Value In SLI? GTX 260 Core 216 Vs. GTX 280

[neiroatopelcc]

[citation][nom]cmmarco[/nom]One thing I don't see mentioned with regards to 32bit or 64bit is the affect the amount of video memory would have on usable memory. I would think SLI on 32bit (i.e. 3 x 512mb cards) would cut into the amount of available memory (due to memory addressing limitations) and thus have an overall reduction in the performance of OS. I have not SLI'd so maybe this is not an issue, but I would not think this limitation has been overcome has it?[/citation]
I've never heard of the addressing limit applying to discreet hardware. My understanding is, that the limitation is in the os kernel. Firstly the actual addressing limit is 4gb - effectively lower and variable depending on bios settings. Secondly I'm understanding that sli management is handled by the graphics drivers, not the hardware. So the addressing limit would at most happen when video memory on each individual card exceeds 4gb - and even so, I still believe the addressing issue is limited to the main ram and caused by how the windows memory manager addresses stuff. Thirdly if it indeed is a problem, it likely can be circumvented by using an equivivalent of pae.
In short - don't think it's an issue at all.

 

[Amphibiousguru]

[citation][nom]neiroatopelcc[/nom]I've never heard of the addressing limit applying to discreet hardware. My understanding is, that the limitation is in the os kernel. Firstly the actual addressing limit is 4gb - effectively lower and variable depending on bios settings. Secondly I'm understanding that sli management is handled by the graphics drivers, not the hardware. So the addressing limit would at most happen when video memory on each individual card exceeds 4gb - and even so, I still believe the addressing issue is limited to the main ram and caused by how the windows memory manager addresses stuff. Thirdly if it indeed is a problem, it likely can be circumvented by using an equivivalent of pae. In short - don't think it's an issue at all.[/citation]

I smell a potential article on this one.

Curiously, I know that using 4GB DDR2, a single GTX 280, and Windows Vista Ultimate 32-bit on MY computer causes issues. It shows up as having just over 3GB system RAM. IDK maybe I messed something up. Using 64-bit OS fixed the problem though.

 

[neiroatopelcc]

[citation][nom]Amphibiousguru[/nom]I smell a potential article on this one.Curiously, I know that using 4GB DDR2, a single GTX 280, and Windows Vista Ultimate 32-bit on MY computer causes issues. It shows up as having just over 3GB system RAM. IDK maybe I messed something up. Using 64-bit OS fixed the problem though.[/citation]
Which graphics adapter you have is almost irrelevant to your issue.
Your problem exclusively is in how windows memory management handles the available memory. You can google PAE extention for more info on this matter. But to be brief - if you'd replace your 280 card with say a mystique 220 or an s3 trio or a race 2c you'll still be missing memory in windows.

On a side note, it's not much more efficient to run 64bit with 4gb than it is to run xp with 3,5gb - cause addresses for variables in software need to be twice as long, making it likely to take up so much of the extra memory that you can't feel a difference. ie. all programs just use a bit more memory without any performance benefit.

 

[ozyman]

One thing that was missed is the CUDA value-add in this build. The 2xGTX280 vs 3xGTX260 systems are almost perfectly matched for gaming performance, but you should get a lot more CUDA compute horsepower out of the 3xGTX260 - that means it'll play PhysX games and do video recoding substantially faster. Given everything else is more or less even, that should tip the balance.

 

[seboj]

[citation][nom]neiroatopelcc[/nom]I've never heard of the addressing limit applying to discreet hardware. My understanding is, that the limitation is in the os kernel. Firstly the actual addressing limit is 4gb - effectively lower and variable depending on bios settings. Secondly I'm understanding that sli management is handled by the graphics drivers, not the hardware. So the addressing limit would at most happen when video memory on each individual card exceeds 4gb - and even so, I still believe the addressing issue is limited to the main ram and caused by how the windows memory manager addresses stuff. Thirdly if it indeed is a problem, it likely can be circumvented by using an equivivalent of pae. In short - don't think it's an issue at all.[/citation]

Hm. It was my understanding that the system would reserve address space allocated for BIOS, motherboard resources, memory mapped I/O, and other PCI / AGP configurations first, hence causing the 32-bit OS to recognize < 4gb.

