idea about gfx i would like some feedback

Gurahl

n00b
Joined
Feb 9, 2008
Messages
7
**would seriously like you input good bad question anything**

a few years ago i was curious why graphics cards were use and toss kind of parts it seems to cost ineffective to me. why not just have a modular system where the parts of the graphics system are more like that of the computer because quite simply thier pretty much the same thing in a simple look at them. you have you gpu like a cpu ram pretty universal in the long run and in put and output connections to and from the motherbaord for data and output to the moniter.

here is my thought

use a datacard to connect to the video bus with a fibre channel or equivelant data tranfer to a seperate case card simply so it can be changed out as the motherbaords get upgraded,

as i see it now the actual megahertz of the graphics card are not really progressing very fast so we grab a socket thats moving pretty fast and might have some longevity for this scheme use it to mount a gpu heck we might as well throw an extra or three on the board as well

we use daughterboard riser cards to house ram modules this way as the tech gets going we atleast have an option of changing out the card to upgrade the type of ram we use in our graphics system

basically what im saying here is why not make it modular this way we can get some longevity and out of our base graphics sure we might need to upgrade here and there but it gives us alot more options oh you want a dual gpu 2 gig graphics system fine no problem you want a single gpu 4 gig graphics system also fine we can build it to what we need rather then have to take whats out there and hope its what we need ive heard people complain about not having enough frames for years but that all they do oh well this system build is better then this one if ya want true upgradeability we have to get away from out of the box graphics solutions and start a truly upgradeable solution that can work for everyone graphics cards have had bios's for years so its not like they cant be upgraded its just not practical for the end user under the current market plan with a modular system we could upgrade our graphics for years till the technology comes out that no longer compatible modular would let people upgrade at thier own pace as well making it feasible for people to but parts as they need them as fast as they need them 512 gig of mem here gpu there chipset here new data card there initial cost for r&d might be a bit high but in the long run graphics companies would make money because the people turned off by the high cost of say card x wouldnt be that bad because they could essentially make it themselves in bits and peices

yes my grammar is poor and my typing sucks but its not a bad idea it worth looking into
 
an example of what i was thinking of rather then just having to go sink another x amount of money on buying 9800gx2's or whatever correct me if i have the jist of it wrong but the 9800gx2 is pretty much an 8800 chipshrunk and multiplied by 2 with some better ram chips why not just go and grap another gpu the better ram and slap it in a case as it comes out and upgrading the bios on the board probably saing time and money for both us and the developers of the hardware
 
It already is modular enough for the current pace of development.
You dont need to replace the whole innards of your PC to get better graphics and you dont need to replace 5 things on your graphics card to upgrade it.
1 simple change of the card itself is all that is needed and is really all that is wanted with the fast pace of graphics card development.

Many more things are required to change/improve throughout the evolution of a graphics processor, here are some:
1) Power delivery / connections to the card
2) Power conversion/stability to the GPU/Ram
3) Bus speeds between the components (requires new higher speed backing PCB for the GPU/Ram to sit on)
4) More advanced DACs and output circuitry to give higher resolution/refresh rates/picture stability/new digital interfaces
5) Bus connections to the PC, we are now on a 2 way V1.1 PCI-E x16 for a single card with version 2.0 emerging.
6) expansion in bus bandwidth is often delivered by having more parallel connections (ie 128bit vs 256bit memory), this requires a redesign of the backing PCB to handle more data lines and may need an increase in the number of PCB layers.
7) Driver development would be a lot tougher as there will be many more variations possible. Making the driver modular too will slow the graphics system further AND make it more complex..
8) there is much more...

For each newer generation, there are so many changes to be made that making the graphics card itself modular is impractical due to the complexity and stability of having so many extra higher speed connections that are not hard soldered. Making these buses into connectors not only makes them less stable by not having a direct connection but the line traces will be longer and therefore harder to get stable at higher speed.
Direct shorter connections rule.

It will ADD complexity by making the card more modular as there are too many changes still occurring.
ie to change just your graphics processor, you would have to change many components, not just one and any one of these will have its own issues related to physical problems through to even more compatibility problems!
Thats saying nothing of getting the driver(s) to function correctly.

Once we have enough power to manage practically everything we want to throw at the graphics system, I'm sure a more modular approach can work as the pace of development will slow.
Not yet though and probably for the next 10 to 20 years, theres a long way to go :)

There is a lot more to add but I had a good night last night and it took me a while to pull this from the ether :)
 
It already is modular enough for the current pace of development.
You dont need to replace the whole innards of your PC to get better graphics and you dont need to replace 5 things on your graphics card to upgrade it.
1 simple change of the card itself is all that is needed and is really all that is wanted with the fast pace of graphics card development.

Many more things are required to change/improve throughout the evolution of a graphics processor, here are some:
1) Power delivery / connections to the card
would say a cube case size case with a independant powersupply work for this need ?

