Why does the GTX 295 fail in vRAM?!

Jacez44

Limp Gawd
Joined
May 9, 2007
Messages
417
It has 1792MB of vRAM but it somehow fails when vRAM is needed.
Such as in Crysis @ 2560x1600:
GTX295-44.JPG

Even if they somehow made the terrible mistake of splitting the vRAM into two separate banks for each GT200b core, that would still not explain how the GTX 260 does just fine while the GTX 295 does not. And heck, 896MB of RAM is so close to 1GB, that I don't understand how the 4870X2 does any better.

I'm starting to wonder if I should just get GTX 280 Triple SLi for about 600$, instead.
 
Even if they somehow made the terrible mistake of splitting the vRAM into two separate banks for each GT200b core, that would still not explain how the GTX 260 does just fine while the GTX 295 does not. And heck, 896MB of RAM is so close to 1GB, that I don't understand how the 4870X2 does any better.

I'm starting to wonder if I should just get GTX 280 Triple SLi for about 600$, instead.

(emphasis mine)

The moment you made this statement, you opinions and theories lost any validity they might have had. You have no idea how SLI works. This is SLI on a card. There's no other way to do it -- if they did it another way, it wouldn't be SLI. This is how it's been on every dual-GPU card from Nvidia /and/ ATI since the beginning. Get with it.

Also, GTX 280 Tri SLI would be more like $800.
 
It has 1792MB of vRAM but it somehow fails when vRAM is needed.
Such as in Crysis @ 2560x1600:

Even if they somehow made the terrible mistake of splitting the vRAM into two separate banks for each GT200b core, that would still not explain how the GTX 260 does just fine while the GTX 295 does not. And heck, 896MB of RAM is so close to 1GB, that I don't understand how the 4870X2 does any better.

I'm starting to wonder if I should just get GTX 280 Triple SLi for about 600$, instead.

Erm the ram *is* split on all NVIDIA and ATI dual gpu boards. If you get a board which has "1GB of memory" it has 512MB dedicated to each GPU, and the memory contents are identical between the two 512MB memory sets.
 
Even if they somehow made the terrible mistake of splitting the vRAM into two separate banks for each GT200b core

Terrible mistake? Do you even understand how SLI works?

It doesn't have 1792MB any more than a pair of 280's has 2GB. It just doesn't.
 
Yes, thanks for bashing my lack of knowledge. That was fun.

My point is that if they centered the 2GB of ram of say the 4870X2 in a single bank where both GPUs could load and unload the textures to and from, that would increase performance "terribly", considering that both vRAM banks have identical contents. It's just a very smart idea that I assumed would be implemented on the GTX 295, let alone on the 4870X2, which can do it more easily since it has only a single PCB.

And as for the GTX 280 Triple SLi for 600$ - Try Buy.com. They sell each for 200$ (190$ AR).
What reason do I have to get GTX 295 Quad-SLi for 1000$ when the 6 month-old 280-Tri SLi gets better performance @ 2560x1600, which is the resolution I will use (if not higher - with triple monitors). I mean.. why does it simply die at that resolution when all other cards, aside from the ones with 512MB of vRAM, work properly. The only reason I can think up to excuse this is crappy drivers, which makes my decision harder..
 
And as for the GTX 280 Triple SLi for 600$ - Try Buy.com. They sell each for 200$ (190$ AR).
Why do people keep referencing this?
Buy.com has yet to ship a single one of those GTX 280's and it's more than likely a pricing error.
 
My point is that if they centered the 2GB of ram of say the 4870X2 in a single bank where both GPUs could load and unload the textures to and from, that would increase performance "terribly", considering that both vRAM banks have identical contents. It's just a very smart idea that I assumed would be implemented on the GTX 295, let alone on the 4870X2, which can do it more easily since it has only a single pcb

Every smart idea we think we have has been thought about by these guys 10 years ago. If this were easy, cheap or even feasible to do it wouldve been done already.
 
the point is you don't understand how it works so you don't have a point. Also something is clearly wrong with the drivers they used or they just faked the benchmarks. How does a single 192 260gtx have a higher minimum frame rate?
 
The content of the two memory banks is *not* the same, that's why they can't combine them. Each GPU needs its own 'workspace' so to speak.
 
