I say its two cards that fit into a single slot with all the limitations of SLI except the aforementioned, and it does not represent taking the crown for single card performance away from ATI's X1900XTX 
And how did they find that out? (whispers: Shhh...they compared them already, that's how they know that it is faster)
If that is truly the case then you really don't know what you're doing here and are simply proving yourself to be a known rambler and flip-flopper. The 7950GX2 IS compared with the X1900 in benchmarks already. So if that is your agenda, then you've already got what you wanted from numerous review sites.
This is why I said you were hopeless. You STILL think that a driverless 7950GX2 is really cool and doing just fine. Without a driver, I can NOT play a game. Period. If Windows XP does not have a video card driver, it can still function properly. You can still open programs, browse the internet (although it will be kind of choppy), still control where your files go, etc. A 7950GX2 cannot play a modern 3D game without a driver and is therefore, useless for its intended purpose without the software.
7950GX2 is a video card, its purpose is to decode and display.
It is the operating system that allows you to play games by sending out the correct information to video card.
It's a shame that the 7950GX2 uses two PCI-E slots then, isn't it.
Addition: Here is a nice and huge picture of the 7950GX2's second PCI-E slot, which proves without a doubt that it is simply one card functioning as an expansion board and a second functioning as an add-on board.
It's a fantastic card, very powerful, it even beats out two 7900GTs for less money... But what some fanbois in this thread need to realise is that it's just two mobile GO 7900 cores on two heavily modified PCBs, nothing more.
If you want to see a true single-card solution involving two GPUs, take a look at the old Galaxy 6600GT and Asus Dual 6800GT, and the upcoming Radeon Gemini.
Rangoric
Senior member
- Apr 5, 2006
- 530
- 0
- 71
Then ATI wouldn't have been able to boost the FPS in various games by 6% with just a driver upgrade.
Software RUNS hardware, and hardware RUNS software. Either one can break the other.
Bad software can blow out your speakers.
Even better.
A bad bios can fry your computer. And your bios is nothing more then software.
Uninstall your video card drivers and attempt playing some games.
Go try it now otherwise you probably wont believe me.
Now what happens is with D3D you are dumped to the REF rasterizer and with OGL you are dumped to Microsoft's software renderer. If the games even start up (OGL ones might not if they are expecting extensions, not sure about D3D ones either).
See this is where the driver comes in handy. The driver directly communicates with the hardware, what the OS can do is extremely limited.
Maybe you should go buy a beginners book to Computers/Hardware/Software/Windows, might help you.
Rangoric
Senior member
- Apr 5, 2006
- 530
- 0
- 71
Even when it has improper drivers for the GC, windows will use its "best fit" driver.
Althouth, I do think it would be a fun exercise to remove all video drivers, and see what windows does (looks at his spare comp that is soon to be a test server).
Hmmmm. Just one more thing to try
numbered by me to make it easier
1. Care to take a guess at what percentage it is where a GX2 doesnt block the second slot?
2. Why would I? Its not even the same thing.
3. I never said it did.
4. Its one point out of several, that you and begger seem to hinge your whole arguement on.
5. I dont care which has the faster card. Its "obviously isnt"? According to the poll, most think the GX2 is two cards slapped together.
6. Once again, you hinge your whole post, and argument on one point out of several in my post. Ill post it again, since you dont seem to want to respond to the whole post.
Once again, stop trying to use only one point of my post, and try thinking about the rest of it. Calling the GX2 the same as a dual core CPU is pretty silly to me. But hey, everyone has their own opinion.
Keysplayr
Elite Member
- Jan 16, 2003
- 21,211
- 50
- 91
I believe he can use any point he wishes to. Or will you just try to force him to do what you want him to do?.... You posted the point(s), now they are each and every one of them, open for debate, and not all necessarily at the same time. Agreed?
BFG10K
Lifer
- Aug 14, 2000
- 22,709
- 3,002
- 126
Maybe you should repeat that to beggerking because he seems to think it's one card.
