But! If you are using the SLI option on the MLB linked above, the PCI slot would probably be inaccessible, and (while admitting that I have NOT looked at all) I do not know if they make PCI-X FW cards...
I don't mean to torture you (much) with these 'predictions', but it can all only point to pro OpenGL cards coming to the PowerMac really soon now!
Now Both the Intel Dual Xeon motherboards, and AMD 64bit Dual Opteron motherboard configuration have an SLI option available. Via the nForce4 AMD now has a Dual Opteron Motherboard with an SLI graphics option available, and according to nVidia "Support for Future Dual-Core Opteron CPUs "
"NVIDIA is working closely with AMD to validate the upcoming dual-core versions of the AMD Opteron CPU."
SLI is a dumb technology.
It's like strapping two engines into a Ferrari.
It only improves performance when applications are fill-rate bound. Because two cards sharing the display buffer can fill faster than one.
However most professional graphics applications are *not* fill rate bound. Although videogames often are.
More pro applications preview in wireframe - for this very reason and are typically transform-bound or cpu-bound.
For an application to become fill-rate bound (that means limited in performance ONLY by the rate the CPU can generate pixels) it means that a lot of optimization has to be done at the processing and transformation stages.
We have already established that this optimization is not happening. Which is a problem - but its not a problem that SLI will solve.
I think the G5 case is plenty big enough. a slight redesign the internals of it, and the MB is probably what they would need to do, but the MB would be new for PCI-E any way.
Quote:
Originally posted by MacRonin
Beating this dead horse...
The board also only has DDR1 slots, not DDR2...
;^p
That's just one board used as an example, but it can still allow for 16GB of DDR which would suffice.
It only improves performance when applications are fill-rate bound. Because two cards sharing the display buffer can fill faster than one.
However most professional graphics applications are *not* fill rate bound. Although videogames often are.
More pro applications preview in wireframe - for this very reason and are typically transform-bound or cpu-bound.
For an application to become fill-rate bound (that means limited in performance ONLY by the rate the CPU can generate pixels) it means that a lot of optimization has to be done at the processing and transformation stages.
We have already established that this optimization is not happening. Which is a problem - but its not a problem that SLI will solve.
Carni
Before we go, and decide that Graphics cards are not necessary anymore for use in highend 3D applications. Would you care to explain to us all why there are Quadro FX Pro 3D cards, and why they out perform standard graphics cards for use with 3D applications? And save the conspiracy theory BS. There is no need to get into that nonsense again.
I like this line the best.
Quote:
It's like strapping two engines into a Ferrari.
It almost makes sense, but It's not a Ferrari. It's a Scarab (boat). Dual i/b - o/b motors. That's more like it. But it's better in a computer. It would never help in a car. You'd never go back to single if you had it in a boat. And if you had ever used it in a computer. You may feel the same way.
Actually, I saw a Honda CRX that Motor Trend (or was it Car & Driver...?!?) put a second stock engine/transaxle in (driving the rear wheels) that helped the performance quite a bit...
As for SLI in an Apple rig, couldn't come soon enough...!
Before we go, and decide that Graphics cards are not necessary anymore for use in highend 3D applications. Would you care to explain to us all why there are Quadro FX Pro 3D cards, and why they out perform standard graphics cards for use with 3D applications? And save the conspiracy theory BS. There is no need to get into that nonsense again.
There you go - all face jumpy again.
Do you know anything about real time 3d?
I will explain
Applications are limited by one of three performance bottlenecks.
CPU,
Transform
or Fill rate
Applications are rarely fill rate bound. Programmers have to do a lot of optimization before they hit this wall. Typically fill-bound apps will have to shove a lot of filled polygons with fancy shaders attached. IE Games.
Pro apps tend not to be fill-rate bound. Easy to demonstrate. Does MAYA run faster on an SLI? Quadro boards you will notice have ultra fast wireframe support for precisely this reason.
The only non-gaming app I think SLI would help is flight simulation which I am sure is why NVidia are supporting SLI on Quadro.
Modern SLI is a technology to primarily improve fill rate. It won't improve things on the Mac.
Looking at the recently published performance stats for Mac versus PC 3d - There is a long road of optimization to go before SLI will make a smidgen of difference.
Like I said, there is a performance problem on the Mac - but SLI won't help.
