theBishopp
14-03-08, 11:41
A thread from another forum, which greatly interested me, and I wondered what you lot thought... A few of the replies have been included too... The title is the title from the other forum too...
Ati/AMD are soon going to release 2 high end GFX cards. Not only that but AMD now have a genuine rocking chipset. This is going to hit Nvidia hard. Then this larabee is coming from Intel. AMD and Intel are going for the GPCPU. Intel and now AMD have competent die shrinking technologies. Nvidia hasn't got a clue how to make a CPU. I think nvidia could well begin to falter in just 12months time.
On the plus side the 9800s have been benchmarked and are not a total disappointment as was feared.
Nvidia will try and say that you can only get high end GPU performance with an add in card. Though how that will work in a longer period of time will be interesting as the market for cards will only stay in high end. Mid range may well be covered by GPCPUs.
There have been rumours of Nvidia either buying AMD (not likely) or entering into the CPU market themselves (they have a low powered phone processor recently launched, so maybe they are thinking about that).
IMO nVidia would be mad to try to break into CPUs in any major way. The investment requirements are huge, credit is very expensive now, and there's not a lot of profit margin in CPUs presently. They wouldn't be allowed to buy AMD anyway, due to anti-monopoly laws.
I'd be surprised if CPGPUs (or whatever they're called) will alter the discrete graphics card market significantly - I suspect they'll be similar in effect to on-board GPUs that currently live on motherboards. People generally will get tired of falling behind with graphics performance long before their CPU has reached the end of its life, with obvious consequences. In the mobile sector, CPGPUs may extend battery life, as CPU power saving technologies are automatically extended to GPUs on the same die. Motherboards will be simplified, or have room for more functions (including extra serial bandwidth for discrete graphics cards).
Not necessarily earth-shattering for nVidia, though. AMD/Ati are just returning to the market position for graphics cards that they held pre-takeover. They probably expected a temporary dip in design output and competitiveness.
The only way that I could see CPGPUs becoming a real threat to discrete video cards would be if a similar approach was taken to the early 386s where a separate maths co-processor was optionally dropped into another socket next to the main CPU.
You could then have super-fast 2D environment acceleration for the desktop with modest 3D acceleration from the main GPCPU with super-accelerated 3D tasks taken on by the optional GPU co-processor with its own heatsink and fan. This could then be replaced for little more than the cost of a GPU chip at will by the end-user.
Not just the 386, but also the 8088, 8086, 286, and 486 SX. The maths co-pro only became a fixed feature across an entire x86 range with the advent of the Pentium.
I can't imagine any better way for the current GPU manufacturers to maximise their potential horsepower and also their profits whilst simultaneously cutting out the middleman. Bear in mind the relatively cheap offers we've seen from both camps on the "X2" variants of their graphics cards which seems to infer that the cost of GPU manufacture and supply only constitutes a small fraction of the total cost of a graphics card.
If Ati and nVidia were given the ability to directly interface their GPUs with the CPU rather than crossing the PCI-E bus to get the two talking to each other then we'd all stand to win. We could kiss goodbye to XFX, Sapphire, etc. RAMDACs and the rest of the supporting hardware found on current graphics cards are nothing special and could easily and cheaply be integrated onto any motherboard eliminating the potential bottleneck of crossing over the current PCI-E bus architecture with a little clever design work due to the GPU being tied directly to the CPU.
GPU RAM could be socketed on the motherboard in exactly the same way that CPU RAM is currently socketed in order to give the end-user the choice of the exact speed and type of graphics RAM to use with their personal choice of GPU. It would open up an entire new world of possibilities whilst leaving the GPU manufacturers open to develop as they choose within much less limited constraints than exist within the current format.
Maybe nVidia know something that Ati doesn't because they've been steaming ahead bearing scant regard for the limitations of PCI-E while Ati have been making very conservative use of it. Although the approach that Ati have taken is arguably superior within the limitations of a PCI-E bus, nVidia seem to have been almost ignoring these limitations. Is this in preparation for something to come that's just over the horizon?
I'm totally convinced that this is the way graphics hardware is headed long-term. If an approach along these lines was to be taken then you could quite reasonably expect to see enormous gains in graphics processing performance within the next couple of generations.