This is nothing more than semantics. The point is the replication level at both cores is such that it makes them individual CPUs.
This is unlike a HT CPU which simply presents two logical CPUs but doesn't have the same level of replication to physically have two CPUs on one die.
The device ID at the very least would stop you and then there's the fact that Windows couldn't even execute the code given it couldn't even initialize the device properly.
Then if you somehow forced it Windows would probably hang on startup or it would disable the device in the Device Manager stating it's not working properly.
But you would not be operating the device within its standard operating parameters just like you don't operate the 7950 properly when you disable SLI or when you run it under VGA mode.
Thus the hardware has very much to do with the software (and vice versa) unlike the claims here that it doesn't.
BFG10K
Lifer
- Aug 14, 2000
- 22,709
- 3,002
- 126
So you admit we need to change the software the moment we go from a single card to a 7950?
In otherwords the software needs to change because we are now running a 7950 (aka SLI) so by your own admission your "hardware has nothing to do with the software" comment is complete nonsense.
The OS doesn't "send" anything, it simply provides the API interface to interact with the driver which directly interacts with the hardware.
No, the driver does. The OS has nothing to do with SLI, it's all the driver doing the load distribution and programming the hardware.
The OS just provides the necessary layer(s) for programs to interface with the drivers and also to load the hardware to make it an available resource to the system.
Yet by your own admission ("os needs to send "different" data to each GPU") we have to change the software when the 7950 is installed compared to a real single card, proving hardware has very much to do with software (and vice versa).
To the Nvidiots : Wake up you boneheads two cards in one. No matter how you'll try to manipulate the situation the fact remains that we are talking about "two cards"
To the fanATIcs: Who gives a fook if it is a single card solution? If you blame Nvidiots being unreasonable and out of reality because they want to count it as a single card solution so that Nvidia can have the high end lead, then you're equally retarded for answering to such a claim... PPL DO have judgement you know..
I am an idiot for posting to this vicious circle of stupidity.. lol
Reagardless of all these nonsense 7950GX2 is a high performance "card/s" at a very good price. If I didn't care for the I.Q (but unfortunately I do when I pay big bucks and that's why I went ATI this time) and I was to buy a gpu now I'd probably have taken it.. But others find the IQ differences too unimportant so they find this card a great offer. Who the fook honestly cares if it should be considered a single or a dual card? Mercy!!
To the fanATIcs: Who gives a fook if it is a single card solution? If you blame Nvidiots being unreasonable and out of reality because they want to count it as a single card solution so that Nvidia can have the high end lead, then you're equally retarded for answering to such a claim... PPL DO have judgement you know..
I am an idiot for posting to this vicious circle of stupidity.. lol
Reagardless of all these nonsense 7950GX2 is a high performance "card/s" at a very good price. If I didn't care for the I.Q (but unfortunately I do when I pay big bucks and that's why I went ATI this time) and I was to buy a gpu now I'd probably have taken it.. But others find the IQ differences too unimportant so they find this card a great offer. Who the fook honestly cares if it should be considered a single or a dual card? Mercy!!
Lets assume I wrote a driver which is specifically designed to proove the point
I never said it would work, I was just pointing out you could use one if you wanted to. Perhaps this is what was meant by the "hardware has nothing to do with software" comment? The hardware will just respond to whatever you tell it to do and has no knowlege of the driver except by it's output.
I'm not here to defend the comment though, "hardware has nothing to do with software" seems like a dumb thing to say in any case.
Sure he can. Its just silly to ignore all the other points, and focus only on one. The only one the each of them have tried to refute, was the two slot point. Nothing on the fact that a dual core CPU looks the same as a single core CPU from the outside, nothing about the GX2 having two PCB's, GPU's, fans, sets of ram, etc. If they're going to try and say a dual core CPU is the same as a GX2, then they need to look at more than they both just use one "slot". Its really pretty easy.
