View Full Version : R700 Launches on 14th July
CaptNKILL
07-10-08, 08:04 PM
It doesn't add up.
If it took extra 1.5GB just to get the other GPU on the same board working, than the real crossfire with 2x 512 Mb single cards would be in big trouble.
Hm?
No, I was talking about quad SLI. 4x512Mb still gives you the same amount of usable memory as a single 512Mb card since each card stores the same exact information.
That's how SLI and crossfire have always been.
I'm not sure where the argument is here...
A 2Gb X2 card has a lot of memory on it but it still has a 1Gb frame buffer. A 1Gb X2 card has a 512Mb frame buffer.
Praising 2Gb on an X2 or GX2 card as some sort of achievement doesn't really make sense. The cards may be fast but its actually a terribly inefficient use of memory and a single GPU card with an actual 1Gb of memory will have the same size framebuffer without all the extra heat and cost generated by having another gigabyte of memory stuck to one of the boards.
Heinz68
07-10-08, 09:09 PM
Hm?
No, I was talking about quad SLI. 4x512Mb still gives you the same amount of usable memory as a single 512Mb card since each card stores the same exact information.
That's how SLI and crossfire have always been.
I'm not sure where the argument is here...
A 2Gb X2 card has a lot of memory on it but it still has a 1Gb frame buffer. A 1Gb X2 card has a 512Mb frame buffer.
Praising 2Gb on an X2 or GX2 card as some sort of achievement doesn't really make sense. The cards may be fast but its actually a terribly inefficient use of memory and a single GPU card with an actual 1Gb of memory will have the same size framebuffer without all the extra heat and cost generated by having another gigabyte of memory stuck to one of the boards.
Sorry I missed the 4x 512MB part. But I clearly stated it would make much more difference if it was shared memory.
I was more concentrating on the part "it just takes an extra 1.5Gb to get the other GPUs working."
Now the other way to look at it. When it's not shared than each GPU needs memory so if the HD 4870x2 was with 1GB mem total (2 x 512mB) and the other with 2GB (2 x 1GB), there sure should be a difference.
Question is couldn't they accomplish EXACTLY same with shared memory, I don't know, but maybe the technology was not there yet.
Or does anybody knows 100% the memory is not shared at all.
One thing I believe they're not so stupid to use the 2GB if it was a waste and the only benefit was, it looks good on the BOX.
Anyway all this is just speculation on my part but i believe we will see
both 1GB and 2GB cards variation sooner or later and see for real what the benefits are.
CaptNKILL
07-10-08, 09:20 PM
Sorry I missed the 4x 512MB part. But I clearly stated it would make much more difference if it was shared memory.
I was more concentrating on the part "it just takes an extra 1.5Gb to get the other GPUs working."
Now the other way to look at it. When it's not shared than each GPU needs memory so if the HD 4870 was with 1GB mem total (2 x 512mB) and the other with 2GB (2 x 1GB), there sure should be a difference.
Question is couldn't they accomplish EXACTLY same with shared memory, I don't know, but maybe the technology was not there yet.
Or does anybody knows 100% the memory is not shared at all.
One thing I believe they're not so stupid to use the 2GB if it was a waste and the only benefit was, it looks good on the BOX.
Anyway all this is just speculation on my part but i believe we will see
both 1GB and 2GB cards variation sooner or later and see for real what the benefits are.
Sorry, the 1.5Gb thing was bad wording on my part. I just meant that the extra memory was needed because each GPU needed its own frame buffer. :)
I agree that shared memory would be a huge improvement. But so far the rumors have told us that the X2 will not have shared memory. The fact that they are releasing a 2Gb model makes me think they still need 1Gb per GPU to have a 1Gb frame buffer. I can't imagine any other reason to put that much memory on the card.
The 1Gb X2 model will probably have memory problems before too long, just like any other 512Mb setup that is available now. The 2Gb model will probably stay consistent for much longer. This is the main reason I've stuck to a 768Mb card for almost 2 years when there are faster 512Mb cards available.