Yes, thanks for bashing my lack of knowledge. That was fun.

My point is that if they centered the 2GB of ram of say the 4870X2 in a single bank where both GPUs could load and unload the textures to and from, that would increase performance "terribly", considering that both vRAM banks have identical contents. It's just a very smart idea that I assumed would be implemented on the GTX 295, let alone on the 4870X2, which can do it more easily since it has only a single PCB.

And as for the GTX 280 Triple SLi for 600$ - Try Buy.com. They sell each for 200$ (190$ AR).
What reason do I have to get GTX 295 Quad-SLi for 1000$ when the 6 month-old 280-Tri SLi gets better performance @ 2560x1600, which is the resolution I will use (if not higher - with triple monitors). I mean.. why does it simply die at that resolution when all other cards, aside from the ones with 512MB of vRAM, work properly. The only reason I can think up to excuse this is crappy drivers, which makes my decision harder..

Sli does not support multiple monitors anyways. So while you will be able to use 3+ monitors it will not be in Sli. And since you are talking about 3 cards you wont be running XP for spanning.

So the highest single display res is 2560 currently.
 
Sli does not support multiple monitors anyways. So while you will be able to use 3+ monitors it will not be in Sli. And since you are talking about 3 cards you wont be running XP for spanning.

So the highest single display res is 2560 currently.

SLI supports multi-monitor.
 
Since when (havent been in the nvidia loop)? Regardless, he wants more than two GPU's he will be in vista which does not support spanning.
 
Since when (havent been in the nvidia loop)? Regardless, he wants more than two GPU's he will be in vista which does not support spanning.

~October of last year? Something like that. The 180 series added multi-monitor support. The trick is, it just assigns a GPU to each screen... so it's really disabling SLI without having to restart. All fine and good, but not what people were asking for at all.
 
this always did annoy me. I understand that yes these cards do technically have 2x memory and 2x width but in reality they have to load the same thing twice so they really don't. ALMOST lying but then it is.
 
The Crysis engine seems to have issues with the GTX 295 in general. I've tested it a lot in the last day and sometimes it works great and sometimes it doesn't at all. Hopefully a driver will come out soon to patch it up.
 
(emphasis mine)

The moment you made this statement, you opinions and theories lost any validity they might have had. You have no idea how SLI works. This is SLI on a card. There's no other way to do it -- if they did it another way, it wouldn't be SLI. This is how it's been on every dual-GPU card from Nvidia /and/ ATI since the beginning. Get with it.

Also, GTX 280 Tri SLI would be more like $800.

I agree with everything said here.

Unless the OP can link to $200 GTX 280s.

edit:
And as for the GTX 280 Triple SLi for 600$ - Try Buy.com. They sell each for 200$ (190$ AR).

Now your just making stuff up.

Understand, theres a huge emphasis from the public on how much RAM a video card has, and thats completly skewed its importance.

Furthermore a video card drivers are completly taken for granted, cuz they're always there and they almost always work.
 
My point is that if they centered the 2GB of ram of say the 4870X2 in a single bank where both GPUs could load and unload the textures to and from, that would increase performance "terribly", considering that both vRAM banks have identical contents.

That idea has been brought up over the last few years or so throughout the forums. In fact I think that idea has been mentioned at least once in a HardOCP GPU review. Can't remember which one though.

Anyway, while yes it is a good idea, I'm fairly sure that at least one engineer, either at Nvidia or ATI, have thought up of the idea, actually gone through the trouble of working it out, and have their own conclusions why it can or cannot work.

I mean.. why does it simply die at that resolution when all other cards, aside from the ones with 512MB of vRAM, work properly. The only reason I can think up to excuse this is crappy drivers, which makes my decision harder..

A few reasons:
- Drivers
- Whether or not the game itself (dependent on the game engine) scales well with SLI.

The GTX295 was released about a week ago and the drivers are a bit immature. Every single card in that HardwareCanucks has been on the market for at least 4 months and have relatively mature drivers.

Oh and no hotlinking images from other sites that you do not own or have an image account with. Edit your first post to make that hotlink into a regular link.
 
Hopefully future drivers will fix that issue, and yes, I suspect it ran out of RAM, yet again NVIDIA screwed up with this bit, they should have made it a full 2GB card like the 4870x2.
 
