GTX480 -> 2-Way SLI Scaling -> Almost 100%!!!!!

[MagickMan]

And this is an "enthusiast" site so the numbers are bound to be skewed in your favor but the numbers are still bad for your argument

No doubt. I skews the numbers by a factor of 2-3 at least. I'd venture to guess that less than 10% of PC gamers have ever had SLI/CF.

 

[Martimus]

I am having flashbacks to posts by Rollo for some reason.

 

[Daedalus685]

http://www.hardware.info/nl-NL/articles/amdnampoZGCa/Clash_of_the_Titans_3way_SLI_GTX_480_test/13

For what it is worth they use a different sample and directly compare the scaling (it is from the Netherlands though... but just as dated as the guru 3d one you link).

2:1 scaling for gtx480 is jst shy of 80% overall, the 5870 is hovering at about 70%.

There is no doubt that the SLI scales AMAZING.. but you have to give the sales BS a rest. We are all enthusiasts and yet we still care about things like value, efficiency, and so on...

Oh, and for the record.. the 5870 trifire scales about 152% compared to 128% for the 3way SLI... Care to add how "Enthusiasts who really count would only ever buy 3, and the 5870 scales MUCH better in triple set ups" to the thread? or how quadfire shows very little diminishing returns from crossfire. Granted that it is not as high up front but ATI has done an exceptional job in reducing the penalty for more than 2 (which they need to perfect in order to really drive a small die strategy home).

/sigh to the idea that an enthusiast only counts with 5 GPUs, 9 CPU sockets and no less than 256GB of ram.... I suppose some would do that... mind you some just as enthusiastic will spend just as much cash on a rig 1/100 the speed to ensure it is dead silent and consumes as little power as possible.

 

[Skurge]

http://www.hardware.info/nl-NL/articles/amdnampoZGCa/Clash_of_the_Titans_3way_SLI_GTX_480_test/13

For what it is worth they use a different sample and directly compare the scaling (it is from the Netherlands though... but just as dated as the guru 3d one you link).

2:1 scaling for gtx480 is jst shy of 80% overall, the 5870 is hovering at about 70%.

There is no doubt that the SLI scales AMAZING.. but you have to give the sales BS a rest. We are all enthusiasts and yet we still care about things like value, efficiency, and so on...

Oh, and for the record.. the 5870 trifire scales about 152% compared to 128% for the 3way SLI... Care to add how "Enthusiasts who really count would only ever buy 3, and the 5870 scales MUCH better in triple set ups" to the thread? or how quadfire shows very little diminishing returns from crossfire. Granted that it is not as high up front but ATI has done an exceptional job in reducing the penalty for more than 2 (which they need to perfect in order to really drive a small die strategy home).

/sigh to the idea that an enthusiast only counts with 5 GPUs, 9 CPU sockets and no less than 256GB of ram.... I suppose some would do that... mind you some just as enthusiastic will spend just as much cash on a rig 1/100 the speed to ensure it is dead silent and consumes as little power as possible.

Am I reading that correctly? 2 to 4 GPUs give 65% additional performance? Thats pretty good actually.

 

[tincart]

/sigh to the idea that an enthusiast only counts with 5 GPUs, 9 CPU sockets and no less than 256GB of ram.... I suppose some would do that... mind you some just as enthusiastic will spend just as much cash on a rig 1/100 the speed to ensure it is dead silent and consumes as little power as possible.

Agreed. I have always seen a computer hardware enthusiast as someone who has a specific attitude towards technology - not someone who owns the expensive product of the month. I could buy two 5870's or I could stick with what I have and tweak to squeeze out extra performance (and "performance" can be FPS, low temperatures, low noise, whatever threshold you set for your hardware to perform at). Either approach is valid for "enthusiasts."

 

[thilanliyan]

**

You still haven't given any facts to back up your claim of people with 5870s and 480s have more than one. Is that coming any time soon? I'm asking honestly as I didn't think that was the case, but I could be wrong.