Heinz68
07-10-08, 09:51 PM
Sorry, the 1.5Gb thing was bad wording on my part. I just meant that the extra memory was needed because each GPU needed its own frame buffer. :)
I agree that shared memory would be a huge improvement. But so far the rumors have told us that the X2 will not have shared memory. The fact that they are releasing a 2Gb model makes me think they still need 1Gb per GPU to have a 1Gb frame buffer. I can't imagine any other reason to put that much memory on the card.
The 1Gb X2 model will probably have memory problems before too long, just like any other 512Mb setup that is available now. The 2Gb model will probably stay consistent for much longer. This is the main reason I've stuck to a 768Mb card for almost 2 years when there are faster 512Mb cards available.
All clear to me now.
This is great news. Sold the 260 on ebay and will get one of these when it surfaces.
abtomat74
07-11-08, 12:04 AM
Wow, the first 2GB card!
Well, 2x1GB...which leads me to my question:
Will the memory limitations of a 32bit O/S count this as 2GB or just 1GB? If the former, then that card would basically "require" a 64bit O/S as you would have around 1.7-1.8GB leftover.
If the performance is that good and the driver support is there I'll be putting the 260 in the spare rig and buy one FO SHO!
Well, 2x1GB...which leads me to my question:
Will the memory limitations of a 32bit O/S count this as 2GB or just 1GB? If the former, then that card would basically "require" a 64bit O/S as you would have around 1.7-1.8GB leftover.
I do not think that there would be any limitation from the OS, as the OS will see the video card as a device. There will be no "OS" requirement.....
CaptNKILL
07-11-08, 02:53 AM
I do not think that there would be any limitation from the OS, as the OS will see the video card as a device. There will be no "OS" requirement.....
I'm not sure about that.
I don't really know much about it myself but others have told me that all of the memory used in the system, including video memory, counts toward the total address space supported by the OS.
It seems like people on 32bit operating systems would be running into problems pretty often if that were true though given the amount of video memory on modern graphics cards.
I'm not sure about that.
I don't really know much about it myself but others have told me that all of the memory used in the system, including video memory, counts toward the total address space supported by the OS.
It seems like people on 32bit operating systems would be running into problems pretty often if that were true though given the amount of video memory on modern graphics cards.
Contrary to what I have read, addressable memory for the system is limited to I dunno whatever it is 3.5 GB or so thing for the a 32 bit OS. The memory on the card is independent from the system memory the OS addresses/tracks/uses etc from what I have read. Maybe I am wrong but,...
Just FYI. ATI already sells 2GB fireGL card which can work on Windows XP 32 bit and Linux 32 bit and Vista 32 bit
http://ati.amd.com/products/workstation/techspecs3.html
ChrisRay
07-11-08, 03:39 AM
The addressable memory issue is only an issue with 32 bit OS. If you have 3 8800GTX cards in a system with 4 gigs of memory. You only get like 1.5 gigs of memory ;p
Haha, so with one of these cards you're limited to less than 2GB of system memory in a 32-bit OS. I guess 64-bit just became more necessary.
To me the fact they're going for 2GB proves there won't be memory sharing. Oh well, should be a great card.
The addressable memory issue is only an issue with 32 bit OS. If you have 3 8800GTX cards in a system with 4 gigs of memory. You only get like 1.5 gigs of memory ;p
Ah ok, so I am wrong in my inference. OS will affect vid memory availability?
CaptNKILL
07-11-08, 04:29 AM
Haha, so with one of these cards you're limited to less than 2GB of system memory in a 32-bit OS. I guess 64-bit just became more necessary.
To me the fact they're going for 2GB proves there won't be memory sharing. Oh well, should be a great card.
Agreed all around.
Redeemed
07-11-08, 05:13 AM
Haha, so with one of these cards you're limited to less than 2GB of system memory in a 32-bit OS. I guess 64-bit just became more necessary.
To me the fact they're going for 2GB proves there won't be memory sharing. Oh well, should be a great card.
Well, you very well could be right, but at the same time the memory can still be shared. If you're running a 32-bit OS and have 2GB of RAM, you could still use up to 1.5GB of the framebuffer. And I'm not sure there's much of a reason why that framebuffer cannot be shared still. The driver can just "disable" 512MB or however much is necessary to fall within the addressable space.
CaptNKILL
07-11-08, 05:25 AM
Well, you very well could be right, but at the same time the memory can still be shared. If you're running a 32-bit OS and have 2GB of RAM, you could still use up to 1.5GB of the framebuffer. And I'm not sure there's much of a reason why that framebuffer cannot be shared still. The driver can just "disable" 512MB or however much is necessary to fall within the addressable space.
By memory sharing he's referring to both of the GPUs using the same memory, rather than 1Gb each like current CF and SLI setups.
There were rumors that the 4870 X2 would share its memory (ie, a 1Gb card with 2 GPUs would actually give you a 1Gb frame buffer rather than 512Mb) but its looking like they are going the other way since they are releasing a 2Gb card (ie 1Gb frame buffer).
Redeemed
07-11-08, 05:39 AM
By memory sharing he's referring to both of the GPUs using the same memory, rather than 1Gb each like current CF and SLI setups.
There were rumors that the 4870 X2 would share its memory (ie, a 1Gb card with 2 GPUs would actually give you a 1Gb frame buffer rather than 512Mb) but its looking like they are going the other way since they are releasing a 2Gb card (ie 1Gb frame buffer).
I understand that. ;) I was merely implying that a portion of the framebuffer could possibly be disabled from the driver level to accomodate the addressable memory requirement. For example, under a 32-bit OS:
System RAM = 3GB. The driver then disables 1.5 GB and the two GPUs share the remaining 512MB of framebuffer (that'd suck :|).
System RAM = 1GB. The driver does not disable any of the framebuffer, and both GPUs share the total 2GB.
System RAM = 2GB. The driver disables 512MB of framebuffer, and the GPUs then share the remaining 1.5GB.
I mean, that is doable, isn't it? I'm sure it'd require some really complex drivers... but I don't see any real reason why it couldn't be done. Or maybe I'm just being too optimistic? :o
CaptNKILL
07-11-08, 05:57 AM
I understand that. ;) I was merely implying that a portion of the framebuffer could possibly be disabled from the driver level to accomodate the addressable memory requirement. For example, under a 32-bit OS:
System RAM = 3GB. The driver then disables 1.5 GB and the two GPUs share the remaining 512MB of framebuffer (that'd suck :|).
System RAM = 1GB. The driver does not disable any of the framebuffer, and both GPUs share the total 2GB.
System RAM = 2GB. The driver disables 512MB of framebuffer, and the GPUs then share the remaining 1.5GB.
I mean, that is doable, isn't it? I'm sure it'd require some really complex drivers... but I don't see any real reason why it couldn't be done. Or maybe I'm just being too optimistic? :o
I think you misunderstood what we're talking about with sharing...
Current X2 cards do not "share" memory. They each use separate banks of memory that store the same thing. 1Gb of memory = 512Mb usable frame buffer.
People thought that the new X2 cards would share memory, in that there would be a single bank of shared memory that both cards would access. 1Gb of memory = 1Gb frame buffer. If this were true, it'd be a huge step for multi-GPU setups.
Because there is a 2Gb card coming out, it is highly unlikely that the memory will be shared giving it a real 2Gb frame buffer because that is completely unnecessary at this point. It'd also needlessly complicate things for 32bit OS users by decreasing their system memory in favor of another gig of video memory that most likely won't be used.
The only possible reason for them to be releasing a 2Gb 4870X2 is that the memory is not shared and that they need 2Gb of memory to give the card a 1Gb frame buffer.
As far as the cards decreasing their framebuffer to give more system memory, it doesn't seem likely. The graphics cards are wired to access their memory a certain way, they would probably lose memory bandwidth if a portion of the memory was disabled.
Redeemed
07-11-08, 06:25 AM
I think you misunderstood what we're talking about with sharing...
Current X2 cards do not "share" memory. They each use separate banks of memory that store the same thing. 1Gb of memory = 512Mb usable frame buffer.
People thought that the new X2 cards would share memory, in that there would be a single bank of shared memory that both cards would access. 1Gb of memory = 1Gb frame buffer. If this were true, it'd be a huge step for multi-GPU setups.
Because there is a 2Gb card coming out, it is highly unlikely that the memory will be shared giving it a real 2Gb frame buffer because that is completely unnecessary at this point. It'd also needlessly complicate things for 32bit OS users by decreasing their system memory in favor of another gig of video memory that most likely won't be used.
The only possible reason for them to be releasing a 2Gb 4870X2 is that the memory is not shared and that they need 2Gb of memory to give the card a 1Gb frame buffer.
As far as the cards decreasing their framebuffer to give more system memory, it doesn't seem likely. The graphics cards are wired to access their memory a certain way, they would probably lose memory bandwidth if a portion of the memory was disabled.
I fully understand how current dual-GPU cards opperate... I'm thinking you're not understanding what I'm saying.
By sharing, I mean this:
Each GPU will have full access to the total available framebuffer. If that framebuffer is 2GB, then each GPU will have full access to all 2GB (sharing the same pool of RAM, if you will). Under a 32-bit OS with 2GB or more of RAM, the full 2GB of framebuffer will not be addressable. In this situation the driver could "disable" however much of the framebuffer is neccessary to fit within the addressable space. Example... with 2GB of system RAM 512MB of framebuffer would be "disabled" by the driver. This then leaves 1.5GB total available framebuffer that each GPU would have full access to. With current dual-GPU cards the 1.5GB I mentioned above would be divided into about 768MB per GPU- each GPU would not see the 1.5GB total but instead only the 768MB it had access to... that's not really sharing the framebuffer at all. With what I'm saying, each GPU on the 4870X2 would have full access to that 1.5GB. Remove 512MB of system RAM and the driver then "enables" the remaining 512MB framebuffer and now each GPU can access the full 2GB. Does that make more sense now?
slaWter
07-11-08, 07:16 AM
Very interesting! Hopefully this is true and we'll see reviews next week!
I might get one just to try it out, maybe I'll keep it and add a second card once I get the X58.
As captn said I'm not sure if that would work seeing it would reduce the bus width. I think people buying these cards just need to go with a 64-bit OS. Still, I'd rather sacrifice system memory over video memory, for gaming, anyway.
Ninja Prime
07-11-08, 11:04 AM
I understand that. ;) I was merely implying that a portion of the framebuffer could possibly be disabled from the driver level to accomodate the addressable memory requirement. For example, under a 32-bit OS:
System RAM = 3GB. The driver then disables 1.5 GB and the two GPUs share the remaining 512MB of framebuffer (that'd suck :|).
System RAM = 1GB. The driver does not disable any of the framebuffer, and both GPUs share the total 2GB.
System RAM = 2GB. The driver disables 512MB of framebuffer, and the GPUs then share the remaining 1.5GB.
I mean, that is doable, isn't it? I'm sure it'd require some really complex drivers... but I don't see any real reason why it couldn't be done. Or maybe I'm just being too optimistic? :o
It doesn't work that way. Priority is given to devices, it will only limit system RAM. If you have 2gigs of cards, the system will limit you to like 1.5gigs of system RAM. Worse yet a tri-sli set up with GTX 280s would only see like 512mb system RAM left. :eek:
They should have made Vista a 64 bit only OS, wouldnt have these problems.:bleh:
hell_of_doom227
07-11-08, 12:12 PM
Vista x64 and XP 64 is a way to go with. Not sure why people bother with 32bit OS.
From Sampsa, great news! :)
Good news! I can confirm that based on my own tests microstuttering is gone on R700!
I've tested with R700 (ATI Radeon HD 4870 X2) and R680 (ATI Radeon HD 3870 X2) in Crysis (1600x1200 and High settings). I used Fraps and enabled Frametimes logging. I recorded 2 seconds from exactly the same point in game (loaded from save game). Based on my recorded data, with ATI Radeon HD 3870 X2 frames are rendered after ~21,5 and every other frame after ~49,5 ms. With ATI Radeon HD 4870 X2 all framres are rendered after ~ 21,9 ms.
http://www.xtremesystems.org/forums/showpost.php?p=3132311&postcount=469
From Sampsa, great news! :)
http://www.xtremesystems.org/forums/showpost.php?p=3132311&postcount=469
YES! :D My next card this is! /yoda
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.