2) Power conversion/stability to the GPU/Ram
3) Bus speeds between the components (requires new higher speed backing PCB for the GPU/Ram to sit on) how many pcbs are there currently being used on the graphics cards ?
4) More advanced DACs and output circuitry to give higher resolution/refresh rates/picture stability/new digital interfaces
we could use some kind of output card as well for this simliar to a tv tuner type device allowing for the port to be changed
5) Bus connections to the PC, we are now on a 2 way V1.1 PCI-E x16 for a single card with version 2.0 emerging.

thats why i meantioned fibre channel coonection to the modular system so that the card adapter could be changed out as the interface ot upgraded
6) expansion in bus bandwidth is often delivered by having more parallel connections (ie 128bit vs 256bit memory), this requires a redesign of the backing PCB to handle more data lines and may need an increase in the number of PCB layers.

could the pcb layers be tooled into the daughter board configuration i mentioned if we used a different type of connection
7) Driver development would be a lot tougher as there will be many more variations possible. Making the driver modular too will slow the graphics system further AND make it more complex..

as i understand it now the drivers are primarily written for the gpu involved rather then the amount of ram on the card im prolly wrong so dont shoot me im here for ideas
8) there is much more...

For each newer generation, there are so many changes to be made that making the graphics card itself modular is impractical due to the complexity and stability of having so many extra higher speed connections that are not hard soldered. Making these buses into connectors not only makes them less stable by not having a direct connection but the line traces will be longer and therefore harder to get stable at higher speed.
Direct shorter connections rule.

It will ADD complexity by making the card more modular as there are too many changes still occurring.
ie to change just your graphics processor, you would have to change many components, not just one and any one of these will have its own issues related to physical problems through to even more compatibility problems!
Thats saying nothing of getting the driver(s) to function correctly.

Once we have enough power to manage practically everything we want to throw at the graphics system, I'm sure a more modular approach can work as the pace of development will slow.
Not yet though and probably for the next 10 to 20 years, theres a long way to go :)

There is a lot more to add but I had a good night last night and it took me a while to pull this from the ether :)

i never said it was going to be easy but im a big fan of lateral thinking
 
I just want to add to the thorough post Nenu made.

You can't use modular memory because high-end graphics cards inherently use bleeding-edge memory tech. The buses are custom-made for the speeds and memory chips they're designed to interface with; you really can't build-in an upgrade path without making the bus spec faster (and thus upping the cost).

If you want to know what happens when you design a memory-type agnostic bus, look no further than Intel's FB-DIMM. This technology adds an interface between the DRAM and the memory controller, allowing the controller to support any type of memory, and also allowing the controller to support more devices on the bus. The downside: poor performance, and lots of heat generated by the buffers. In fact, the cost, poor performance and high heat are the reason Intel is removing FB-DIMM support from their low-end Xeon workstation chipset this year.

The reasons we can have interchangeable standards for main computer memory (but not graphics memory) is because these buses are typically much lower-speed than the ones you see on graphics cards, which allows you to populate the bus with multiple modules without causing instability.

I can remember several topics on this stretching back to the 90s, especially when embedded DRAM was ready to take the graphics world by storm (like that actually happened). eDRAM would make this easy, because you could use the AGP design with a fast on-chip frame buffer, and texturing from main memory without much performance hit. When you wanted a frame buffer/graphics speed upgrade, you could just pop-out the chip and replace it with a new one; when you wanted better texturing performance, you would get more (or faster) main memory.

Unfortunately, large amounts of fast external memory made AGP texturing redundant, and eDRAM never surfaced in the PC graphics world. However, you can see this architecture at-work on the Xbox 360, where they use the 12MB eDRAM for frame buffer and anti-aliasing, and use the external shared ram for triangle data and texturing. Maybe we will eventually see eDRAM on the desktop, but it's not today.
 
i never said it was going to be easy but im a big fan of lateral thinking

Cool, theres many of us who do too.
Its not that this hasnt been thought of, its just impractical to implement.
 
ok i see what your talking about have yall heard about the phase change ram corsair is working by chance?

http://en.wikipedia.org/wiki/Phase-change_memory

give ya lil bit of a precurser into the stuff
ibm is currently researching it use in something similar ro a ramsan harddrive but instead since the current has to alternate attenuation to frequency it would save in it current state at power loss failure
corsair and samsung were noted on the stock report for one of the largest manufacturers of the material as having placed large orders as well as three of the largest military contractors in the world also placed orders
that was early last year
 
The reasons we can have interchangeable standards for main computer memory (but not graphics memory) is because these buses are typically much lower-speed than the ones you see on graphics cards, which allows you to populate the bus with multiple modules without causing instability.

