Sign in with
Sign up | Sign in
Your question

idea....

Last response: in CPUs
Share
July 10, 2006 2:24:40 AM

what if on a mobo there would be a socket for the cpu and another socket for the gpu and the gpu would come as a chip and a heatsink/fan just like any cpu. and there would be slots for ddr1/2/or better memory and gddr/1/2/3/or better memory and the graphics memory would come like system memory sounds stupid....any comments on it

More about : idea

July 10, 2006 2:44:36 AM

are you interested in amd's upcomming ideas by any chance?
Related resources
July 10, 2006 2:47:09 AM

but u could like say have a 6600 with 1gb or a 7900gt with 2gb u could mix and match lol....this is really becomming a stupid idea rite lol
July 10, 2006 2:54:35 AM

While that may seem like a good idea there are a few things that restrict that. One issue is access to memory. The GPU benefits from having a large amount of memory very close by so it can run at high speeds. The graphics memory is usually ahead of standard RAM. Having a chip only solution would either strain the memory bus or these parts would need to start integrating cache onto the chip.
July 10, 2006 2:57:15 AM

o.........didnt thing about that
July 10, 2006 2:58:55 AM

A graphics processor is essentially a second CPU, except it's used to process graphics. A graphics card is basically a daughter card with an integrated CPU and memory which plugs into a slot on the motherboard.
July 10, 2006 4:04:11 AM

i think hes striking at the idea of upgradeable builtin components. Peer inside your case, past the lucite window and look at your motherboard. Glance to your videocard and now imagine it melting into your motherboard, and your motherboard growing by about 36 square inches. What do you suppose all those nifty bits and pieces on your videocard are for, and where are they going to go?
July 10, 2006 4:23:36 AM

It's not a good idea for a couple reasons:

1. Airflow would be poor
2. VIVO, DVI and a slew of other connections would have to be standardized and put on the motherboard. (unless u use a PCI connector plate)
3. GPU's and memory are designed to work together more so than CPU/Memory. In other words, nVidia/ATI put a lot of work into which specific memory chips work best w/ the GPU. If you have DIMM slots on the mobo for the GPU, anyone could make the memory and you'd have performance/compatibility issues.
5. real estate on ATX/BTX motherboards is already slim, there's no room for 2/4 more DIMM slots and a socket for a GPU.
4. there's no technological need for it PCI-E 16x is plenty fast enough and PCI-E 32x is already being worked on. There are no advantages to your idea.

Nevertheless, it was an interesting one.
July 10, 2006 5:28:07 AM

Quote:
It's not a good idea for a couple reasons:


I'm not advocating a socketed GPU just yet, but will play devil's advocate for fun:

Quote:
1. Airflow would be poor


Workarounds wouldn't have to be all that difficult. It's not like GPU HSFs are the most elegant solution imagineable.

Quote:
2. VIVO, DVI and a slew of other connections would have to be standardized and put on the motherboard. (unless u use a PCI connector plate)


AMD might have chosen 4x4 to bring RHT and to be first to set a GPU socket standard.

Quote:
3. GPU's and memory are designed to work together more so than CPU/Memory. In other words, nVidia/ATI put a lot of work into which specific memory chips work best w/ the GPU. If you have DIMM slots on the mobo for the GPU, anyone could make the memory and you'd have performance/compatibility issues.


So sell the DIMM with the socketed GPU or put the memory on-die.

Quote:
5. real estate on ATX/BTX motherboards is already slim, there's no room for 2/4 more DIMM slots and a socket for a GPU.


Is there a need for more than one DIMM slot ? You don't need more than 256 or 512 MB do you? Maybe a new kind of compact slot? How about a 256 or 512 chip or chip array on the mobo? I'm not claiming this will work, just asking if it is feasible.

Quote:
4. there's no technological need for it PCI-E 16x is plenty fast enough and PCI-E 32x is already being worked on. There are no advantages to your idea.