Who cares about SLI, there isn't enough games on the Mac platform to warrant the cost of buying two (DirectX optimized) graphic cards. What I would like to see are professional graphic cards like a Wildcat, Quadro or FireGL with optimized OSX drivers.
Who cares about SLI, there isn't enough games on the Mac platform to warrant the cost of buying two (DirectX optimized) graphic cards. What I would like to see are professional graphic cards like a Wildcat, Quadro or FireGL with optimized OSX drivers.
SLI is not for gaming performance. Look at benchmarks, not what was speculated. SLI is for 3D apps, nothing else.
The current implementation of SLI splits the existing 16 lanes between two cards... so you're not gaining any bandwidth.
From an architectural standpoint, this has nothing to offer performance wise. It perhaps presents an interesting upgrade path, but that is all. The same performance is possible from simply strapping the same hardware on a single board. This would cut engineering and produciton costs significantly as compared to working with SLI systems.
In general, you get the best price/performance ratio when driving a single display with a single display board.
PCI-E has much to offer users.
SLI has little to offer unless you're into meaningless stat-mongering with other gamers.
Which benchmarks? The Doom3 ones or the Halo ones?
Anything.
dfiler, yes it seems as though SLI is in its infancy at the moment, but it does seem to have a lot of potential. However with the advent of cell processor technology SLI could possibly become redundant sooner than we think.
SLI has been around for many years. Although, ironically the acronymn now means something different.
For decades now, systems designers have experimented with using multiple cards to drive a single display. It's never turned out to be economical.
While it's possible that the price/performance tradeoff may change, there's yet to be a case made as to why this time is different. It has always been cheapest to drive a display from a single card rather than multiple cards. This trend can be observed throughout the industry. Fewer boards and chips equates to better price/performance.
The idea of strapping two graphic cards into a PC to improve graphics performance seems an obvious way of improving graphics throughput.
But the difficulty is sharing the load between the two cards. Old Voodoo style SLI did this by rendering the same scene on both cards.
Both cards had to do identical transform and lighting calculations (no saving) Both cards have to host a duplicate texture cache (no saving) - and only when it came to rasterizing (filling ) the screen was there a benefit to be had.
One card rendered the odd scanlines the other took the even. This is how it made things go faster. In fill-bound applications this saving would be significant. Although rarely would it be double.
As soon as the fill performance improves, it not long befire your application is soon going to be hogtied by vertex performance.
New style SLI is supposed to do some sharing of geometry processing between the two cards - although I am a bit skeptical about that.
There are applications where SLI *can* clearly make a benefit. Most notably games where there are high numbers of polys, a high level of overdraw - and expensive shaders.
But any application which needs the following will benefit:
1) Large frame buffers- HD and above - where the number of pixels to fill is significant.
2) Expensive pixel shaders. If a pixel shader is very expensive then any application can become fill-rate bound. SLI will help here too. The use of Normal map shaders in Doom 3 is a good example of "expensive".
Typically, pro 3D apps use wireframe previews or simple shaded previews. And although the performance of some 3D apps is not great on the Mac - it isn't from lack of fill rate.
3DS Max and Softimage now allow you to use pixel shaders in their 3D preview. So SLI might help here. However neither of these applications to my knowledge runs on the Mac.
SLI in a Mac , not soon people! Were still bitching about Nvidia 5200 ultras being Apples pride and choice. Apple needs to give us some professional cards before it starts giving us toys to play with.
SLI in a Mac , not soon people! Were still bitching about Nvidia 5200 ultras being Apples pride and choice. Apple needs to give us some professional cards before it starts giving us toys to play with.
I agree. I also think that as they update the graphic specs on the high end PowerMacs Apple needs to do the same thing with the middle and lower end models as well. I mean, look at the iMac. It has a 1.8gig G5,up to 2 gigs of memory, and it maxes out at 250GB of HD storage but it is crippled by its low end graphics chipset. If I'm going to spend over $2000 on a Mac I want it to have a high end graphics chipset like a 256MB 5900 Ultra. And dont even say Apple cant do it,I've seen them in laptops and if they can put them in laptops they can put them in the iMac,its just that Apple chooses not to do it.
Comments
Originally posted by MacRonin
But! If you are using the SLI option on the MLB linked above, the PCI slot would probably be inaccessible, and (while admitting that I have NOT looked at all) I do not know if they make PCI-X FW cards...
I don't mean to torture you (much) with these 'predictions', but it can all only point to pro OpenGL cards coming to the PowerMac really soon now!