BFG10K
Lifer
- Aug 14, 2000
- 22,709
- 3,002
- 126
That's the whole point though. What use is a driver if it doesn't work?
Can it even be called a driver if it can't operate the device at all?
He's using the comment because he's playing semantic games in his posts.
He ignores the software when it suits his agenda but then turns around and admits himself the driver (err "programming", "data path", or whatever he chooses to call it) needs to change when the hardware changes.
Then he flip-flops back to the OS, "hardware has nothing to do with software" and the cycle begins anew.
Argueing about this would be even more pointless than argueing about whether the 7950 is two "cards".
I declare this thread pointless. *goes back to OT*
Thankyou for that summurization, but I don't see the last [insert correct # of pages because I don't want to bother finding out and I'm sure no one else does] being about Nvidiots nor fanATIcs. Instead, it seems to mainly be trying to understand the magnificent hardware that can function to its fullest with or without software that beggerking is trying to invoke. Everything else in between was simply him getting on the "technical term" bandwagon in an attempt to distort his primary argument.
If indeed he would have shown how to do HT on an A64, I would have apologized for my doubt as well as praised him for the find.
If indeed he would have shown how the 7950GX2 can function on the level that a realistic purchaser would want it to (an enthusiast level) without drivers nor software for that matter, I would once again attempt to pry my foot from my mouth and apologize as well as praise. I'm always eager to learn new things even if the majority of the people don't deem it to be possible--just as long as it actually is possible and beneficial.
But, he didn't do that. When answering how to do HT on an A64, he merely told me how to find out how to do that. He didn't provide any method whatsoever. In fact he only later stated that it would be useless to try and achieve that since the A64 itself wouldn't allow it, even with the right approach (i.e. emulator).
When answering how software can be "fooled" into thinking that there are more components present than there actually are, the explanations he gave became pointless since there yeilds no gain in performance when "fooling" it.
Then when telling us his primary agenda, which was only to COMPARE the 7950GX2 (not claim victory, but just to compare it) to other single/GPU's/cards/whatever else is single ESPECIALLY the X1900XTX, he must have forgot that that has already happened back when the card first debuted: http://www.anandtech.com/video/showdoc.aspx?i=2769
So, why did he argue for so long only to come to the desire to see the 7950GX2 COMPARED to the X1900XTX, something that had already happened? My reasoning: Because he knew he was wrong for the "SOFTWARE HAS NOTHING TO DO WITH HARDWARE!!!!" comment and wanted to try to ramble his way out of it rather than admit it. But everyone has their own speculations as to why I'm sure.
Your analogy is flawed beggerking. GPUs and CPUs are entirely different monsters, direct comparasons between the two cannot be made. A dual core CPU (AMD ones) are two seperate processing engines made on the same die. For Intel, (the 65nm netburst stuff) the processors are actually on two seperate pieces of silicon and are merely mounted to the same piece of substrate. They are really a double CPU package and not really dual core. What does single CPU mean? The same die? How many processing engines it has? How many pieces of substrate there are?
The 7950GX2, has two (2) seperate processing engines on two (2) different substrates, on two (2) different PCBs.
Beggerking, do you consider a Dual Socket Opteron motherboard, filled with two Single core CPUs (dual processor SMP system and let's assume that the second processor shares ram with the first one) a single processor system?
No? Niether are dual core processors. Sure they are manufactured on the same die (AMD X2s), but they are two seperate processors that behave in almost exactly the same way as a dual processor SMP system does. The GX2 doesn't have the luxury of having everything on the same die, nor can it say that everything it only has 1 substrate, it can't even boast of having only one (1) PCB.
On the other hand, I don't consider the 7950GX2 to be double card either. It only has 1 PCI-e connector (but that doesn't mean much, because PCI-Express is a serial connection so it can be split up and divided in many ways). I think of the 7950GX2 as a 1.5-1.75 card deal.