Have you guys any thoughts on this?
Ati/AMD are soon going to release 2 high end GFX cards. Not only that but AMD now have a genuine rocking chipset. This is going to hit Nvidia hard. Then this larabee is coming from Intel. AMD and Intel are going for the GPCPU. Intel and now AMD have competent die shrinking technologies. Nvidia hasn't got a clue how to make a CPU. I think nvidia could well begin to falter in just 12months time.
On the plus side the 9800s have been benchmarked and are not a total disappointment as was feared.
Nvidia will try and say that you can only get high end GPU performance with an add in card. Though how that will work in a longer period of time will be interesting as the market for cards will only stay in high end. Mid range may well be covered by GPCPUs.
There have been rumours of Nvidia either buying AMD (not likely) or entering into the CPU market themselves (they have a low powered phone processor recently launched, so maybe they are thinking about that).
IMO nVidia would be mad to try to break into CPUs in any major way. The investment requirements are huge, credit is very expensive now, and there's not a lot of profit margin in CPUs presently. They wouldn't be allowed to buy AMD anyway, due to anti-monopoly laws.
I'd be surprised if CPGPUs (or whatever they're called) will alter the discrete graphics card market significantly - I suspect they'll be similar in effect to on-board GPUs that currently live on motherboards. People generally will get tired of falling behind with graphics performance long before their CPU has reached the end of its life, with obvious consequences. In the mobile sector, CPGPUs may extend battery life, as CPU power saving technologies are automatically extended to GPUs on the same die. Motherboards will be simplified, or have room for more functions (including extra serial bandwidth for discrete graphics cards).
Not necessarily earth-shattering for nVidia, though. AMD/Ati are just returning to the market position for graphics cards that they held pre-takeover. They probably expected a temporary dip in design output and competitiveness.
The only way that I could see CPGPUs becoming a real threat to discrete video cards would be if a similar approach was taken to the early 386s where a separate maths co-processor was optionally dropped into another socket next to the main CPU.
You could then have super-fast 2D environment acceleration for the desktop with modest 3D acceleration from the main GPCPU with super-accelerated 3D tasks taken on by the optional GPU co-processor with its own heatsink and fan. This could then be replaced for little more than the cost of a GPU chip at will by the end-user.
Not just the 386, but also the 8088, 8086, 286, and 486 SX. The maths co-pro only became a fixed feature across an entire x86 range with the advent of the Pentium.
I can't imagine any better way for the current GPU manufacturers to maximise their potential horsepower and also their profits whilst simultaneously cutting out the middleman. Bear in mind the relatively cheap offers we've seen from both camps on the "X2" variants of their graphics cards which seems to infer that the cost of GPU manufacture and supply only constitutes a small fraction of the total cost of a graphics card.
If Ati and nVidia were given the ability to directly interface their GPUs with the CPU rather than crossing the PCI-E bus to get the two talking to each other then we'd all stand to win. We could kiss goodbye to XFX, Sapphire, etc. RAMDACs and the rest of the supporting hardware found on current graphics cards are nothing special and could easily and cheaply be integrated onto any motherboard eliminating the potential bottleneck of crossing over the current PCI-E bus architecture with a little clever design work due to the GPU being tied directly to the CPU.
GPU RAM could be socketed on the motherboard in exactly the same way that CPU RAM is currently socketed in order to give the end-user the choice of the exact speed and type of graphics RAM to use with their personal choice of GPU. It would open up an entire new world of possibilities whilst leaving the GPU manufacturers open to develop as they choose within much less limited constraints than exist within the current format.
Maybe nVidia know something that Ati doesn't because they've been steaming ahead bearing scant regard for the limitations of PCI-E while Ati have been making very conservative use of it. Although the approach that Ati have taken is arguably superior within the limitations of a PCI-E bus, nVidia seem to have been almost ignoring these limitations. Is this in preparation for something to come that's just over the horizon?
I'm totally convinced that this is the way graphics hardware is headed long-term. If an approach along these lines was to be taken then you could quite reasonably expect to see enormous gains in graphics processing performance within the next couple of generations.
Have you guys any thoughts on this?