Unfortunately, large amounts of fast external memory made AGP texturing redundant, and eDRAM never surfaced in the PC graphics world. However, you can see this architecture at-work on the Xbox 360, where they use the 12MB eDRAM for frame buffer and anti-aliasing, and use the external shared ram for triangle data and texturing. Maybe we will eventually see eDRAM on the desktop, but it's not today.[/QUOTE]

ok if im reading this correctly even a modular on the daughterboard memory controller wouldn't work correctly never said i was a guru of hardware so can break down the speed difference your talking about specifically is the number of bandwidth lanes too and from the gpu and then from the gpu to the adapters


whats the order of execution for the gfx card specifically im guessing its something like fetch the program data to ram with an overflow buffer and then that info is processed by the gpu and out to the display adapter

again dont shoot me im trying to wrap my head around it

i know its not the same but i was told that i wouldnt be able to build a freestanding dedicated lapping machine at home and i did this is way more complex but if i can understand why its not possible either ill drop it or ill keep going till it can be
 
ok i see what your talking about have yall heard about the phase change ram corsair is working by chance?

Yes, its interesting ideas but not commercial... yet!
Wikipedia has a fair write up on the history
http://en.wikipedia.org/wiki/Phase-change_memory
It certainly will give a performance leap if they can release it soon.

As changes in the memory state are delivered by a temperature difference rather than electronically, temperature isolation needs to be very good to prevent anomolous changes.
The memory will need to be impervious to radiation too.
Especially if it is to be used in a lot of commercial environments as the energy from stray radiation can cause heating and a change of memory state.
This kind of memory currently needs a constant supply of energy to keep its state so I dont see it as the holy grail... yet.
 
Yes, its interesting ideas but not commercial... yet!
Wikipedia has a fair write up on the history
http://en.wikipedia.org/wiki/Phase-change_memory
It certainly will give a performance leap if they can release it soon.

As changes in the memory state are delivered by a temperature difference rather than electronically, temperature isolation needs to be very good to prevent anomolous changes.
The memory will need to be impervious to radiation too.
Especially if it is to be used in a lot of commercial environments as the energy from stray radiation can cause heating and a change of memory state.
This kind of memory currently needs a constant supply of energy to keep its state so I dont see it as the holy grail... yet.

a lil bit better info its based on amplitude of the voltage not heat though
http://domino.research.ibm.com/comm/pr.nsf/pages/news.20061211_phasechange.html?Open&printable

and corsair published an article last year in pcgamer near the back about ramm dimms using this type of ship in thier new ram lines sometimes this year

intel report from this month on the stuff
http://www.trustedreviews.com/cpu-memory/news/2008/02/07/Intel-Sampling-Phase-Change-Memory/p1
 
This kind of memory currently needs a constant supply of energy to keep its state so I dont see it as the holy grail... yet.

No, it doesn't need a constant supply of voltage to hold the state, it is nonvolatile. The only major problem is it is susceptible to temperature changes.

This means that there is a line between performance and reliability that manufacturers will have to be careful about. There are ways to reduce the activation temperature of the phase change, but if you do this, you make the device more susceptible to external temperature changes.

I've been pretty disappointed with the whole NV memory front because there's not a single technology without drawbacks.

Flash: requires high voltages to trap charge inside the cell (write), and makes things worse because if you want to write one cell, you have to erase an entire block. Less important: the charge bleeds away over time, so Flash has a finite short life of 10 years or less, unless you rewrite the data.

MRAM and FeRAM: can't get the densities up because the MEMS constructs don't shrink well. Also requires some exotic materials that are very rare. But if you don't need density and have the cash, very nice.

PRAM: temperature makes this hard to implement. We can make this work on rewritable media, with DVD-RWs withstanding large temperature ranges. But the write speed of a DVD-RW is slower than one bit per 100ns, which is really not capable of competing with Flash, let alone DRAM. If you want faster speeds, you run into that annoying temperature wall I mentioned above.
 
Yeah I didnt explain that too well.
The memory isnt truly non volatile yet.

As the ram is still susceptible to localised temperatures, they will need to use power to maintain the temps/data while the ram is operational until the problem is solved.
This will also mean more heat differences in the system that can further affect it.
They can otherwise suffer a shorter life of the data in memory if it fits the application but that does limit its uses.
A quick refresh of the data state to keep the data intact works fine but it isnt truly non volatile.
The amount of power necessary to keep the memory state intact while operational looks to be acceptable now due to lower power developments with the write process in the last year.
I imagine this will limit the environments the memory can be used in though unless extra care is taken to reduce the temp the memory sits in when running and when not in use.

When the memory is off, it is non volatile as long as it isnt exposed to high temps for extended periods.
As I dont know the temps used to change the phase of the material, I cant dig much deeper :)
 
well from the data ive been readin looks like its 150 and 400 milliamperes which raises the temperature to 185 degrees c but thats on a 15nm insulated crystalline wafer so im not sure what that would translate to when its layered into a pcb
and 175 degrees c and below is 0 data bit or reset
 
Back
Top