 

[SlowSpyder]

OP, as has been asked numerous times, please provide proof that most users who buy a 5870 or GTX480 level card intend to use at least two of them. I simply do not believe that. While I'm sure 5870 and GTX480 users are more likely to use multiple cards than say Radeon 5400 users or 9500GT users, I'd be very suprised if a majority of 5870 or GTX480 users were using two or more cards. What you are saying clearly goes against what most of us here think, so if you can't provide some sort of proof of this than I don't believe it.

I have a 5870 and would not consider a second unless I go some sort of crazy good deal on it, and even then I may pass it by as my single 5870 is pretty capable at 1920x1200. My computer room/office is the sercond floor of our home, it only has 15 amp service which is an issue with the wife at her computer, the TV on and a space heater in winter or window AC in summer going. I don't know that adding another 180 watts of power consumption would be a good idea until I get a dedicated outlett for the space heater/AC

 

[tweakboy]

That is insane. Still however I rather go with a dual GPU single PCB , nVidia of course, ATI drivers are never gonna mature.. gl

 

[lavaheadache]

scaling does look great indeed.

 

[luv2increase]

http://www.hardware.info/nl-NL/articles/amdnampoZGCa/Clash_of_the_Titans_3way_SLI_GTX_480_test/13

For what it is worth they use a different sample and directly compare the scaling (it is from the Netherlands though... but just as dated as the guru 3d one you link).

2:1 scaling for gtx480 is jst shy of 80% overall, the 5870 is hovering at about 70%.

There is no doubt that the SLI scales AMAZING.. but you have to give the sales BS a rest. We are all enthusiasts and yet we still care about things like value, efficiency, and so on...

Oh, and for the record.. the 5870 trifire scales about 152% compared to 128% for the 3way SLI... Care to add how "Enthusiasts who really count would only ever buy 3, and the 5870 scales MUCH better in triple set ups" to the thread? or how quadfire shows very little diminishing returns from crossfire. Granted that it is not as high up front but ATI has done an exceptional job in reducing the penalty for more than 2 (which they need to perfect in order to really drive a small die strategy home).

/sigh to the idea that an enthusiast only counts with 5 GPUs, 9 CPU sockets and no less than 256GB of ram.... I suppose some would do that... mind you some just as enthusiastic will spend just as much cash on a rig 1/100 the speed to ensure it is dead silent and consumes as little power as possible.

I appreciate the post, but both the ATI config and Nvidia config were bottlenecked by their test CPU in that review. The GPUs simply weren't given the breathing room they needed to run at their fullest potential.

In the Guru3D review, it is no doubt that the 2-way and 3-way tests were somewhat bottlenecked as well, just not to the extent as they were in the review you posted.

@thilan
You point to where I said said people with 480s and 5870s will have more than one. You can't do it. I simply stated "MANY". You are about the fourth person who took what I said completely out of context.

For the record, I never stated that those without the cream of the crop GPUs were "not" enthusiasts. I never said that. Seriously, I think people forgot how to read.

It must have something to do with humanity losing their communication skills due to texting. Although, this would mainly affect the younger generation.

 

[Outrage]

So unless you have a i7 @ more then 3.4Ghz you shouldent by a nvidia card, you never know... you might end up with 3 gpu's. Afterall, many enthusiasts who buy 2 cards will also end up getting a third card.

 

[Daedalus685]

I appreciate the post, but both the ATI config and Nvidia config were bottlenecked by their test CPU in that review. The GPUs simply weren't given the breathing room they needed to run at their fullest potential.

In the Guru3D review, it is no doubt that the 2-way and 3-way tests were somewhat bottlenecked as well, just not to the extent as they were in the review you posted.
.

Please, PLEASE tell me what would have been a better choice in the review than the i7 980x at 3.47ghz......

You have to be kidding... or you didn't actually look at the configuration they used... Because if that is bottlenecked too much to make 2 3 or 4 way crossfire not worth it then every computer but for a multiple socket server would be in the same boat...

 

[VirtualLarry]

Many AMD boards now support SLI..

WHAT??? Name one.

 

[luv2increase]

So unless you have a i7 @ more then 3.4Ghz you shouldent by a nvidia card, you never know... you might end up with 3 gpu's. Afterall, many enthusiasts who buy 2 cards will also end up getting a third card.

This is a very basic thing we are talking about here. It is called a GPU bottleneck. Now, we are living in an age with multi-threaded gaming so multi-threaded CPUs are very important for that as well, and you gaming performance will go up from an OC'ed multi-threaded CPU as well since many games "today" are multi-threaded unlike how they were just 3 short years ago.

If you are going to have 2 x 5870s or 2 x 480s, you should give them at least 4ghz on an i7 CPU to let them work to their fullest capabilities. To go any less would be not getting your money's worth.

These are the most powerful GPUs in the world. They need powerful CPUs to coincide with them.

 

[Daedalus685]

Am I reading that correctly? 2 to 4 GPUs give 65% additional performance? Thats pretty good actually.

Yes, you are reading it correctly. ATI currently has exceptional scaling for more than 2 cards, their 2 cards is not bad either but handily beaten by SLI.

 

[Daedalus685]

This is a very basic thing we are talking about here. It is called a GPU bottleneck. Now, we are living in an age with multi-threaded gaming so multi-threaded CPUs are very important for that as well, and you gaming performance will go up from an OC'ed multi-threaded CPU as well since many games "today" are multi-threaded unlike how they were just 3 short years ago.

If you are going to have 2 x 5870s or 2 x 480s, you should give them at least 4ghz on an i7 CPU to let them work to their fullest capabilities. To go any less would be not getting your money's worth.

These are the most powerful GPUs in the world. They need powerful CPUs to coincide with them.

Well lucky for you, the review I linked does jsut that.

If one actually goes to the next page in the review they bring the i7 980x (the most powerful CPU in the world) up to 4.4ghz and see how much the triple SLI scores increase.

The displayed GeForce GTX 480 3-way SLI scores will make many people raise the question whether the performance still be higher if the processor is overclocked. That we examined: the Gulf Town arts might be a 33% higher clock frequency display - 4.4 GHz to be exact - and we ran four of the benchmarks again. Only in 3DMark Vantage is still a more-or-more significant increase in performance. In both game benchmarks, the difference in the score less than two percent. We may therefore well conclude that such heavy graphics benchmarks, the cards are clearly the bottleneck and not the processor.

*Translated through Google Chrome

I don't think you understand just how difficult it is to bring the CPU to the bottleneck at 30' resolutions on a 6 core i7... Bottlenecks are fun little things and are not linear (they are absolutely not basic as you say), but more often than not in any game at the highest settings all the video cards you can stuff into the computer won't pull the limiting factor away from the GPU. The idea that you need a 4ghz i7 6 core for SLI or crossfire is garbage.

 

[luv2increase]

Well lucky for you, the review I linked does jsut that.

If one actually goes to the next page in the review they bring the i7 980x (the most powerful CPU in the world) up to 4.4ghz and see how much the triple SLI scores increase.

The problem with that is it is a "known" problem right now that the "drivers" are causing bad performance in 3-way and 4-way SLI configs with the 480. Until the driver situation is worked out, the review you posted is:

1. Irrelevant in the 2-way CF/SLI tests due to the bottleneck
2. Irrelevant in the 3-way SLI tests due to drivers


I thought everyone knew Nvidia was having problems with 3-way SLI and the 480? I guess I thought wrong

 

[Daedalus685]

The problem with that is it is a "known" problem right now that the "drivers" are causing bad performance in 3-way and 4-way SLI configs with the 480. Until the driver situation is worked out, the review you posted is:

1. Irrelevant in the 2-way CF/SLI tests due to the bottleneck
2. Irrelevant in the 3-way SLI tests due to drivers


I thought everyone knew Nvidia was having problems with 3-way SLI and the 480? I guess I thought wrong

You have to be kidding me....

You came on here spouting how people who are enthusiasts buy 2 or 3 or 4 cards..... I posted a review looking at the scaling of this (not even as dated as the review in your OP) and you blame drivers... Well sir, if the drivers are bad then don't mention 3 or 4 way SLI right now....

The review you linked blames the CPU on the poor scaling in triple SLI, not the drivers.... You wanted to compare scaling numbers... Well triple crossfire looks to scale better, so does quad... You can blame drivers and the CPU all day but one does not draw an arbitrary line in the sand... If comparing 1 on 1 was not valid then one cannot complain when 3 on 3 is compared because of "the CPU and drivers"

Please tell me how a CPU that is above and beyond FASTER than the one in the GURU article is bottlenecked for 2 way SLI/CF... It is also far faster than the CPU in your rig.. I guess you are bottlenecked from SLI so there is no validity in your comments on it eh?

 

[NoQuarter]

This is a very basic thing we are talking about here. It is called a GPU bottleneck. Now, we are living in an age with multi-threaded gaming so multi-threaded CPUs are very important for that as well, and you gaming performance will go up from an OC'ed multi-threaded CPU as well since many games "today" are multi-threaded unlike how they were just 3 short years ago.

If you are going to have 2 x 5870s or 2 x 480s, you should give them at least 4ghz on an i7 CPU to let them work to their fullest capabilities. To go any less would be not getting your money's worth.

These are the most powerful GPUs in the world. They need powerful CPUs to coincide with them.

Those aren't CPU bottlenecks. As Daedalus685 pointed out the 4.4Ghz OC'd results are negligibly different. The way to tell if the CPU is bottlenecking the system is to look at the benches from 1 resolution lower. In this review the benches at 1920x1200 are much higher than the benches at 2560x1600, meaning the CPU isn't bottlenecking the system because it's capable of pushing many more frames per second at 1920x1200 than the GPUs can handle at 2560x1600.

edit: wait, what? The results are irrelevant because the drivers don't work better than they do? How do you get the card to run faster with nonexistant faster drivers? Both reviews used 197.41 btw.

 

[luv2increase]

You have to be kidding me....

You came on here spouting how people who are enthusiasts buy 2 or 3 or 4 cards..... I posted a review looking at the scaling of this (not even as dated as the review in your OP) and you blame drivers... Well sir, if the drivers are bad then don't mention 3 or 4 way SLI right now....

Hey genius, I never did mention 3 or 4-way SLI right now. Notice the title of the thread???????

You have failed reading comprehension. Also, it is UNDISPUTED that Nvidia is having problems with 3-way and 4-way scaling right now. It is the truth.

You have to realize that I look at everything from an "unbiased" standpoint. I could care less for either company. It is just computer hardware for crying out loud.

@NoQuarter
I took it as the 980x was only tested on the 3-way 480 setup and "not" the 2-way 480. It doesn't matter for the 3-way setup because at this time, the scaling is messed up anyway and no amount of CPU horsepower can change that. It is in the drivers. Until they fix the scaling, that linked review is "pointless".

 

[Daedalus685]

Hey genius, I never did mention 3 or 4-way SLI right now. Notice the title of the thread???????

You have failed reading comprehension. Also, it is UNDISPUTED that Nvidia is having problems with 3-way and 4-way scaling right now. It is the truth.

You have to realize that I look at everything from an "unbiased" standpoint. I could care less for either company. It is just computer hardware for crying out loud.

@NoQuarter
I took it as the 980x was only tested on the 3-way 480 setup and "not" the 2-way 480. It doesn't matter for the 3-way setup because at this time, the scaling is messed up anyway and no amount of CPU horsepower can change that. It is in the drivers. Until they fix the scaling, that linked review is "pointless".

"Many people seem to "want" to focus on pitting a single 5870 up against a single 480. But, why would we do this when more likely than not, those who buy the 480 and 5870 are probably going to opt for 2 of them and possibly even 3 or 4!"

You don't want to compare 1:1, yet 3:3 is not valid....

Thanks for the compliment as well .

As insane as it might be I'll even give you the benefit that you are trying to be unbiased but are just ignorant.

So, lets go through this shall we:

- The situation here is NOT bottlenecked because a MASSIVE over clock (again I point out it was on the most powerful CPU in the world) provided almost no increase. It is VERY difficult to bottleneck something at these resolutions with a CPU. The relationship is not linear but generally speaking if we get x FPS at a lower resolution the CPU is capable of doing that (a bit less) at higher resolutions before it would be the bottleneck. This is absolutely not the case here, high res crushes the FPS in half in many cases. The GPU (even triple crossfire) is the bottleneck.

- The drivers may very well be at fault... but if we are going to discount 3 way for driver issues we may as well go straight back to 1:1 as it is just as likely that driver fixes will improve CF scaling as well... There are two things we can compare.. what we have now and nothing... If we are not allowed to compare what we have now for whatever reason then we are back to nothing.



The review used the 980x for everything.. They used to use a 965 but replaced it for the 980x for this review. It is all written in the configuration page. Despite all of this, the review is no more pointless than the one in the OP... If this one fails based on some merit then all others do as well. I'd like to see a source on it being a driver issue that is known as well. As it stands it simply does not compete, which doesn't matter as the 3 (or 2) GPU configuration systems are not exactly selling like hot cakes...

 

[luv2increase]

"Many people seem to "want" to focus on pitting a single 5870 up against a single 480. But, why would we do this when more likely than not, those who buy the 480 and 5870 are probably going to opt for 2 of them and possibly even 3 or 4!"

You don't want to compare 1:1, yet 3:3 is not valid....

Thanks for the compliment as well .

As insane as it might be I'll even give you the benefit that you are trying to be unbiased but are just ignorant.

So, lets go through this shall we:

- The situation here is NOT bottlenecked because a MASSIVE over clock (again I point out it was on the most powerful CPU in the world) provided almost no increase. It is VERY difficult to bottleneck something at these resolutions with a CPU. The relationship is not linear but generally speaking if we get x FPS at a lower resolution the CPU is capable of doing that (a bit less) at higher resolutions before it would be the bottleneck. This is absolutely not the case here, high res crushes the FPS in half in many cases. The GPU (even triple crossfire) is the bottleneck.

- The drivers may very well be at fault... but if we are going to discount 3 way for driver issues we may as well go straight back to 1:1 as it is just as likely that driver fixes will improve CF scaling as well... There are two things we can compare.. what we have now and nothing... If we are not allowed to compare what we have now for whatever reason then we are back to nothing.



The review used the 980x for everything.. They used to use a 965 but replaced it for the 980x for this review. It is all written in the configuration page. Despite all of this, the review is no more pointless than the one in the OP... If this one fails based on some merit then all others do as well.

You don't have a good understand about how GPU drivers work.

When I had my tri-Fire 5870 rig, it wasn't until the 9.12 Hotfix drivers that my 3-way scaling actually became VERY GOOD! However, the 1-way and 2-way WERE GOOD before the 9.12 Hotfix. When the 3-way scaling was horrible, were we supposed to take the 3-way results at that time of truly indicative of how 3 x 5870s performed in tri-Fire? Absolutely not. The drivers were well optimized for single and 2-way CF but nothing more.

Just because drivers do well in 2-way configs does not automatically mean that the 3-way and 4-way are going to be optimized as well. That isn't how it works Daniel Son

 

[Daedalus685]

You don't have a good understand about how GPU drivers work.

When I had my tri-Fire 5870 rig, it wasn't until the 9.12 Hotfix drivers that my 3-way scaling actually became VERY GOOD! However, the 1-way and 2-way WERE GOOD before the 9.12 Hotfix. When the 3-way scaling was horrible, were we supposed to take the 3-way results at that time of truly indicative of how 3 x 5870s performed in tri-Fire? Absolutely not. The drivers were well optimized for single and 2-way CF but nothing more.

Just because drivers do well in 2-way configs does not automatically mean that the 3-way and 4-way are going to be optimized as well. That isn't how it works Daniel Son

Good guess on my name... lol

But I know how drivers work... The point is we either compare them now or we don't. We can never use magic to determine whether 3 way scaling will get better or not. We can hope, sure, but we only have what we have. And yes, the old trifire results WERE indicative of the performance of triple crossfire. We can all hope it was a bug (mind you I never saw a review until this year) but we can only compare what we have at any point in time. At the time it might have been terrible. Now it doesn't seem to be as bad, and triple SLI is. t would be a HUGE gamble to get 3 480s and hope for the same performance increase, just as it would have been with 5870s in November.

At what point is the poor scaling from limiting factors and not drivers?

Can we compare 3 way SLI when it reaches 50% scaling? I could just as easily claim that until ATI drivers are better (i.e. 80% scaling) it is pointless to compare crossfire to SLI.

Of course those that do well in 2-way may not do well in 3-way.... That is not the point. If we are going to go high end until we get to a fault in a configuration (such as going from single to SLI/crossfire) then we can bloody well go one level higher to find a fault in the other team.

Despite all of this.. hardly anyone will buy 2 or 3 of these things. 1:1 is by FAR the most useful metric for folks.

 

[NoQuarter]

You don't have a good understand about how GPU drivers work.

When I had my tri-Fire 5870 rig, it wasn't until the 9.12 Hotfix drivers that my 3-way scaling actually became VERY GOOD! However, the 1-way and 2-way WERE GOOD before the 9.12 Hotfix. When the 3-way scaling was horrible, were we supposed to take the 3-way results at that time of truly indicative of how 3 x 5870s performed in tri-Fire? Absolutely not. The drivers were well optimized for single and 2-way CF but nothing more.

Just because drivers do well in 2-way configs does not automatically mean that the 3-way and 4-way are going to be optimized as well. That isn't how it works Daniel Son

Yes, you are suppose to take the 3-way results as indicative of how 3 x 5870 performed in tri-Fire because that is how they performed. You can caveat it with prospective improvements in drivers, and fully expect to see performance gains in later drivers, but to someone who just bought or is thinking of buying 3 x 5870's that it what they are going to see when they hook it up! That is an influence on the buyer's decision and very relevant.

 

[Skurge]

Hey genius, I never did mention 3 or 4-way SLI right now. Notice the title of the thread???????

You have failed reading comprehension. Also, it is UNDISPUTED that Nvidia is having problems with 3-way and 4-way scaling right now. It is the truth.

You have to realize that I look at everything from an "unbiased" standpoint. I could care less for either company. It is just computer hardware for crying out loud.

@NoQuarter
I took it as the 980x was only tested on the 3-way 480 setup and "not" the 2-way 480. It doesn't matter for the 3-way setup because at this time, the scaling is messed up anyway and no amount of CPU horsepower can change that. It is in the drivers. Until they fix the scaling, that linked review is "pointless".

Wait, the review is pointless cause nV drivers suck with 3xSLI? ok Gotcha

That means your review is pointless too, cause AMD's drivers suck with 2xCF.

That also means until every review shows 100% scaling with both AMD and nV setups.

Seriously, you can't first blame the CPU then when that is disproved you go on to blame drivers and say a review is pointless when it doesn't show a certain setup performing in a certain way.