What if AMD has bought into Physix in a big way and are looking to have a more efficient means of CPU to GPU to PCP communication? Again, I'm just asking questions. Clearly, increasing GPU clock and graphics memory speed is of interest. If graphics memory capacity or speed is decoupled from the GPU, maybe that is considered an advantage by ATI or AMD? Each time I think about 4x4, I wonder if AMD has been collectively drinking too much or if they actually have something really cool headed down the pipe. It's hard for me to accept that they lost the ability to find intelligent solutions.
July 10, 2006 5:52:48 AM

Quote:
Each time I think about 4x4, I wonder if AMD has been collectively drinking too much or if they actually have something really cool headed down the pipe. It's hard for me to accept that they lost the ability to find intelligent solutions.


What do you think 4x4 is? There is no special hardware except for a desktop motherboard that supports dual processors. This is already the standard in the server market. And 4x4 will not have RHT! It will have 2 sockets, and therefor RHT is not possible unless it is done in software, and in this case, it would be possible on Intel CPUs too, therefor it would not add any advantage to AMD that it doesn't add to Intel. For RHT to be done on the CPU, it would have to be done on a single die, not across multiple sockets.
July 10, 2006 5:58:04 AM

Quote:
Each time I think about 4x4, I wonder if AMD has been collectively drinking too much or if they actually have something really cool headed down the pipe. It's hard for me to accept that they lost the ability to find intelligent solutions.


What do you think 4x4 is? There is no special hardware except for a desktop motherboard that supports dual processors. This is already the standard in the server market.

How do you know that 4x4 will mimic server mobo design?

Quote:
And 4x4 will not have RHT!


How do you know this? I'm not saying it will but speculation is on the rise. Yes, yes, rumors are meaningless, but what do we really know right now and what is speculation?

Quote:
[For RHT to be done on the CPU, it would have to be done on a single die, not across multiple sockets.


Why is it mandatory to be single socket?
July 10, 2006 6:28:17 AM

Quote:
Each time I think about 4x4, I wonder if AMD has been collectively drinking too much or if they actually have something really cool headed down the pipe. It's hard for me to accept that they lost the ability to find intelligent solutions.


What do you think 4x4 is? There is no special hardware except for a desktop motherboard that supports dual processors. This is already the standard in the server market.

How do you know that 4x4 will mimic server mobo design?

Quote:
And 4x4 will not have RHT!


How do you know this? I'm not saying it will but speculation is on the rise. Yes, yes, rumors are meaningless, but what do we really know right now and what is speculation?

Quote:
[For RHT to be done on the CPU, it would have to be done on a single die, not across multiple sockets.


Why is it mandatory to be single socket?

Because, thats what AMD has published! It is 2 AMD dual core processors. That means that there are 2 CPU sockets. If RHT is going to be implemented because 4x4 is NOT a new chip!!! Therefor we KNOW that it will not be implemented on 4x4.

If and when It is implemented, it is mandatory to be on a single socket because it is physically IMPOSSIBLE. It is either on the CPU or not.
July 10, 2006 7:00:18 AM

Intel aims for 32 cores by 2010:
Quote:
Chicago (IL) and Westlake Village (CA) - Five years ago, Intel envisioned processors running at 20 GHz by the end of this decade. Today we know that the future will look different. CPUs will sacrifice clock speed over core count: The "Gulftown" processor, Intel's first "many core" CPU, will run at only two thirds of the clock speed of today's fastest Xeon CPU - but achieve 15x the performance, thanks to 32 core

http://www.tgdaily.com/2006/07/10/intel_32_core_processor/
July 10, 2006 8:12:16 AM

Quote:
Workarounds wouldn't have to be all that difficult. It's not like GPU HSFs are the most elegant solution imagineable.


Where is it going to go?

Quote:
So sell the DIMM with the socketed GPU or put the memory on-die.


Its going to be tough to have a 256bit bus with DIMM slots, on-die or substrate.

Quote:
What if AMD has bought into Physix in a big way and are looking to have a more efficient means of CPU to GPU to PCP communication?


Its not a bottleneck on AGP so whats there to do?

Quote:
If graphics memory capacity or speed is decoupled from the GPU


????
July 10, 2006 1:03:17 PM

Quote:
Workarounds wouldn't have to be all that difficult. It's not like GPU HSFs are the most elegant solution imagineable.


Where is it going to go?

You can ship it out a couple of slots like a 1900XT or out the case door like the Lian Li 60plus if you want to use existing kinds of solutions. It would be possible to push it towards a blowhole. Problem is, to come to any CPU HSF kind of solution, you'd either need to target the mobo layout to a specific kind of case design or get a case maker to work with you. That's not been done before that I know of, but moving hot air around a metal box is simple tech.