It's also my understanding that PAE is not supported by desktop versions of Windows, due to driver compatibility issues.

 

[deanjo]

[citation][nom]Shadow703793[/nom]Again, can some one please run BadaBOOM on a tri-SLI set up?[/citation]


Badaboom cannot yet take advantage of more then one card per encode. Spanning one encoding process the load over multiple cards is not yet supported.

 

[coopchennick]

[citation][nom]neiroatopelcc[/nom]I've never heard of the addressing limit applying to discreet hardware. My understanding is, that the limitation is in the os kernel. Firstly the actual addressing limit is 4gb - effectively lower and variable depending on bios settings. Secondly I'm understanding that sli management is handled by the graphics drivers, not the hardware. So the addressing limit would at most happen when video memory on each individual card exceeds 4gb - and even so, I still believe the addressing issue is limited to the main ram and caused by how the windows memory manager addresses stuff. Thirdly if it indeed is a problem, it likely can be circumvented by using an equivivalent of pae. In short - don't think it's an issue at all.[/citation]

Just to clear things up a bit, I am pretty sure that it works like this...
A 32 bit OS can only address 4GB of memory (2^32). Theoretically, if you had no components that had ram on them (which would need to be addressed by the OS), then you would see all of your 4GB of system ram. But a video card (lets say a 280 with 1GB) has such memory that must be addressed and other things like a sound card, etc also use up space from that 4GB limit. The OS sees these system components' memory more necessary than system ram, it doesn't address all of the system ram (rather than those components' ram).

So if I had a 32 bis OS on a system with 4GB ram installed, and a video card with 2GB, Windows sees that card's ram more important and cuts into the system ram so the video card can use all of its ram.

Also, in SLI setups, all of the information on the video cards' ram is redundant so it only uses the addressing for the size of 1 of those card's ram.

Hope this helps

 

[coopchennick]

Oh, also the PAE thing - You're right, the OS uses a few extra bits so it can map more memory but I somewhat read up on it and am a little weary of the whole "unnoficiality" and lacking support for it.

In the meantime, 3.3 GB is working fine for me.

 

[xianghuanzhong]

[citation][nom]Pei-chen[/nom]A 4870 X2 costs around $500. A GTX 280 $330. $500 x 2 = $1000 > $330 x 2 = $660 so what extra funds are you talking about?[/citation]
agree

 

[neiroatopelcc]

[citation][nom]seboj[/nom]Hm. It was my understanding that the system would reserve address space allocated for BIOS, motherboard resources, memory mapped I/O, and other PCI / AGP configurations first, hence causing the 32-bit OS to recognize < 4gb.It's also my understanding that PAE is not supported by desktop versions of Windows, due to driver compatibility issues.[/citation]
Dunno about the pae limitation, as I understood it wouldn't make much of a difference having xp 32 with 3,5gb or xp/vista 64 with 4gb - and I only have 4gb in my home pc. At work I've got 6gb memory though, and am using vista 64. Anyhow, the other part I'm quite sure you're right about.

[citation][nom]coopchennick[/nom]Just to clear things up a bit, I am pretty sure that it works like this...A 32 bit OS can only address 4GB of memory (2^32). Theoretically, if you had no components that had ram on them (which would need to be addressed by the OS), then you would see all of your 4GB of system ram. But a video card (lets say a 280 with 1GB) has such memory that must be addressed and other things like a sound card, etc also use up space from that 4GB limit. The OS sees these system components' memory more necessary than system ram, it doesn't address all of the system ram (rather than those components' ram).So if I had a 32 bis OS on a system with 4GB ram installed, and a video card with 2GB, Windows sees that card's ram more important and cuts into the system ram so the video card can use all of its ram.Also, in SLI setups, all of the information on the video cards' ram is redundant so it only uses the addressing for the size of 1 of those card's ram.Hope this helps[/citation]
You may not know so, but the video framebuffer is a very small buffer. It consists of 128kb memory in the 'upper memory region' (b0000-bffff), so the amount of video memory is completely irrelevant in this scenario. However it isn't irrelevant when talking direct3d applications, as these keep a copy of video memory in local memory as far as I've understood - or at least a portion of it.

