Some general questions about PhysX cards

T

Tuf

Limp Gawd
Joined
May 18, 2008
Messages
491
I was considering picking up an inexpensive card for PhysX processing since I have a few games that offer it (Batman AA, Batman AC, Metro 2033, Mafia2). Which card specs have greater effects than others? Processor cores, core clock, memory bus, etc? I saw a GT 430 on sale at newegg but it's for a 64 bit memory bus card instead of 128 and am not sure how much that matters. And then there is the 450 with 192 stream processors ...

I realize that there are several variables that will recommend a different solution but I'm just curious where a sweet spot would be for a GTX 580.

I know the future of the technology is questionable and not many games use it but I really do enjoy the effects and the games that offer it so if it would be worth investing a few bucks to enhance the effect and keep frames reasonably high.
 
hmm... based on my experience, i once had a 9800gt paired with a gtx285, and the combo works well...
and then i upgrade the gtx285 to a 6990 (using hybrid physx) and the 9800gt can work well and not bottlenecking the 6990....

a few months later i had a chance of a very cheap 2nd hand gtx460, and so i upgrade my 9800gt with a gtx460 and i do experience a slight increase in fps in my physx enabled game (roughly about 10fps - 15fps)

so overall, i think anything above 9800gt will work well with your gt580, try getting a 2nd hand gtx460, i got mine for only less then around $75 :)
 
Thanks, I figured I would either grab a used card or get a 4xx from the Egg if a good deal pops up .... it's not like I'm missing out on very much contend lol.
 
The innate flaw with PhysX dedicated GPU is that there is a 1:1 relationship between it and the GPU. There is no such thing as an 'inexpensive PhysX card', especially if you have a modern GPU. Unfortunately there is likely no solution. A slow PhysX card will bottleneck your entire system and in many cases running the PhysX directly on the primary GPU is faster than using the older slow card.

Basically the way it's designed is if you upgrade your GPU you need to upgrade your PhysX card or else you're not going to get the full potential of your GPU. If a person is willing to not get the full potential of their GPU, rather than having two cards, you might as well run it directly on the GPU and save yourself the slot, and PSU requirements.
 
OK, thanks. I will probably just end up putting another 580 in at some point and not worrying about the PhysX card.
 
bigdogchris said:
The innate flaw with PhysX dedicated GPU is that there is a 1:1 relationship between it and the GPU. There is no such thing as an 'inexpensive PhysX card', especially if you have a modern GPU. Unfortunately there is likely no solution. A slow PhysX card will bottleneck your entire system and in many cases running the PhysX directly on the primary GPU is faster than using the older slow card.
Click to expand...

I don't know man, a GT 430 doesn't seem to bottleneck a Radeon HD 6970 much at all: http://hardforum.com/showthread.php?t=1645682

The largest gain seen was in Batman Arkham Asylum. Jumping up from a GT 430 to a GTX 460 as the dedicated PhysX card only gained them 5 FPS... not much of a difference at all, the GT 430 was obviously keeping up just fine.

The GT 430 is a pretty cheap card. You could go slightly higher-end and assure that you're covered without spending a whole lot.
 
Unknown-One said:
I don't know man, a GT 430 doesn't seem to bottleneck a Radeon HD 6970 much at all: http://hardforum.com/showthread.php?t=1645682

The largest gain seen was in Batman Arkham Asylum. Jumping up from a GT 430 to a GTX 460 as the dedicated PhysX card only gained them 5 FPS... not much of a difference at all, the GT 430 was obviously keeping up just fine.

The GT 430 is a pretty cheap card. You could go slightly higher-end and assure that you're covered without spending a whole lot.
Click to expand...
um thats with an AMD card so you are comparing it to the cpu having to do the physx. if you had a gtx580 then it could handle BOTH graphics and physx just as well if not better than trying to use a gt430 for physx.

with a gtx580 card for graphics, you will want at least gts450 and preferably a gtx460 to handle the physx. heck in Batman Arkham City they recommend a gtx460 for physx while using a gtx570 for graphics.
 
You seem to be under the false assumption that AMD cards can only use software PhysX. The person in that thread has patched his Nvidia drivers to allow the use of a GT 430 as a dedicated PhysX card for his HD 6970.

Might want to read again, you're missing the point a bit...The comparison is between a slow PhysX card and a fast PhysX card with a VERY fast rendering card. The slow PhysX card (GT430) barely held back a card as fast as the HD 6970, and there was no worthwhile gain from upgrading the dedicated PhysX card to something as ridiculous as a GTX 460 (maximum gain was an additional 5 FPS).