It suffers from the same limitations as a dual card setup. Namely, these two come to mind.
--If the game doesn't support SLI, then you'll see performance similar to a 7900GT.
--Each GPU has it's own framebuffer, yes there is 512MB of memory on each of the two PCBs, but total usable memory is 512MB, NOT 1GB.
it is the OS that sends out data, driver is only a encoder that encodes data so hardware can understand.
"The driver object is a portion of memory allocated by Windows, which describes where the actual driver is loaded into memory plus contains information readable by Windows about the driver"
"When you decide to print a document or connect to the Internet, Windows will examine the Function Dispatch Table to see which major function code will be able to fulfil the request. "
Text[
Correct. But I'm referring to " A 7950GX2 cannot play a modern 3D game without a driver and is therefore, useless for its intended purpose without the software."
7950gx2 is a hardware, it doesn't play games.
its the software( Moder 3d game) that would be useless without a driver.
we did the same when we went from single core to dualcore. I consider dualcore a single CPU.
software doesn't "need" to change for the hardware to work. hardware works as it is. software, on the other hand, changes to improve performance.
There is somthing within that text link of yours beggerking that you failed to point out:
Are you going to argue with your own source now too? Let me exaggerate what your own evidence points out:
Did you get that?
The OS may have its own role to play in the driver and the hardware, but software by no means has nothing to do with hardware like you keep defending. They must exist together in order to be beneficial.
Stop arguing against it already, do you not yet notice you are the ONLY one here that is defending your bogus claim of software and hardware being independent from one another? That says something. If it were really true, don't you think more would be in here at your defense?
Last get the simple perspective out of the way first. IF by single card you mean *SINGLE POINT OF INTERACTION* (PCIe slot) then yes, 7950GX2 is a single "card".
However this simplistic approach is problematic (at the very least). For example my priority (DB client) "thinks" that my DB server is a single device and my browser "thinks" the internet is a single being, for my browser IP is only the logical mapping of the device just like the memory consumed by an application is viewed by the application itself like just one single memory chip.
It (the application) doesn?t know and doesn?t care if there is an abstraction layer (the OS) that "makes it look" like a single device, it gets the logical address of the memory as it is presented by the OS, just like IP in a TCP/IP network, as apposed to MAC address/physical memory address.
Additionally my browser can work with the more then one computer with the same "physical" IP address through NAT but for my browser the two computers/network devices will have a different logical IP address, just like an application will see a different data on the same physical memory address at different points in time because the OS did some memory swap under the hood at some point, but for the application the logical address of the two blokes of data will always be different.
(I know the IP/internet example is a stretch and probably highly debatable but -> ) The point of the last three paragraph is that whenever an abstraction layer exist (in our case there are two abstraction layers which I will get to later) defining/categorizing 7950GX2 as a single/multi card doesn?t cut it anymore. A further analysis of the product and any abstraction layer that the products (or some of its function) utilize is in need, and maybe even a new category/set of definition will be required.
The MAIN problem here is NOT if this is single card (PCB) because that doesn?t matter that much, you CAN build the same final product (7950GX2) on a SINGLE PCB while retaining its MAIN characteristics (END inputs and outputs, work being done).
(Btw when old graphics adaptors where able to receive memory upgrades/add-on in the form of an additional PCB is was never considered a dual PCB card, some of the people here let themselves get confused by fact that the additional PCB uses an additional slot space, I can only hope we all agree here that 7950GX2 is a dual slot solution.)
The MAIN problem here IS the fact that there are two abstraction layers at two different levels, and if that wasn?t enough there is some (very little) overlapping between the two.
The first abstraction layers level (lets call it the hardware or physical level) exist in the form of the PCIe bridge, it changed the U connection schema to a Y (sometime also referred to a T) schema. Its advantages are two fold, it takes only one PCIe (x16) slot, and it eliminate the need for the chipset to support the PCIe to PCIe full duplex communication (read AND write from PCIe device to PCIe device, as apposed to PCIe device to PCIe controller) that SLI needs to function properly, thus eliminate the need for a SLI capable/supporting chipset.
