new arch like when the 480 launched. This is based on the same platform, so drivers really are not going to do anything at all.

However, I will say that waiting for a more official review would be wise... But blaming the performance on drivers is a last ditch attempt at anyone trying to explain away crappy performance. And to be fair, if it is priced right, this will be just fine as far as performance goes. This version of the card will need to be priced around $179 if they want to sell them, though. The 1GB/256 probably will need to be priced around $199. The 5830 is an overpriced piece of turd... Bring that thing down to the $179 range too.

 

[dookulooku]

I'm not sure if it works exactly that way with Cypress.

Look at 5850 - OC it to the 5870 clock speeds and you lose like 3%, and you have the same rops but only 90% of the texture units and 90% of the shader power.

Look at 5970 vs 5850 CF - same story, 90% of the shader power, 90% of the texture units but pretty much the same performance.

Look at 5830 vs 4890/4870/5770 - more texture units (40% more), more shaders (42.5% more), more bandwidth same rops and the performance isn't that much different (I know that many people think that the 5000 series shaders are weaker vs 4000 series shaders, but please bear in mind that 5000 series has better IQ quality and that probably costs that performance amount - read BFG10K articles on AF/AA of 5000 series).

I think less ROPs is what kills 5830 performance.

lol, I'm not trying to be mean to you Gaia, but you just contradicted part of what you said earlier about 16 fewer texture being the reason why the 5830 was a disappointment.

Anyways, if it is less ROPs that hurts performance, then it's likely that "ROP" is a big simplification of what's really being cut out. Both the 58/57xx GPUs may be heavily reliant on ROPs to achieve maximum performance.

 

[GaiaHunter]

lol, I'm not trying to be mean to you Gaia, but you just contradicted part of what you said earlier about 16 fewer texture being the reason why the 5830 was a disappointment.

Those texture units appeared in there more to explain a probable reason of bad performance on the GTX 460 - less rops and texture units - I don't really know if is only texture units or rops that limits GXT 460 (plus the obvious bandwidth and memory difference) performance, alongside sps - and all these seem interconnected in fermi - so basically as you chop(fuse) SMs off, all these can potentially create a big drop.

For me it seems evident that amongst the 5000 series it is the ROPs and core clock that matter more, although cutting the rest doesn't help either. Fermi, whoever is a different architecture, so can be limited in a different way.

Sorry for the less than then clear point on my part (and I'm not trying to be cocky or act as a victim with this sentence).

 

[dookulooku]

Those texture units appeared in there more to explain a probable reason of bad performance on the GTX 460 - less rops and texture units - I don't really know if is only texture units or rops that limits GXT 460 (plus the obvious bandwidth and memory difference) performance, alongside sps - and all these seem interconnected in fermi - so basically as you chop(fuse) SMs off, all these can potentially create a big drop.

For me it seems evident that amongst the 5000/4000 series it is the ROPs that matter more, although the rest doesn't help either. Fermi, whoever is a different architecture, so can be limited in a different way.

Sorry for the less than then clear point on my part (and I'm not trying to be cocky or act as a victim with this sentence).

Ok understood. And I don't think the 460 looks that bad in that link, outside of the Unigine Heaven benchmark, which is likely a limitation of the 768 MB. And you don't get a great picture of performance when benchmarking at 8xAA -- it just favors cards that have the most memory bandwidth and pixel fillrate. The results could be very different at the more common 4xAA setting.

 

[GaiaHunter]

Ok understood. And I don't think the 460 looks that bad in that link, outside of the Unigine Heaven benchmark, which is likely a limitation of the 768 MB.

Basically loses some/win some compared to the GTX 275.

Now the curious is that the 5830 lose some/win some vs the 4890..

Even more curious it was a wash between the 4890 and the GTX 275.

But now the 5830 that is supposed to be, give or take, the performance of a 4890 gets trashed?

So I'll just wait for more reviews (if you can call this one a review) and more games/resolutions/ AA settings.

 

[Lonyo]

I hope some site does some testing on something that isn't a 4GHz Core i7 CPU, like maybe the sort of CPU someone buying a $200 graphics card might use. It would be nice to see some real world comparisons between all these cards.

 

[nemesismk2]

I am waiting for reviews from sites like Anandtech, Tech Report etc for a decent GeForce GTX 460 review.

 

[tviceman]

http://www.xtremesystems.org/forums/showthread.php?t=255160

Wasn't this card supposed to be faster than what's shown there?

If those results hold true in the majority of benchmarks, it's actually a pretty damn good card vs. it's immediate competition. Of course, how much it prices out will determine how good the card is, but less ram and memory bandwidth than a 5830 and still generally outperforming it bodes really, really well for the 1 gig 256-bit version.

 

[formulav8]

I wouldn't really expect general performance to improve much with the wider memory bus. 192 bit with DDR5 memory provides a huge amount of bandwith compared to the basic specs of this card. AA and such may end up improving though from a wider path. The jump from 768mb to 1gb wouldn't make much of a difference either if they run at similar clockspeeds. Just my opinion of course...


Jason

 

[sanzen07]

Will this card be faster than my 9800 GTX with the configuration in my signature?

 

[Russwinters]

It looks like the 768 w/o any OC will be roughly GTX 275/280 performance depending on the game, etc.

The new architecture seems to scale really well with OC; so if your comfortable with OCing you may be able to squeeze some pretty good performance out of this thing for the price.


I am much more interested in the 1GB/256bit version.

 

[toyota]

Will this card be faster than my 9800 GTX with the configuration in my signature?

in most cases yes but if your cpu is at stock 2.4 then it will severally limit any gpu upgrade you make.