Cheers!
It has 1- PCI, 3 PCI-X, and 2 PCI-E I think.
Originally posted by onlooker
It has 1- PCI, 3 PCI-X, and 2 PCI-E I think.
From left to right:
PCI-X 133MHz
PCI-X 100 MHz
PCI-X 100MHz
PCI-e
PCI
PCI-e
So the PCI slot is between the PCI-e slots, which probably kills it's usage if you are running dual GPUs in SLI...
I would figure that one of the PCI-X 100MHz slots might be crowded out also...
Same image shows only the lone FW400 port...
The board also only has DDR1 slots, not DDR2...
;^p
Originally posted by BenRoethig
Apple would need a redesigned case for SLI.
Good gives them the chance to add another HD bay while they are at it.
Originally posted by onlooker
Now Both the Intel Dual Xeon motherboards, and AMD 64bit Dual Opteron motherboard configuration have an SLI option available. Via the nForce4 AMD now has a Dual Opteron Motherboard with an SLI graphics option available, and according to nVidia "Support for Future Dual-Core Opteron CPUs "
"NVIDIA is working closely with AMD to validate the upcoming dual-core versions of the AMD Opteron CPU."
SLI is a dumb technology.
It's like strapping two engines into a Ferrari.
It only improves performance when applications are fill-rate bound. Because two cards sharing the display buffer can fill faster than one.
However most professional graphics applications are *not* fill rate bound. Although videogames often are.
More pro applications preview in wireframe - for this very reason and are typically transform-bound or cpu-bound.
For an application to become fill-rate bound (that means limited in performance ONLY by the rate the CPU can generate pixels) it means that a lot of optimization has to be done at the processing and transformation stages.
We have already established that this optimization is not happening. Which is a problem - but its not a problem that SLI will solve.
Carni
Originally posted by BenRoethig
Apple would need a redesigned case for SLI.
I think the G5 case is plenty big enough. a slight redesign the internals of it, and the MB is probably what they would need to do, but the MB would be new for PCI-E any way.
Originally posted by MacRonin
Beating this dead horse...
The board also only has DDR1 slots, not DDR2...
;^p
That's just one board used as an example, but it can still allow for 16GB of DDR which would suffice.
Originally posted by Carniphage
SLI is a dumb technology.
It's like strapping two engines into a Ferrari.
It only improves performance when applications are fill-rate bound. Because two cards sharing the display buffer can fill faster than one.
However most professional graphics applications are *not* fill rate bound. Although videogames often are.
More pro applications preview in wireframe - for this very reason and are typically transform-bound or cpu-bound.
For an application to become fill-rate bound (that means limited in performance ONLY by the rate the CPU can generate pixels) it means that a lot of optimization has to be done at the processing and transformation stages.
We have already established that this optimization is not happening. Which is a problem - but its not a problem that SLI will solve.
Carni
Before we go, and decide that Graphics cards are not necessary anymore for use in highend 3D applications. Would you care to explain to us all why there are Quadro FX Pro 3D cards, and why they out perform standard graphics cards for use with 3D applications? And save the conspiracy theory BS. There is no need to get into that nonsense again.
I like this line the best.
It's like strapping two engines into a Ferrari.
It almost makes sense, but It's not a Ferrari. It's a Scarab (boat). Dual i/b - o/b motors. That's more like it. But it's better in a computer. It would never help in a car. You'd never go back to single if you had it in a boat. And if you had ever used it in a computer. You may feel the same way.
Originally posted by onlooker
It would never help in a car.
Actually, I saw a Honda CRX that Motor Trend (or was it Car & Driver...?!?) put a second stock engine/transaxle in (driving the rear wheels) that helped the performance quite a bit...
As for SLI in an Apple rig, couldn't come soon enough...!
Originally posted by onlooker
Before we go, and decide that Graphics cards are not necessary anymore for use in highend 3D applications. Would you care to explain to us all why there are Quadro FX Pro 3D cards, and why they out perform standard graphics cards for use with 3D applications? And save the conspiracy theory BS. There is no need to get into that nonsense again.
There you go - all face jumpy again.
Do you know anything about real time 3d?
I will explain
Applications are limited by one of three performance bottlenecks.
CPU,
Transform
or Fill rate
Applications are rarely fill rate bound. Programmers have to do a lot of optimization before they hit this wall. Typically fill-bound apps will have to shove a lot of filled polygons with fancy shaders attached. IE Games.