1. We know that A HD 6970 is a bit faster than a GTX 480.
2. We know that A GT 430 performed almost identically to a GTX 460 as a dedicated PhysX card for a HD 6970.
3. Therefor, a GT 430 did not bottleneck a HD 6970, and would certainly not bottleneck a GTX 480.
4. Therefor, a GT 430 could also be used as a dedicated PhysX card for a GTX 480 without causing a bottleneck.

Conclusion: You don't need an uber-expensive PhysX card even if you have an uber-expensive rendering card. Fairly cheap ones seem to be able to keep up with HD6970 / GTX480 class GPU's without a problem.

With a GTX580 as the rendering card, I'd probably pick a GT 440 as a dedicated PhysX card, just to be future proof :)
 
Last edited:
sigh, you still do get it and have no idea what you are talking about. AGAIN with an AMD card, just about any card for physx will help because the other option is the cpu. things are completely different when using an Nvidia card. a gt430 will most certainly not be a good physx card while using a higher end Nvidia card. with something like a gtx480 or faster it could actually bring performance down compared to letting the gtx480 or faster card handle both graphics are physx.

and please stop talking out of your butt as a gt440 would be little to no better than letting a gtx580 handle both graphics and physx.

EDIT: maybe this will finally make you understand. see the gt240 as physx is hardly any faster than the gtx480 doing both graphics and physx. a gt240 is 20% faster than a gt430 so maybe now you will get it. if that was a gtx580 for the main card than the gt240 would be lucky to even break even. and if a gt240 only breaks even then what do you think the slower gt430 will do? well it will be slower than letting the gtx580 do both graphics and physx. and btw a gt440 is also slightly slower than a gt240 so it would basically be useless as a physx card compared to just letting a gtx580 do both graphics and physx.



Uploaded with ImageShack.us
 
Last edited:
cannondale06 said:
sigh, you still do get it and have no idea what you are talking about. AGAIN with an AMD card, just about any card for physx will help because the other option is the cpu. things are completely different when using an Nvidia card. a gt430 will most certainly not be a good physx card while using a higher end Nvidia card. with something like a gtx480 or faster it could actually bring performance down compared to letting the gtx480 or faster card handle both graphics are physx.
Click to expand...
Nope... you're still not getting it...being an AMD card doesn't matter in my comparison because I'm not comparing against an Nvidia card that's handing both graphics and PhysX at the same time.

Scenario 1a and 1b:
An HD 6970 with either a GT 430 or a GTX 460 dedicated to PhysX.

Scenario 2a and 2b:
A GTX 480 with either a GT 430 or a GTX 460 dedicated to PhysX.

I can guarantee you that upgrading either system's dedicated PhysX card from a GT 430 to a GTX 460 would not give you any appreciable gain in minimum framerate (because the GT430 isn't really a bottleneck).

Edit: and a GTX 480 doing both graphics and PhysX at once would be slower than all of the above.

cannondale06 said:
and please stop talking out of your butt as a gt440 would be little to no better than letting a gtx580 handle both graphics and physx.
Click to expand...
There's no need to be rude. And I'm quite sure a GT 440 would be sufficient as a dedicated PhysX card for a GTX580.

Edit: See the table below. The GT 440 should be a much faster PhysX card than the GT 240 could ever hope to be.

cannondale06 said:
EDIT: maybe this will finally make you understand. see the gt240 as physx is hardly any faster than the gtx480 doing both graphics and physx. a gt240 is 20% faster than a gt430 so maybe now you will get it. if that was a gtx580 for the main card than the gt240 would be lucky to even break even. and if a gt240 only breaks even then what do you think the slower gt430 will do? well it will be slower than letting the gtx580 do both graphics and physx. and btw a gt440 is also slightly slower than a gt240 so it would basically be useless as a physx card compared to just letting a gtx580 do both graphics and physx.



Uploaded with ImageShack.us
Click to expand...
I'm not sure what you're trying to point out, because your graph backs up what I've been saying...