Now there is an additional comment to be made here, theoretically, given a big enough PCIe bridge, you can build a single card that have a GPU and DSP designed for audio processing AND a USB and/or a Firewire hub chip AND a network controller and so on and so on?(you get the point).
The second Nvidia engineers opted for a PCIe bridge that is the second that single/dual card debate has taken on a whole new meaning, and probably the second that the term "card" has stop being good enough to define (or discriminate from others) the 7950GX2.
Hell I remember when there was a PCI card that hosted a propriety PCI-to-ISA bridge which hosted on top of that an I/O chip (plain old I/O) that the OEM didn?t want to redesign for the PCI interface, it had bios extension for DOS (just like SCSI card have) and a DOS driver (which was the SAME binary that was used for the original ISA card) for a DOS application (which controlled an outside industrial box) so is that an ISA card or a PCI card?
And for the second abstraction layer AKA SLI, well I'm no expert but it look to me like the game/application sees the driver, or whatever the driver present to him (don?t forget that somewhere in the middle there is also the directX layer), and the driver see 2 GPU, the driver don?t see card(s) it see GPUs(!), in what way the presence of the PCIe bridge change the way SLi works? I mean every GPU has distinct physical lanes to the rest of the system, the only difference is in the convergence point of the lanes, when the data need to get to or from the CPU or/and the main memory to GPU(s) it (always) moves through a single link to the chipset, in old/regular SLI the convergence point is in the chipset logic and in 7950GX2 the convergence point is in the PCIe bridge.
Now about the CPU analogies, IMHO there isn?t any point comparing the 7950GX2 to Intel dual core CPUs, which work a lot more like old/regular SLI. AMD Athlon 64 X2 processor on the other and, because of the onchip crossbar switch, resemble 7950GX2 much more from a schematic POV (crossbar switch == PCIe bridge).
However this simplistic approach is problematic (at the very least). For example my priority (DB client) "thinks" that my DB server is a single device and my browser "thinks" the internet is a single being, for my browser IP is only the logical mapping of the device just like the memory consumed by an application is viewed by the application itself like just one single memory chip.
It (the application) doesn?t know and doesn?t care if there is an abstraction layer (the OS) that "makes it look" like a single device, it gets the logical address of the memory as it is presented by the OS, just like IP in a TCP/IP network, as apposed to MAC address/physical memory address.
Additionally my browser can work with the more then one computer with the same "physical" IP address through NAT but for my browser the two computers/network devices will have a different logical IP address, just like an application will see a different data on the same physical memory address at different points in time because the OS did some memory swap under the hood at some point, but for the application the logical address of the two blokes of data will always be different.
(I know the IP/internet example is a stretch and probably highly debatable but -> ) The point of the last three paragraph is that whenever an abstraction layer exist (in our case there are two abstraction layers which I will get to later) defining/categorizing 7950GX2 as a single/multi card doesn?t cut it anymore. A further analysis of the product and any abstraction layer that the products (or some of its function) utilize is in need, and maybe even a new category/set of definition will be required.
The MAIN problem here is NOT if this is single card (PCB) because that doesn?t matter that much, you CAN build the same final product (7950GX2) on a SINGLE PCB while retaining its MAIN characteristics (END inputs and outputs, work being done).
(Btw when old graphics adaptors where able to receive memory upgrades/add-on in the form of an additional PCB is was never considered a dual PCB card, some of the people here let themselves get confused by fact that the additional PCB uses an additional slot space, I can only hope we all agree here that 7950GX2 is a dual slot solution.)
The MAIN problem here IS the fact that there are two abstraction layers at two different levels, and if that wasn?t enough there is some (very little) overlapping between the two.