Every smart idea we think we have has been thought about by these guys 10 years ago. If this were easy, cheap or even feasible to do it wouldve been done already.

Exactly right. This guy is such a genius.. he thought of the only thing ATI and NVIDIA never did! lmao. Oversimplification leads to "everyones a genius" syndrome.

The content of the two memory banks is *not* the same, that's why they can't combine them. Each GPU needs its own 'workspace' so to speak.

Right, but the point is resources need to be transferred between the GPUs (duplicated) at times when the rendering doesn't give hints that they don't need it anymore. So some of the resources are exactly the same, some of them are not.

Sli does not support multiple monitors anyways. So while you will be able to use 3+ monitors it will not be in Sli. And since you are talking about 3 cards you wont be running XP for spanning.

So the highest single display res is 2560 currently.

SLI supports up to 2 monitors with 180.xx drivers. Keep up.
 
Hopefully future drivers will fix that issue, and yes, I suspect it ran out of RAM, yet again NVIDIA screwed up with this bit, they should have made it a full 2GB card like the 4870x2.

Finally, an actual answer to my question. I don't know why everyone is taking this issue so seriously, but I thank you all.

Oh yes, and to the douche who said I was making stuff up - the Buy.com-GTX 280-200$-giveaway has ended. Maybe if you're very special you can find a similar one on Slickdeals.com. Good luck.
 
SLI supports up to 2 monitors with 180.xx drivers. Keep up.

I hope you're done trying to make yourself feel better by insulting others, because I have a question.

If I buy two GTX 295's in Quad-SLi and leave my 8800 GTS 512 to do physics, can I run 2 monitors off the 295 SLi and another 2 on the G92, or by inputing "SLi" my system forgets any number above 2 in monitor quantity?
 
Furthermore a video card drivers are completly taken for granted, cuz they're always there and they almost always work.

That's BS... I had issues with drivers for my Xfired 4850's. They aren't always there (BSOD during install) and they sure as hell don't always work if I don't get the most perfect set that plays nice...
 
If I buy two GTX 295's in Quad-SLi and leave my 8800 GTS 512 to do physics, can I run 2 monitors off the 295 SLi and another 2 on the G92, or by inputing "SLi" my system forgets any number above 2 in monitor quantity?

If you buy a pair of GTX 295s and keep the 8800 GTS 512 for PhysX you can run 2 monitors, total, off one of the 295s.
 
First of all, everyone seems to have forgotten the rumors of shared memory floating around before the 4870x2 launch. It's entirely possible with GDDR5, but like sideport wasn't implemented because it just wasn't needed enough to delay the launch and it would have increased price, which was ATI's way of strongarming their way back into the market last year.



If I buy two GTX 295's in Quad-SLi and leave my 8800 GTS 512 to do physics, can I run 2 monitors off the 295 SLi and another 2 on the G92

Yes, cards outside of the SLI set can still drive monitors.


Not true, the 4870x2 only has 1GB of truly usable vram, the 1GB on each card are mirrors of each other.

Read the whole damned sentence to avoid looking stupid.


Try checking the HotDeals section of the forum once in a while. Granted it was probably a mistake and should have been $100 higher, but the cards were still being sold at that price this morning. Buy probably learned from Dell and sold a greatly under-priced item to get a few hundred thousand dollars for a few days.
 
Hopefully future drivers will fix that issue, and yes, I suspect it ran out of RAM, yet again NVIDIA screwed up with this bit, they should have made it a full 2GB card like the 4870x2.

It didn't run out of VRAM. It is a known bug that the GTX 295 just craps out at 2560x1600. Resolution has a VERY VERY SMALL impact on VRAM usage. *EXTREMELY* small impact. 2560x1600 only uses 2.6% of the available VRAM (double buffered no AA). Everything else that is stored in VRAM is independent of the resolution.

This notion that higher resolutions == higher VRAM usage (while technically correct) is extremely misleading. The fact is that the framebuffer (which is the only thing that uses up more VRAM at higher resolutions) uses a tiny, tiny amount of VRAM.

AA will increase the amount of VRAM usage used, but not by a significant amount (and I don't feel like calculating it) afaik (at least 2x AA won't, which is what this was about).
 