First of all, you're making a false assumption about the performance of the GT 240 and the GT 430. The GT 240 is a faster rendering card than the GT 430, but the GT430 whips the GT 240 when it comes to PhysX calculations. Here's the reason why:

GT 240:
Stream processors: 96
Special Function Units: 6
Core clock: 550 MHz
Shader Clock: 1340 MHz

GT 430:
Stream processors: 96
Special Function Units: 12
Core clock: 700 MHz
Shader Clock: 1400 MHz

Edit: (Throwing this in for good measure) GT440:
Stream processors: 96
Special Function Units: 12
Core clock: 810 MHz
Shader Clock: 1620 MHz

CUDA and PhysX care about the number of stream processors (cores) on the card, the efficiency of those cores, and the clockspeed of those cores. Memory type and bandwidth shouldn't have much effect on PhysX performance at all. The GT430 has the same number of cores as the GT240, but they're on a newer revision of the architecture, and running at a higher clockspeed... so the GT 430 ends up as a faster PhysX card than the GT 240.


Now, with that out of the way, back to your graph. The first thing I notice, even adding a dedicated GT240 is better than trying to let the GTX 480 handle graphics and PhysX at the same time.
You claimed earlier that such a small dedicated card would bring performance down, seems you were mistaken.

A GT 430 would fall fairly close to the 9800 GTX on your chart. The 9800 GTX has 128 stream processors rather than 96, but they're clocked slower and they're from a MUCH older architecture.


Now, this is the important part:
1. Look at what the 9800 GTX (our GT430 analogue) did to their framerate... it already brought the minimum up to 42 FPS.
2. This is much faster than a GTX 480 attempting to do both graphics and physics on its own.
3. Upgrading to a GTX 260 only improved minimum framerate by 2 FPS.
4. Upgrading to a GTX 275 (or anything faster) only improved minimum framerate by 5 FPS over the 9800 GTX (GT430)

As I said before, a 5 FPS boost is pretty much nothing (especially when you consider how much more all those cards cost than a GT 430), and it indicates that the GT 430 doesn't really pose much of a bottleneck to cards in the GTX 480's performance range.
 
Last edited:
look at reviews because a gt430 is 20% SLOWER than a gt240. that means the gt430 would barely break even at best if used with a gtx480 and would likely be slower than a gtx580 handling both graphics and physx. a gt440 is about 15% faster than a gt430 so that makes it still slightly slower than a gt240. again look at reviews and stop making up your own performance. its beyond foolish to suggest a gt430 or gt440 for physx card when using a high end Nivdia card. if you cant figure that out then fine but stop giving out bad advice.

the minimum card you want with a gtx580 would be a gts450 or there is no point even even bothering. and to get the most out of physx you would need at least a gtx460 while using a gtx580 as the main card. I get that and so do the game makers because that's why they recommend a gtx460 for physx while using a gtx570 as the main card in Batman AC.
 
Last edited:
cannondale06 said:
look at reviews because a gt430 is 20% SLOWER than a gt240.
Click to expand...
I already explained why your reasoning here is flawed.

Dedicated to Graphics: GT 240 > GT 430 (The GT 240 can render frames faster)
Dedicated to Physics: GT 430 > GT 240 (The GT 430 can crunch numbers faster)

It's not that hard to understand. The GT 430 has the same number of cores, they're from a newer architecture that's better tuned for computational performance, and they're running at a higher clockspeed. This ends up making the GT 430 a MUCH faster PhysX card than the GT 240.


Time for a history lesson: The GeForce 9 series, the GeForce 200 series, the GeForce 400 series, and their computational (CUDA / PhysX) performance:
- The GeForce 9 series (G92 core and its derivatives) are actually very efficient when it comes to PhysX calculations.
- The GeForce 200 series (GT200 core and its derivatives) actually de-optimized PhysX performance when compared to the GeForce 9 series.
- That's right, if you compare the PhysX performance a G92 based card and a GT200 based card with the SAME number of stream processors at the SAME clockspeed, the GT200 based card would be slower.
- The GeForce 400 series (GF100 core and its derivatives) greatly improved PhysX performance, making it more more efficient than G92, and WAY more efficient than GT200.

The GT 240 is based on the GT200 core (the GT215 to be precise), which has nerfed PhysX performance. The GT 430 is based on the GF100 core (specifically, the GF108), which has MUCH better PhysX performance with the same number of stream processors.


If you want the nitty-gritty technical reasons why GF100-based cards walk all over GT200-based cards when it comes to computational (CUDA / PhysX) performance, here you go:
- The GT 430 has dedicated L1 load/store cache, the GT 240 doesn't.
- The GT 430 has 768 KB of L2 cache with read/write access from all client types, the GT 240 only has 256 KB and it can only be used for textures.
- The GT 430's integer units support full 32-bit precision for all instructions, the GT240 only supports 32bit precision for some instructions.
- Each SM (unit of 32 CUDA cores) on the GT 430 has four SFUs (special function units), while the GT 240 only has two SFUs for every SM.
- The above means that the GT430 has 12 special function units, while the GT240 only has 6. SFUs are where a lot of PhysX calculations go on, so the GT240 is seriously hampered here.