How video memory is used in sli is beyond me, but I'm sure if it's sli'd using tiling, it'd have to have the same content on all cards. I'm not so sure if it was alternate frames or the like though.

 

[Thor]

It's pity you don't use the excellent Enemy Territory:Quake Wars in your benchmark...


Hope next you will do that!

 

[jaragon13]

How about GTX 260 in SLI versus GTX 260 core 216 in SLI,versus a single GTX 280? lol,that'd be single sided. funny how people are so bitchy at this comment...yet for no good damned reason.Really? Grow up.

 

[jaragon13]

[citation][nom]cangelini[/nom]Just an FYI, all testing in the Tom's Hardware US lab is being transitioned to 64-bit Vista. We hear your feedback and are responding accordingly.Just an interesting aside, though--Windows XP x32 remains the favorite for overclocking. We even received the heads-up from AMD not to expect Phenom II to overclock as aggressively under a 64-bit OS. Remember that when our Phenom II vs. i7 overclocking comparison launches. Just something to keep in mind for the enthusiasts clamoring for a native 64-bit environment with 32-bit tests *duck*[/citation]
Does it really matter? Sounds like something AMD shouldn't do.(or they are messing around the experts reviewers with their 64 bit systems)
32 bit is becoming more useless day by day,my dual boot XP/Vista X64 works wonders.

 

[GizmoPunk]

If you're going to benchmark games, benchmark games people are currently buying and playing. The games benchmarked here are definitely not what most gamers are spending their time on, and are probably not representative of latest rendering engines used in the games that are popular.

 

[seboj]

I went 64-bit a while ago and never looked back. Currently running WinServ '08 on an overclocked 4800+ and 6gb RAM. Makes me smile every day.

I just don't see the point in 32-bit anymore, excluding benchers.

That part about AMD is interesting from a PR standpoint. AMD tells review sites not to expect much OC, so then the sites don't hype the product. Launch day comes, the products reviewed, and no one's disappointed, because no one was expecting much. I just hope they're not trying to cover their ass for releasing a sub-par product.

 

[coopchennick]

[citation][nom]neiroatopelcc[/nom]You may not know so, but the video framebuffer is a very small buffer. It consists of 128kb memory in the 'upper memory region' (b0000-bffff), so the amount of video memory is completely irrelevant in this scenario. However it isn't irrelevant when talking direct3d applications, as these keep a copy of video memory in local memory as far as I've understood - or at least a portion of it.[/citation]

That's interesting, thanks. I was under the impression that the exact size of the video memory is the amount of memory the OS had to handle/address. I guess I have some reading to do...

 

[Guest]

That framebuffer you mentioned is the VGA framebuffer. It's used exclusively when the graphics card is operating in VGA compatibility mode - that is, it has a highest resolution of 320x200 8-bit or 640x480 4-bit. 320x200 8-bit is what Windows XP uses to draw it's loading bar.

When the OS loads the actual graphics drivers, it assigns a region of the address space to the graphics card. If there is enough unused address space, it gives it that, else it will overlap some of it with system RAM. This RAM is then mapped into an application's address space when taking advantage of hardware acceleration - so it can copy textures and write graphics commands - and hidden behind OpenGL and Direct3D.

 

[Movian]

Can i just mention that people are focusing too much on The Crysis Benchmark. To me from reading the review the choice of 3 260 cards over the 2 280 cards was largely factored by the current pricing in the market. As they state at the time of purchase it saved $100 which ban buy allot of beer to drink while playing Crysis
However at the time of this new article the price changes reduced that margin.

Making it almost a tie in comparison... However i would argue in favor of the 2 cards as at a later time you can add in a third 280. Enabling a future performance boost when the cards get yet again reduced in price

Great Review