Quote:
So sell the DIMM with the socketed GPU or put the memory on-die.


Quote:
Its going to be tough to have a 256bit bus with DIMM slots, on-die or substrate.


OK, difficult. Do you think either is possible?

Quote:
What if AMD has bought into Physix in a big way and are looking to have a more efficient means of CPU to GPU to PCP communication?


Quote:
Its not a bottleneck on AGP so whats there to do?


That's part of my question. If for example Physix is a focus, is a multi-socket approach part of the solution? I'm not claiming to have the answers, just wondering if we might be mising something. Sure, the talk may be baseless rumors, but what if it isn't. Just brainstorming, that's all.

Quote:
If graphics memory capacity or speed is decoupled from the GPU


????[/quote]

Meaning that if you can update graphics memory independantly of GPU or vice versa, as new tech arrives, that's a potential performance advantage. Look at it another way - if one presupposes that game programmers will continue their quest for more realistic games, what GPU, memory, coprocessor, etc., changes will be needed to keep up? How close are we to AGP/PCI-e-based designs from hitting the wall? Are there multisocket approaches that will help for projected needs? For my own personal needs, current hardware and software are almost there, but that doesn't mean that the industry will stop progressing.
July 10, 2006 5:32:34 PM

wouldnt u think that the comp would run cooler because there wouldnt be that huge card blocking all the air....
July 10, 2006 6:23:41 PM

I like the idea to some extent. If there was a way to just swap the chip out and add more GPU ram as an upgrade.

You wouldn't have to standardize ports because you could just have one of those little cables that attach to the card and then to a faceplace with all the outputs the card supports.

Sounds good to me.
- The motherboard would be more efficient
- less non-soldered connections
- cheaper
- upgradeable ram
- easier to add bigger heatsinks
July 10, 2006 6:36:01 PM

Quote:
Because, thats what AMD has published! It is 2 AMD dual core processors. That means that there are 2 CPU sockets. If RHT is going to be implemented because 4x4 is NOT a new chip!!! Therefor we KNOW that it will not be implemented on 4x4.


I appreciate your feedback, but still I'll just sit back and see how it unfoilds.

Quote:
If and when It is implemented, it is mandatory to be on a single socket because it is physically IMPOSSIBLE. It is either on the CPU or not.


What specifically are you saying is physically impossible?
July 10, 2006 7:22:11 PM

think i should submit it to nvidia?haha
July 10, 2006 7:25:28 PM

ya just a little connector on the mobo that connects to a dvi/vga/s-video/etc on a face plate
July 10, 2006 7:27:28 PM

way easier to watercool just pop a little water block on
July 10, 2006 7:30:22 PM

and this would all fit on a btx mobo so there wont have to be any new cases that they would have to make
a b à CPUs
July 10, 2006 8:17:26 PM

Is this all really necessary? I mean correct me if I am wrong, I thought the PCI-Express bus was far from saturated.
July 10, 2006 10:08:09 PM

Quote:
you'd either need to target the mobo layout to a specific kind of case design or get a case maker to work with you.


Thats the problem.

Quote:
OK, difficult. Do you think either is possible?


Well sure but its going to be expensive to boot.

Quote:
That's part of my question. If for example Physix is a focus, is a multi-socket approach part of the solution? I'm not claiming to have the answers, just wondering if we might be mising something. Sure, the talk may be baseless rumors, but what if it isn't. Just brainstorming, that's all.


Well PCIe is faster and can send data both ways. So far the only thing thats been used for is budget cards and equiping them with less memory and using system memory. The physics card is dead in the water the other solutions seem much better.

Quote:
Meaning that if you can update graphics memory independantly of GPU or vice versa, as new tech arrives, that's a potential performance advantage.


That'd also be a pain in the ass for devs and ATI/Nvidia.

Quote:
How close are we to AGP/PCI-e-based designs from hitting the wall?


No where near it. 65nm, GDDR4, better PCIe are all on the way.
July 10, 2006 10:23:41 PM

Quote:
Workarounds wouldn't have to be all that difficult. It's not like GPU HSFs are the most elegant solution imagineable.


Where is it going to go?

You can ship it out a couple of slots like a 1900XT or out the case door like the Lian Li 60plus if you want to use existing kinds of solutions. It would be possible to push it towards a blowhole. Problem is, to come to any CPU HSF kind of solution, you'd either need to target the mobo layout to a specific kind of case design or get a case maker to work with you. That's not been done before that I know of, but moving hot air around a metal box is simple tech.

Quote:
So sell the DIMM with the socketed GPU or put the memory on-die.


Quote:
Its going to be tough to have a 256bit bus with DIMM slots, on-die or substrate.


OK, difficult. Do you think either is possible?

Quote:
What if AMD has bought into Physix in a big way and are looking to have a more efficient means of CPU to GPU to PCP communication?


Quote:
Its not a bottleneck on AGP so whats there to do?


That's part of my question. If for example Physix is a focus, is a multi-socket approach part of the solution? I'm not claiming to have the answers, just wondering if we might be mising something. Sure, the talk may be baseless rumors, but what if it isn't. Just brainstorming, that's all.

Quote:
If graphics memory capacity or speed is decoupled from the GPU


????

You're missing the point. The point I was trying to make is there's no reason to put the GPU in a socket on the motherboard. Have you any idea how much work and money everyone would have to spend to make it happen? PCI-E 32x will provide more than enough bandwidth for years to come. And when it reaches its limit, someone will come up with a new and faster bus. That's the way its worked for going on 20 years now. Why do we keep using this design cycle? Because it works and it's cheap.

You post suggests creating a whole new motherboard layout & design, standardized DIMM design, standardized socket, and a slew of other stuff that simply won't happen.

I know all this is theoretical and you just playing devil's advocate... but let's inject reality into the equation for just one second.
July 10, 2006 10:30:40 PM

Quote:
and this would all fit on a btx mobo so there wont have to be any new cases that they would have to make


Really?

Show me how.

Oh and when you show me, make sure take into account how you plan to fit a heatsink/fan in the case that's BIGGER than a standard CPU's.

Oh and another thing, show me how to you plan to design the whole thing so that it fits in or around the BTX air flow duct.

Remember now, it has to work in a standard BTX case already on the market.

Good luck with that.
July 11, 2006 2:40:31 AM

Quote:
Because, thats what AMD has published! It is 2 AMD dual core processors. That means that there are 2 CPU sockets. If RHT is going to be implemented because 4x4 is NOT a new chip!!! Therefor we KNOW that it will not be implemented on 4x4.


I appreciate your feedback, but still I'll just sit back and see how it unfoilds.

Quote:
If and when It is implemented, it is mandatory to be on a single socket because it is physically IMPOSSIBLE. It is either on the CPU or not.


What specifically are you saying is physically impossible?

I am referring to RHT. You can not do it across multiple packages unless it is implemented in software (in which case it can be applied to both AMD and Intel). And no need to wait to see how it unfolds when AMD has already published documents defining the 4x4 system, but have fun waiting for what everyone else already knows.
July 11, 2006 3:09:50 AM

Does that RHTing require sprinkling of fairy dust? :lol: 
July 11, 2006 3:37:45 AM

Quote:
Does that RHTing require sprinkling of fairy dust? :lol: 


Well, its just been reported that RHT wont be included in K8L...... maybe they'll be able to work some magic and get it into the 4x4 though.. :roll:
July 11, 2006 4:47:14 AM

Quote:
Does that RHTing require sprinkling of fairy dust? :lol: 


Nonono... i think you're confusing RHT with Intel's 32 core chip by 2010. That requires the fair dust my friend.

;-)
July 27, 2006 3:06:41 PM

Quote:
You're missing the point. The point I was trying to make is there's no reason to put the GPU in a socket on the motherboard. Have you any idea how much work and money everyone would have to spend to make it happen? PCI-E 32x will provide more than enough bandwidth for years to come. And when it reaches its limit, someone will come up with a new and faster bus. That's the way its worked for going on 20 years now. Why do we keep using this design cycle? Because it works and it's cheap.

You post suggests creating a whole new motherboard layout & design, standardized DIMM design, standardized socket, and a slew of other stuff that simply won't happen.

I know all this is theoretical and you just playing devil's advocate... but let's inject reality into the equation for just one second.


Well, 4x4 might not have a socketed GPU anytime soon, but check out this and other comments:

"Today, ATI sees the merger with AMD as critical to finally being able to wrest control of the entire PC, to be able to shape all of it to suit its architectural goals, making compromises with no one. It sees this move as potentially elevating the role of the GPU to a position alongside the CPU in both priority and stature, such that the chip responsible for processing any given task may become inconsequential to most users. For AMD, the merger is an acquisition of a key element of intellectual property that a platform producer would need. It enables the co-opting of the GPU for purposes neither company might have been able to undertake, even acting jointly. But that's just a bonus, an extra payoff that may come after AMD integrates the GPU into its platform strategy."

in this article. It makes it look to me like AMD/ATI will have a socketed GPU/CPU, possibly with radically new architecture at the CPU, GPU and chipset level in the future. This is the kind of direction of discussion I was trying to promote. Look back and see how this was shot down in this thread. Now, just a couple of weeks later, it looks like at least part of the industry is headed towards change.
July 27, 2006 3:22:15 PM

Quote:
Because, thats what AMD has published! It is 2 AMD dual core processors. That means that there are 2 CPU sockets. If RHT is going to be implemented because 4x4 is NOT a new chip!!! Therefor we KNOW that it will not be implemented on 4x4.


I appreciate your feedback, but still I'll just sit back and see how it unfoilds.

Quote:
If and when It is implemented, it is mandatory to be on a single socket because it is physically IMPOSSIBLE. It is either on the CPU or not.


What specifically are you saying is physically impossible?

think out side of the box, each cpu could have rht, thus making it seem like 2 cores not 4 and each core having the power of a dual core, but i don't believe it will happen anyway but i will wait and see.

I am referring to RHT. You can not do it across multiple packages unless it is implemented in software (in which case it can be applied to both AMD and Intel). And no need to wait to see how it unfolds when AMD has already published documents defining the 4x4 system, but have fun waiting for what everyone else already knows.
July 27, 2006 8:40:06 PM

http://www.anandtech.com/video/showdoc.aspx?i=2570

Quote:
They claim that NVIDIA is working on flip chip GPU sockets for motherboards. Apparently, inside NVIDIA engineering teams have several prototypes where the GPU, rather than the CPU, is the main focus of a motherboard with two sockets: one for the GPU and another for the CPU.



My biggest problem is that you would have to change your motherboard for every graphics platform (non-incremental) upgrade. Example - Say, ATI is using GDDR3, their next big platform will use GDDR4. An update to the memory controller is not something you can do in a bios update (hence we have AM2). Add to that, it looks as if physics might be pushed onto the gpu around dx11. I don't think a mobo designer can predict what traces will be needed in the future.

Additionally, we have the mobo noise factor. Look how sound performs better as a seperate card, rather than integrated.

I think an upgradable socket is good for IGP systems, but not performance or gaming rigs.
July 27, 2006 8:53:25 PM

Quote:
http://www.anandtech.com/video/showdoc.aspx?i=2570

They claim that NVIDIA is working on flip chip GPU sockets for motherboards. Apparently, inside NVIDIA engineering teams have several prototypes where the GPU, rather than the CPU, is the main focus of a motherboard with two sockets: one for the GPU and another for the CPU.



My biggest problem is that you would have to change your motherboard for every graphics platform (non-incremental) upgrade. Example - Say, ATI is using GDDR3, their next big platform will use GDDR4. An update to the memory controller is not something you can do in a bios update (hence we have AM2). Add to that, it looks as if physics might be pushed onto the gpu around dx11. I don't think a mobo designer can predict what traces will be needed in the future.

Additionally, we have the mobo noise factor. Look how sound performs better as a seperate card, rather than integrated.

I think an upgradable socket is good for IGP systems, but not performance or gaming rigs.

I think the AMD/ATI project is looking towards mobile, right?
July 27, 2006 9:16:45 PM

Quote:
I think the AMD/ATI project is looking towards mobile, right?


Definately. I think that in this area the merger will help AMD, more than even the desktop segment. Centrino has been a powerful brand for Intel, With the ATI merger, AMD might now be able to compete.
!