That's why the GF100 architecture is faster than GT200... but why is the old G92 architecture faster than GT200?:
- SM units on G92 are arranged differently. Each unit of 8 stream processors has 2 SFUs, which means a card like the 9800 GTX has 32 SFUs.
- The 9800GTX might have more SFUs than the GT430, but the GT430 has additional architectural improvements (noted above) and runs the units at a much higher clockspeed.

So,what have we learned? Avoid using anything based on GT200 as a PhysX card, they suck at it, and can hardly be compared to cards based on G92 / GF100 :p


cannondale06 said:
that means the gt430 would barely break even at best if used with a gtx480 and would likely be slower than a gtx580 handling both graphics and physx.
Click to expand...
Incorrect, the GT 430 has no trouble keeping up an HD 6970 (which is faster than a GTX 480). A GT 430 wold be just fine as a dedicated PhysX card for a GTX 480.

Your own chart shows the GT 240 already improving the performance of a GTX 480. Since a GT 430 has better computational performance than a GT 240, therefor faster PhysX performance, therefor it wold improve performance still further (showing gains similar to those from using a 9800GTX as a dedicated PhysX card).

cannondale06 said:
a gt440 is about 15% faster than a gt430 so that makes it still slightly slower than a gt240. again look at reviews and stop making up your own predictions.
Click to expand...
Reviews of these cards generally only compare rendering performance. Since these cards will be doing nothing but PhysX calculations for us, we care about their computational performance.

Rendering performance does not directly equate to PhysX performance. The GT 430 and the GT 440 have higher computational performance than the GT 240, and are therefore much faster PhysX cards than the GT 240.

cannondale06 said:
its beyond foolish to suggest a gt430 or gt440 for physx card when using a high end Nivdia card. if you cant figure that out then fine but stop giving out bad advice.
Click to expand...
I'm not giving out bad advice, you are MASSIVELY underestimating the computational performance of the GT 430 and the GT 440 when they're used as dedicated PhysX cards.

Even your own chart showed that the addition of a lowly GT 240 improve the performance of a GTX 480. A GT 430 is a lot faster at PhysX calculations than a GT 240, so the boost would be even greater.

You're still under the false assumption that rendering performance = PhysX performance. You need to get that out of your head, because it simply isn't true.
 
Last edited:
you are not worth fooling with. the fact that you still keep comparing using a gt430 with an AMD card to using it with an Nvidia card shows you are just too ignorant on this topic.
 
cannondale06 said:
you are not worth fooling with. the fact that you still keep comparing using a gt430 with an AMD card to using it with an Nvidia card shows you are just too ignorant on this topic.
Click to expand...
...You're kidding right? Did you read any of the above? You've called me ignorant multiple times now (you edited it out of another post, classy...) and yet I've come up with a pile of hard facts and logical explainations, and you've brought nothing back to the table except an unfounded and un-backed disagreement.


Anyway... is the fact that I used an AMD card as the rendering card in some exmaples your ONLY remaining point of contention? I've explained multiple times why using an AMD card as the rendering card doesn't matter here.

The primary rendering card is ONLY doing rendering, all PhysX calculations in every single example I've offered are being handled by a dedicated PhysX card.
Replace the HD 6970 with a GTX 480 in my examples... nothing changes. The primary rendering card is still ONLY doing rendering, and all PhysX calculations are STILL being handled by a dedicated PhysX card.

(You do gain the additional option of running both graphics and PhysX on the GTX480 at the same time, but even your graph shows that doing so is is slower than dedicating a GT 240 to PhysX)

The rendering card doesn't matter, all that matters is that it's a very fast rendering card (so we can tell if there's a bottleneck when using these low-end cards as PhysX cards).


Do you still honestly think a GT240 has faster computational (CUDA / PhysX) performance than a GT 430? If so, you've crossed into delusion.
 