Pro apps tend not to be fill-rate bound. Easy to demonstrate. Does MAYA run faster on an SLI? Quadro boards you will notice have ultra fast wireframe support for precisely this reason.
The only non-gaming app I think SLI would help is flight simulation which I am sure is why NVidia are supporting SLI on Quadro.
Modern SLI is a technology to primarily improve fill rate. It won't improve things on the Mac.
Looking at the recently published performance stats for Mac versus PC 3d - There is a long road of optimization to go before SLI will make a smidgen of difference.
Like I said, there is a performance problem on the Mac - but SLI won't help.
Carni
Originally posted by Relic
Who cares about SLI, there isn't enough games on the Mac platform to warrant the cost of buying two (DirectX optimized) graphic cards. What I would like to see are professional graphic cards like a Wildcat, Quadro or FireGL with optimized OSX drivers.
SLI is not for gaming performance. Look at benchmarks, not what was speculated. SLI is for 3D apps, nothing else.
Originally posted by mattyj
SLI is not for gaming performance. Look at benchmarks, not what was speculated. SLI is for 3D apps, nothing else.
Which benchmarks? The Doom3 ones or the Halo ones?
From an architectural standpoint, this has nothing to offer performance wise. It perhaps presents an interesting upgrade path, but that is all. The same performance is possible from simply strapping the same hardware on a single board. This would cut engineering and produciton costs significantly as compared to working with SLI systems.
In general, you get the best price/performance ratio when driving a single display with a single display board.
PCI-E has much to offer users.
SLI has little to offer unless you're into meaningless stat-mongering with other gamers.
Originally posted by Carniphage
Which benchmarks? The Doom3 ones or the Halo ones?
Anything.
dfiler, yes it seems as though SLI is in its infancy at the moment, but it does seem to have a lot of potential. However with the advent of cell processor technology SLI could possibly become redundant sooner than we think.
It'll be interesting to see how this turns out.
For decades now, systems designers have experimented with using multiple cards to drive a single display. It's never turned out to be economical.
While it's possible that the price/performance tradeoff may change, there's yet to be a case made as to why this time is different. It has always been cheapest to drive a display from a single card rather than multiple cards. This trend can be observed throughout the industry. Fewer boards and chips equates to better price/performance.
But the difficulty is sharing the load between the two cards. Old Voodoo style SLI did this by rendering the same scene on both cards.
Both cards had to do identical transform and lighting calculations (no saving) Both cards have to host a duplicate texture cache (no saving) - and only when it came to rasterizing (filling ) the screen was there a benefit to be had.
One card rendered the odd scanlines the other took the even. This is how it made things go faster. In fill-bound applications this saving would be significant. Although rarely would it be double.
As soon as the fill performance improves, it not long befire your application is soon going to be hogtied by vertex performance.
New style SLI is supposed to do some sharing of geometry processing between the two cards - although I am a bit skeptical about that.
There are applications where SLI *can* clearly make a benefit. Most notably games where there are high numbers of polys, a high level of overdraw - and expensive shaders.
But any application which needs the following will benefit:
1) Large frame buffers- HD and above - where the number of pixels to fill is significant.
2) Expensive pixel shaders. If a pixel shader is very expensive then any application can become fill-rate bound. SLI will help here too. The use of Normal map shaders in Doom 3 is a good example of "expensive".
Typically, pro 3D apps use wireframe previews or simple shaded previews. And although the performance of some 3D apps is not great on the Mac - it isn't from lack of fill rate.
3DS Max and Softimage now allow you to use pixel shaders in their 3D preview. So SLI might help here. However neither of these applications to my knowledge runs on the Mac.
Carni
SLI in a Mac , not soon people! Were still bitching about Nvidia 5200 ultras being Apples pride and choice. Apple needs to give us some professional cards before it starts giving us toys to play with.
I agree. I also think that as they update the graphic specs on the high end PowerMacs Apple needs to do the same thing with the middle and lower end models as well. I mean, look at the iMac. It has a 1.8gig G5,up to 2 gigs of memory, and it maxes out at 250GB of HD storage but it is crippled by its low end graphics chipset. If I'm going to spend over $2000 on a Mac I want it to have a high end graphics chipset like a 256MB 5900 Ultra. And dont even say Apple cant do it,I've seen them in laptops and if they can put them in laptops they can put them in the iMac,its just that Apple chooses not to do it.