The first abstraction layers level (lets call it the hardware or physical level) exist in the form of the PCIe bridge, it changed the U connection schema to a Y (sometime also referred to a T) schema. Its advantages are two fold, it takes only one PCIe (x16) slot, and it eliminate the need for the chipset to support the PCIe to PCIe full duplex communication (read AND write from PCIe device to PCIe device, as apposed to PCIe device to PCIe controller) that SLI needs to function properly, thus eliminate the need for a SLI capable/supporting chipset.
Now there is an additional comment to be made here, theoretically, given a big enough PCIe bridge, you can build a single card that have a GPU and DSP designed for audio processing AND a USB and/or a Firewire hub chip AND a network controller and so on and so on?(you get the point).
The second Nvidia engineers opted for a PCIe bridge that is the second that single/dual card debate has taken on a whole new meaning, and probably the second that the term "card" has stop being good enough to define (or discriminate from others) the 7950GX2.
Hell I remember when there was a PCI card that hosted a propriety PCI-to-ISA bridge which hosted on top of that an I/O chip (plain old I/O) that the OEM didn?t want to redesign for the PCI interface, it had bios extension for DOS (just like SCSI card have) and a DOS driver (which was the SAME binary that was used for the original ISA card) for a DOS application (which controlled an outside industrial box) so is that an ISA card or a PCI card?
And for the second abstraction layer AKA SLI, well I'm no expert but it look to me like the game/application sees the driver, or whatever the driver present to him (don?t forget that somewhere in the middle there is also the directX layer), and the driver see 2 GPU, the driver don?t see card(s) it see GPUs(!), in what way the presence of the PCIe bridge change the way SLi works? I mean every GPU has distinct physical lanes to the rest of the system, the only difference is in the convergence point of the lanes, when the data need to get to or from the CPU or/and the main memory to GPU(s) it (always) moves through a single link to the chipset, in old/regular SLI the convergence point is in the chipset logic and in 7950GX2 the convergence point is in the PCIe bridge.
Now about the CPU analogies, IMHO there isn?t any point comparing the 7950GX2 to Intel dual core CPUs, which work a lot more like old/regular SLI. AMD Athlon 64 X2 processor on the other and, because of the onchip crossbar switch, resemble 7950GX2 much more from a schematic POV (crossbar switch == PCIe bridge).
Keysplayr
Elite Member
- Jan 16, 2003
- 21,211
- 50
- 91
All I know is, if I build an AM2 system, using a motherboard with only one PCI-e slot, I can use a 7950GX2.
If the GX2 covers a PCI slot, what's the big deal? We have been dealing with two slot coolers since 5800U/X800's. So is there any particular reason why a GX2 should be penalized for taking up a PCI slot? Answer is no. So, here I am with a motherboard that only has one PCI-e slot and in it sits a GX2. It can be spun for a month of Sunday's, but it is still a single card solution. For the simple fact alone that people with single PCI-e slot motherboards can happily use a GX2. Because it is physically impossible to insert two independant PCI-e video cards into a motherboard with only one PCI-e slot. This is the thing you cannot get around, even with the most ambitious spinners at ATI's disposal. We are not dumb, and can see for ourselves that a GX2 is a single card solution.
I am not spinning here. Just simple, obvious and uncontestable facts. One card solution, in one PCI-e slot.
Spin away if you wish. Not interested in addressing any points you feel are being ignored because frankly, they have no bearing on the final outcome.
EDIT: There are currently no less than 73 (count them) Nforce 4 motherboards with single PCI-e x16 slots available on newegg. That is just Nforce4 chipsets. There are countless others with varying chipsets with single PCI-e x16 slots. They all have the same limitation, being able to support only one PCI-e video card. And all of them will accept a GX2.
So, while the rest of you folks continue to debate technical points on why the GX2 should be considered two cards, lets stay in the real world where end users with single slot PCI-e motherboards could enjoy a GX2 if they so desired. My work is so done here.
Enjoy.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter