Quad SLI - Any benchmarks?

[S_D]

Don't worry too much about it being CPU limited guys - Just imagine running it on a Dell 30" panel at 2560x1600 on full detail on everything and FSAA and AF and that'll tax the gfx cards some more.

Also, I read a story somewhere this week that we were to expect dual cpu socket SLI and crossfire boards in the next 6 - 9 months. So you can have a dual CPU (each dual core) quad SLI system that won't be CPU limited anymore

Simon

 

[chaparral]

Don't worry too much about it being CPU limited guys - Just imagine running it on a Dell 30" panel at 2560x1600 on full detail on everything and FSAA and AF and that'll tax the gfx cards some more.

Also, I read a story somewhere this week that we were to expect dual cpu socket SLI and crossfire boards in the next 6 - 9 months. So you can have a dual CPU (each dual core) quad SLI system that won't be CPU limited anymore

Simon

And how much will that be... ..£5000 by any chance..plus say another £1500 for dell 30" monitor,keyboard,mouse etc

 

[Tommy B]

WTF is the point in DUAL, or QUAD.

Why not have just ONE AWESOME GPU or CPU. It's pathetic tbh.

 

[S_D]

WTF is the point in DUAL, or QUAD.

Why not have just ONE AWESOME GPU or CPU. It's pathetic tbh.

Because it wouldn't be worth it for a manufacturer to develop one ultra ultra high end part that'd sell in such tiny numbers. Much cheaper to take an existing part and team it up with another for relatively low development costs and an appealing easy upgrade path for consumers.

Simon

 

[kdd]

Hey nobody puts a gun to your head and forces you to buy this stuff. It's offered and you have the choice. Quite a lot of choice at the moment.

 

[YooEntSinMeROYT]

WTF is the point in DUAL, or QUAD.

Why not have just ONE AWESOME GPU or CPU. It's pathetic tbh.

The point is that the CPU manufacturers are about reaching the frequency limits on a single processor die. Due to such extreme miniturisation of electronic gates (we're talking atomic level here) phenomena are arising such as electromigration & subthreshold leakage. Electromigration can cause the eventual loss of one or more transistor connections, which obviously lead to errors in calculations & an unstable CPU. Subthreshold leakage comes into play when scaling down the size of transistors (smaller die size). As when scaling down the size of transistors the supply voltage has to be scaled down also. As the supply voltage is reduced, the threshold voltage has to be reduced in the same proportion. As threshold voltages are reduced, subthreshold leakage rises exponentially. Until it gets to a point where there is more power leakage than power consumption.

So how do you get around this?

You can't up the frequency of the one CPU core any further to gain extra performance. As you are about at limit of the current architecture.

So you run more than one CPU core in unison & hey presto......you gain extra performance & you don't have to spend extra R&D in trying to find the next holy grail in architecture.

Which in the end equates to being able to purchase processors at a reasonable price (as they don't have to recoup their R&D costs).

It made interesting reading researching this. Hope some of you find it useful.
Make sure you read the Wikipedia links as I may not have reached the most accurate conclusion above, but it is my best understanding of it

 

[chaparral]

Hey nobody puts a gun to your head and forces you to buy this stuff. It's offered and you have the choice. Quite a lot of choice at the moment.

Thats not the problem...

The problem is went the lastest and greatest games get release to only work on this type of hardware at playable FPS with high detail levels...
Games like fear, COD2 , Oblivion etc,,,are already very close to needing an SLI/crossfire setup..How long will it be before games are going to need quad & PhysX Accelerator card setups.

 

[nbkpsycho]

Waste of money tbh

 

[turbotoaster]

The point is that the CPU manufacturers are about reaching the frequency limits on a single processor die. Due to such extreme miniturisation of electronic gates (we're talking atomic level here) phenomena are arising such as electromigration & subthreshold leakage. Electromigration can cause the eventual loss of one or more transistor connections, which obviously lead to errors in calculations & an unstable CPU. Subthreshold leakage comes into play when scaling down the size of transistors (smaller die size). As when scaling down the size of transistors the supply voltage has to be scaled down also. As the supply voltage is reduced, the threshold voltage has to be reduced in the same proportion. As threshold voltages are reduced, subthreshold leakage rises exponentially. Until it gets to a point where there is more power leakage than power consumption.

So how do you get around this?

You can't up the frequency of the one CPU core any further to gain extra performance. As you are about at limit of the current architecture.

So you run more than one CPU core in unison & hey presto......you gain extra performance & you don't have to spend extra R&D in trying to find the next holy grail in architecture.

Which in the end equates to being able to purchase processors at a reasonable price (as they don't have to recoup their R&D costs).

It made interesting reading researching this. Hope some of you find it useful.
Make sure you read the Wikipedia links as I may not have reached the most accurate conclusion above, but it is my best understanding of it

everybody, listen to this man, very good point you are making there

 

[Tommy B]

The point is that the CPU manufacturers are about reaching the frequency limits on a single processor die. Due to such extreme miniturisation of electronic gates (we're talking atomic level here) phenomena are arising such as electromigration & subthreshold leakage. Electromigration can cause the eventual loss of one or more transistor connections, which obviously lead to errors in calculations & an unstable CPU. Subthreshold leakage comes into play when scaling down the size of transistors (smaller die size). As when scaling down the size of transistors the supply voltage has to be scaled down also. As the supply voltage is reduced, the threshold voltage has to be reduced in the same proportion. As threshold voltages are reduced, subthreshold leakage rises exponentially. Until it gets to a point where there is more power leakage than power consumption.

So how do you get around this?

You can't up the frequency of the one CPU core any further to gain extra performance. As you are about at limit of the current architecture.

So you run more than one CPU core in unison & hey presto......you gain extra performance & you don't have to spend extra R&D in trying to find the next holy grail in architecture.

Which in the end equates to being able to purchase processors at a reasonable price (as they don't have to recoup their R&D costs).

It made interesting reading researching this. Hope some of you find it useful.
Make sure you read the Wikipedia links as I may not have reached the most accurate conclusion above, but it is my best understanding of it

Yeah, so why don't they just do what AMD have done; Dual core. The "Dual" bit but just in one unit. Why don't Nvidia do the "dual" bit in one card too?

 

[james.miller]

for a start gpu's are massively more complex than cpu's. the 7800gtx has 302 million transistors. a single core opteron weighs in at around 100m.

cpu's were being pushed further and further with regards to speed - the fx57 2.8ghz for example. You can do that when you have a small amount of transistors for a core, but there does come a point were it get increasingly difficult to extract the performance when your ramping speeds up and the process size down.

for cpu's, the easiest thing to do was just tack another core on and you get your performance. Gpu's run slowly enough that they dont encounter the same EM problems that cpu's do and besides that, i dont think a dual core 7800gts with over 600 million transistors is neither workable or profitable right now.