Last edited:
I will give you that the gt430 may be better at physx than the gt240.

being an AMD or Nvidia card DOES matter though. I am tired of saying this but when you use an AMD card your only choice for using physx is on the cpu or a dedicated card. with Nvidia, the level of dedicated card you need is based on the main card you are using for graphics. thats why Nvidia does not even make a physx card. its always a moving target so whats good for one user might barely get the job done for someone else.

just like in that graph where the 8600gt for physx is SLOWER than the gtx480 doing both graphics and physx. if that was a gtx260 for the main gpu then the 8600gt for physx would have broke even or helped by a couple of fps.

you can also see that a gtx275 was needed to get good usage for physx while having a gtx480. well a gtx275 is a much much faster card than the gt430 or gt440.

with a gtx580, a gt430 would be of almost no use at all because the gtx580 can handle both graphics and physx as well as trying to just offload to a gt430. a gt440 is basically the same card so results will be within 15% on that too. sure a gt430 would help while using a gtx560 but going to a gtx580 would wipe out what a gt430 can do.

so again recommending a gt430 or gt440 for physx while using a gtx580 for the main card is foolish. at best that would only give you a frame or two better performance. you would want at least a gts450 to get a worthwhile difference and a gtx460 or better to get the most out of physx for that setup.
 
Last edited:
cannondale06 said:
I will give you that the gt430 may be better at physx than the gt240.
Click to expand...
MAY be faster? It has double the SFUs running at higher clockspeed... it's faster for PhysX... a lot faster.

cannondale06 said:
being an AMD or Nvidia card DOES matter though. I am tired of saying this but when you use an AMD card your only choice for using physx is on the cpu or a dedicated card. with Nvidia, the level of dedicated card you need is based on the main card you are using for graphics. thats why Nvidia does not even make a physx card. its always a moving target so whats good for one user might barely get the job done for someone else.
Click to expand...
As I said, if you're using a dedicated PhysX card, then the rendering card being AMD or Nvidia doesn't matter.

The ONLY difference is that, when using an Nvidia card, you also have the option of running both Graphics and PhysX on your primary card... but that's so slow, that almost ANY (even the GT 240) dedicated card improves performance. It basically makes this option worthless.

So...we're right back to a rendering card + dedicated PhysX card. When using a dedicated PhysX card, the rendering card has no bearing on the performance of the PhysX card, so we're all good.


cannondale06 said:
just like in that graph where the 8600gt for physx is SLOWER than the gtx480 doing both graphics and physx. if that was a gtx260 for the main gpu then the 8600gt for physx would have broke even or helped by a couple of fps.
Click to expand...
Yes... and if I used a slower AMD card, then an 8600 GT would have worked fine as a dedicated PhysX card for it too. Same thing...

cannondale06 said:
You can also see that a gtx275 was needed to get good usage for physx while having a gtx480. well a gtx275 is a much much faster card than the gt430 or gt440.
Click to expand...
Uh, no it wasn't? Your graph CLEARLY shows that the difference in minimum FPS between using a 9800GTX (or GT 430) as a dedicated PhysX card, and using a GTX 275 as a dedicated PhysX card, is only 5 FPS.

That's it... 5 FPS.

You can get a GT 430 for under $50. A GTX 275 costs around $125; more than double the price of the 430... You can do the math on that, double the price for a 5 FPS boost isn't worth it.

Also, WHY would you buy a GT200 based card at all? They're terribly inefficient PhysX cards. Snag something from the GeForce 400 family, it'll perform better while running cooler...

cannondale06 said:
with a gtx580, a gt430 would be of almost no use at all because the gtx580 can handle both graphics and physx as well as trying to just offload to a gt430. a gt440 is basically the same card so results will be within 15% on that too. sure a gt430 would help while using a gtx560 but going to a gtx580 would wipe out what a gt430 can do.
Click to expand...
A GTX 580 might (might) be enough to warrant a slightly faster PhysX card, but I highly doubt it.

The GT 430 doesn't bottleneck a GTX 480, it doesn't bottleneck a HD 6970 (which is faster than a GTX480)... even if it does begin to bottleneck a GTX 580, it'll be minor, and I'm sure a GT 440 would cover the difference.

cannondale06 said:
so again recommending a gt430 or gt440 for physx while using a gtx580 for the main card is foolish. you would want at least a gts450 to get a worthwhile difference and a gtx460 or better to get the most out of physx for that setup.
Click to expand...
It isn't foolish, it'll be fine.

I've seen people make claims like yours before, and I can tell you right now, dedicating a GTX 460 to PhysX is always overkill. The only instance where it might not be overkill is if you're using it as the dedicated PhysX card for an SLI or Corssfire system.
 