It didn't run out of VRAM. It is a known bug that the GTX 295 just craps out at 2560x1600. Resolution has a VERY VERY SMALL impact on VRAM usage. *EXTREMELY* small impact. 2560x1600 only uses 2.6% of the available VRAM (double buffered no AA). Everything else that is stored in VRAM is independent of the resolution.

This notion that higher resolutions == higher VRAM usage (while technically correct) is extremely misleading. The fact is that the framebuffer (which is the only thing that uses up more VRAM at higher resolutions) uses a tiny, tiny amount of VRAM.

AA will increase the amount of VRAM usage used, but not by a significant amount (and I don't feel like calculating it) afaik (at least 2x AA won't, which is what this was about).

You sound so sure.. I wish I could believe you as much as I want to.
 
Do you think this can be resolved with drivers?

His statement has no backing. It makes no sense for the GTX 295 to simply "be bad at 2560x1600". But in theory, yes, since it's not a hardware problem, it can and must be solved by drivers.
 
It didn't run out of VRAM. It is a known bug that the GTX 295 just craps out at 2560x1600. Resolution has a VERY VERY SMALL impact on VRAM usage. *EXTREMELY* small impact. 2560x1600 only uses 2.6% of the available VRAM (double buffered no AA). Everything else that is stored in VRAM is independent of the resolution.

This notion that higher resolutions == higher VRAM usage (while technically correct) is extremely misleading. The fact is that the framebuffer (which is the only thing that uses up more VRAM at higher resolutions) uses a tiny, tiny amount of VRAM.

AA will increase the amount of VRAM usage used, but not by a significant amount (and I don't feel like calculating it) afaik (at least 2x AA won't, which is what this was about).
I'll attest that the drivers do need some tweaks. There is a major problem with the Crysis engine I've found, and it doesn't matter what resolution or quality settings you choose. The card will just rapidly stutter for periods at a time for no reason. It's slower than microstutter but faster than paging/memory stutter. Anyway, I'll be curious to see what NVIDIA does to fix this. Overall though, the card runs every other game extremely well at 2560x1600.
 
The fact is that the framebuffer (which is the only thing that uses up more VRAM at higher resolutions) uses a tiny, tiny amount of VRAM.

Don't intermediate render targets scale with resolution? They can easily account for more than the framebuffer itself.

But ignoring those, a single 4MP FP16 framebuffer by itself is not a trivial amount of data. Add 4xAA plus the Z-buffer and you easily get up to 180MB+ for the framebuffer alone. That is closer to 20% of a 1GB pool, not 2%.
 
It is stated that the initial driver release with the gtx 295 are not optimal of coure, and nvidia states they'll be releasing a driver update that will fix the frame drop errors, in some benchmarks i've seen the gtx 295 fails completely and crashed out cause the driver they used were crap. Knowing Nvidia though they'll fix this issue with new drivers.
 
It is stated that the initial driver release with the gtx 295 are not optimal of coure, and nvidia states they'll be releasing a driver update that will fix these errors, in some benchmarks i've seen the gtx 295 fails completely and crashed out cause the driver they used were crap. Knowing Nvidia though they'll fix this issue with new drivers (unlike ATI who nearly always has something wrong with driver release but thats beside the point).

While true, on both counts, newer/more intense games will stress the 295's memory levels and the card will always be handicapped in those games at very high resolutions.
 
Oh yes, and to the douche who said I was making stuff up - the Buy.com-GTX 280-200$-giveaway has ended. Maybe if you're very special you can find a similar one on Slickdeals.com. Good luck.

Until someone actually gets his or her card I wouldn't say that Buy.com actually sold those cards at that price, since all it takes is a single cancellation email. This is coming from someone who jumped on the deal. :p
 
Believe the one that's right -> me :D


I also have supporting evidence ;) See slide in first post

Holy shit. So, you're basically telling me that if I go with Triple SLi GTX 280 I will only be able to have 2 monitors! Confirm this at once, so I can order my two GTX 295's. :D
 
While true, on both counts, newer/more intense games will stress the 295's memory levels and the card will always be handicapped in those games at very high resolutions.

You can't seriously say that 100MB is going to be the defining factor between playable and unplayable.
 
Back
Top