Last edited:
okay for physx a gt430 may be around 9800gt performance at best.

open you eyes and look at that chart though. it takes a gtx275 to really get best use out of physx while using a gtx480 for the main card.

you know damn well a gt430 is NO WHERE NEAR as fast as the gtx275. in fact it would be slower than the 9800gtx which is alreay giving up quite a bit to the gtx275.

now use some common sense and realize that a gtx580 is about 20% faster than that gtx480 so that would mean it would need an even faster card for physx. only a fool would claim a gt430 or gt440 would make a good physx card while using a gtx580 as the main card. :rolleyes:

EDIT: and you still keep going on with the Nvidia to AMD comparisons when it does not work like that for physx. for example if you have a 6870 then going to a 6970 for your main gpu is NOT going to help with physx. if you have a gtx560 for your main gpu and upgrade to a gtx580 then you increase BOTH your graphics AND physx processing power. that is why you need a faster card for physx when you use a higher end Nvidia gpu.
 
Last edited:
cannondale06 said:
now use some common sense and realize that a gtx580 is about 20% faster than that gtx480 so that would mean it would need an even faster card for physx. only a fool would claim a gt430 or gt440 would make a good physx card while using a gtx580 as the main card. :rolleyes:

EDIT: and you still keep going on with the Nvidia to AMD comparisons when it does not work like that for physx. for example if you have a 6870 then going to a 6970 for your main gpu is NOT going to help with physx. if you have a gtx560 for your main gpu and upgrade to a gtx580 then you increase BOTH your graphics AND physx processing power. that is why you need a faster card for physx when you use a higher end Nvidia gpu.
Click to expand...

Your first statement.... false.

The amount of PhysX that the game calculates ALWAYS has a maximum. That is a part of the PhysX engine itself. Once you reach the point where PhysX has the maximum number of objects being calculated in the engine and have maximized PhysX performance, going with a faster and faster card for PhysX provides NO PERFORMANCE benefit.

On your edit post

The primary graphics processing card does NOT matter in this instance. It's a GPU in this case, that is all that matters.

If you have PhysX setup to only run on the secondary COMPUTATIONAL card, the performance of the MAIN RENDERING card has NO effect on the performance of the secondary computational card.

Fyi, you can use PhysX on an Nvidia card in an ATI main card system. It's called Hybrid PhysX, the PhysX is STILL ran on the Nvidia card and the rendering done on the ATI card. Like I said before, the primary graphics processing card does NOT matter in this instance. It's only a RENDERING GPU, increase the performance of the rendering GPU you increase your rendering capacity, NOT PhysX processing needs.

PhysX processing needs are the same in all situations. PhysX has a maximum number of objects. Once you have reached the point where the COMPUTATIONAL card can maximize performance of these objects, increasing the performance of the computational card has NO effect.
 
yes once you reach the maximum level then it does not matter. a gt430 will NOT let you reach the maximum level though. again open your eyes and look at that chart. it takes a gtx275 or better to get the most out of physx while using a gtx480 for graphics. a gt430 is a much slower card than a gtx275 and would perform below the 9800gtx in that game. with a gtx580 for the main card then a gt430 would look even worse as a physx card.
 
cannondale06 said:
open you eyes and look at that chart though. it takes a gtx275 to really get best use out of physx while using a gtx480 for the main card.
Click to expand...

I've looked at the chart, and I've told you before, it shows that using a 9800 GTX as a PhysX card is only 5 FPS slower than using a GTX 275.

Minimum FPS is what really matters. A GTX 275 will inflate maximum framerate (which also artificially inflates the average), but maximum framerate is a totally worthless figure. Minimum framerate is what impacts your gaming experience, so you want minimum to be as high as possible. The 9800 GTX in that table already gets you there...

cannondale06 said:
you know damn well a gt430 is NO WHERE NEAR as fast as the gtx275. in fact it would be slower than the 9800gtx which is alreay giving up quite a bit to the gtx275.
Click to expand...
Of course a GTX 275 is a faster rendering card than a GT 430, but in this instance, they're being used as PhysX cards, not graphics cards. You MUST take the massive charge in architecture into account. See below:

The GTX 275 has 15 Special Function Units (SFUs) running at 1404 MHz.
The GTX 260 has 12 Special Function Units (SFUs) running at 1242 MHz
The GT 430 has 12 Special Function Units (SFUs) running at 1400 MHz
The GT 440 has 12 Special Function Units (SFUs) running at 1620 MHz

a SFU-heavy workload like PhysX will favor the GT 430 / GT 440 over the GTX 260 / GTX 275.

As I keep telling you, over and over, the GeForce 200 series SUCKS at PhysX. I was being conservative when I said the GT 430 would be around the same PhysX performance as a 9800GTX, it should actually be faster. The extremely small number of SFUs on GT 200 based GPUs internally bottlenecks the rest of the graphics card when performing PhysX calculations (especially when it needs to hit cache, which doesn't work nearly as well on GT200 as it does on GF100). Comparing against cards that use the GT200 core is next to pointless because of how the hardware within the core is divided up.

With the GT 200 core, you only get two SFUs for every 32 cores. This meant that you had to get a beastly graphics card like the GTX 275 in order to get your hands on a decent number of SFUs. THIS ISN'T THE CASE ANYMORE! You can get a large number of SFUs on cards like the GT 430.

The GT430 only has 3 fewer SFUs on-board than the GTX 275, and a gigantic pile of architectural improvements that make CUDA/PhysX faster (which easily make up for the missing units). The GT440 has all of the above, and its running those cores and SFUs a LOT faster than the GTX 275.

What I'm getting at, and what you don't seem to grasp, is that the performance of the GT430 is going to vary GREATLY, depending upon the workload.
- The architectural improvements don't help with rendering
- They don't help with video encoding
- They help a bit with folding@home (putting a GT 430 well ahead of the GT 240)
- They REALLY help with raytracing.
- They REALLY help with physics calculations.

cannondale06 said:
now use some common sense and realize that a gtx580 is about 20% faster than that gtx480 so that would mean it would need an even faster card for physx. only a fool would claim a gt430 or gt440 would make a good physx card while using a gtx580 as the main card. :rolleyes:
Click to expand...
I'm using plenty of common sense and hard numbers. You're either not reading, or failing to comprehend the implications of the GF100 architecture's internal changes. The GTX 580 would NOT necessarily need a faster card for PhysX. As I've explained to you countless times now, a GT 430 or GT 440 wont pose the horrific bottleneck that you seem to think they will.

All I can tell you is to read more carefully, because it's plain as day that any card based on GT200 (like many of the cards in your table) will suck at PhysX compared to reasonable card based on GF100 (like the GT 430 and GT 440), which is why you need something ridiculous like a GTX 275 if you go with a GT200 based card, but much lower-end GF100 based cards can pull their physics-processing weight just fine.

You seem to think a GTX 275 is a fast PhysX card... ok, but compare it against the GT 440 real quick. The 440 is only short 3 SFUs, and they're clocked 15% faster than the SFUs on the GTX 275, and their performance is further bolstered by other changes to the architecture. End result? The GT 430 should come in pretty close to the GTX 260 / GTX 275 when comparing their Special Function Unit performance.

cannondale06 said:
EDIT: and you still keep going on with the Nvidia to AMD comparisons when it does not work like that for physx. for example if you have a 6870 then going to a 6970 for your main gpu is NOT going to help with physx. if you have a gtx560 for your main gpu and upgrade to a gtx580 then you increase BOTH your graphics AND physx processing power. that is why you need a faster card for physx when you use a higher end Nvidia gpu.
Click to expand...
Yes, I keep using them because said comparisons work just fine... All of my comparisons were using a DEDICATED PhysX card. When using a dedicated card, the rendering card doesn't matter (it just needs to be a fast rendering card so we can spot a bottleneck). You now have multiple people telling you this...

If you're happy with the performance you get when running both Graphics and PhysX on the same card (even though it's slower than using almost any dedicated PhysX card), then I'm not sure why you're even involved in a discussion about dedicated PhysX cards.

And yes, if you WERE running both graphics and PhysX on a single card, and you upgraded that single card, you'd gain more graphics and PhysX performance... but so what? What's your point? All you've managed to do is emulate having a weaker graphics card (because now some of its resources are going to PhysX processing). Give the poor GTX 580 a GT 430 or GT 440 so it can stretch its legs and not be bogged down by PhysX calculations.
 
Last edited:
cannondale06 said:
yes once you reach the maximum level then it does not matter. a gt430 will NOT let you reach the maximum level though. again open your eyes and look at that chart. it takes a gtx275 or better to get the most out of physx while using a gtx480 for graphics. a gt430 is a much slower card than a gtx275 and would perform below the 9800gtx in that game. with a gtx580 for the main card then a gt430 would look even worse as a physx card.
Click to expand...
Please... please just read the information that's in front of you. We can't help you if you continue to willfully ignore the massive difference in architecture.

Rendering performance DOES NOT equal PhysX performance. The GT 200 core is great at rendering but poor at PhysX. It's as simple as that. Even G92 was more efficient at PhysX than GT200.

I've told you why cards that use the GT 200 core (like the GTX 275) are ridiculously weak at PhysX; that generation of GPU is so terribly weak for PhysX that modern low-end cards based on the GF 100 core (like the GT 430) match and exceed comparatively beefy cards that use the older core.

Edit: As for your suggestion that a GTX 460 is necessary for PhysX on faster cards...here's further clarification as to why that is a totally ridiculous amount of overkill (maybe this will put things into perspective for you): A GTX 460 has 42 SFUs on-board. That's equivalent to using three GTX 285s in tri-SLI as a PhysX card. That's OBVIOUSLY more than even a GTX 580 needs as a dedicated card.
 
Last edited:
would it be wise to get a dedicated PhysX card that had clocks about the same or faster than the primary card? I have a 6850/GTS 450 combo and that isn't working that great and i think the clocks on the GTS 450 aren't fast enough.
 
I ran my 560Ti with a 260GTX for Physix in the Batman AC benchmark. Running (dedicated) Physix on the 560TI ALWAYS gave higher results than running (dedicated) Physix on the 260GTX.

Pulled the 260GTX today, it was useless and wastes power. I need to sell it. Physix needs to be open source. Nvidia is holding back graphics development, they spent all that money for it and hold it up. So dumb.
 
I ran my 560Ti with a 260GTX for Physix in the Batman AC benchmark. Running (dedicated) Physix on the 560TI ALWAYS gave higher results than running (dedicated) Physix on the 260GTX.
Click to expand...
Really? Care to show some bench screen shots?
I have contemplated buying a midrange last gen nvidia card for Physx but that makes me not want to.
 
Actually I am surprised people are still talking about physx, a quick check and newegg doesn't carry them any more, amazon has some but @ $30. Talk about a dead product. Nvidia bought them and that was it.
 
jiminator said:
Actually I am surprised people are still talking about physx, a quick check and newegg doesn't carry them any more, amazon has some but @ $30. Talk about a dead product. Nvidia bought them and that was it.
Click to expand...
what? did you even read anything here? we are talking about graphics cards to use for dedicated physx not the old physx cards.
 
How would an XFX 8800 512MB Alpha Dog Edition do with PhysX? I don't remember if it's a GT or GX or GTX or GTS, but it has a G92 core and is clocked at 640MHz if that helps.
 
jiminator said:
Ouchy!

LOL, still obsolete. :)
Click to expand...

Still obsolete? More like someone can't read.

FFS dude, if you can't wrap it around your head that a card dedicated to physics processing increases the main GPU's ability to do EVERYTHING ELSE better, then you're about as smart as a doorknob. I don't even like nVidia but am considering going to a multi-GPU setup for this purpose.
 
magoo said:
Yes, my popcorn needed more butter though.:D

I actually learned some stuff, which may come in handy for Borderlands 2.
Click to expand...

Seriously? That's EXACTLY why I'm here as well! :p

I have an old build with just an XFX Radeon HD 6850 Black Edition and I came here after reading this thread (http://hardforum.com/showthread.php?t=1645682)

I think my 6850 might be dying, I saw artifacting in Borderlands whenever I enabled Bloom, but I might want to test it again. Long story short, I was ordering some parts for my new build when I saw the Borderlands 2 Physx demo video and ended up here.

Thanks to all the information from Unknown-One and N1GHTRA1N, I'm going to try an XFX Double D 78XX and pairing it with a GT 430 for PhysX! :D
 
{NG}Fidel said:
Really? Care to show some bench screen shots?
I have contemplated buying a midrange last gen nvidia card for Physx but that makes me not want to.
Click to expand...

After benching a 660 ti alone and then paired with a GT 430 I came out losing 3-4 fps having the 430 as a dedicated Physx card. It would be fine for a hybrid system in most cases besides maybe Batman AC for which I would want to get a cheap card with a lot of Cuda cores to run PhysX high without bottle-necking the main GPU too much. I would think something like the GT 640 would be able to handle anything out now and into the foreseeable future with ease.
 
convexion said:
After benching a 660 ti alone and then paired with a GT 430 I came out losing 3-4 fps having the 430 as a dedicated Physx card. It would be fine for a hybrid system in most cases besides maybe Batman AC for which I would want to get a cheap card with a lot of Cuda cores to run PhysX high without bottle-necking the main GPU too much. I would think something like the GT 640 would be able to handle anything out now and into the foreseeable future with ease.
Click to expand...
yep just like I said it would earlier.
 
You must log in or register to reply here.
Back
Top