SLI/Crossfire performance

[0xygenthief]

I just got through reading a review on HardOCP regarding SLI/Crossfire performance in regards to PCIe bandwidth. The tests ran Nvidia and ATI cards against a slew of benchmarks at varying resolutions. All of the results showed that a PCIe x8 lane performs the same as a x16 lane under the new PCIe 2.0 standard.

http://www.hardocp.com/article/2010/08/16/sli_cfx_pcie_bandwidth_perf_x16x16_vs_x16x8

I found this very hard to believe. So I did my own research on the net and found a few other reviews. None of which shared these findings. Then I discovered why...

In the HardOCP review the test system was OC'd to a meager 3.6 Ghz. While other reviews on the same subject were at the 4.0 or higher mark. Essentially the CPU speed was the bottleneck in the tests. It is likely that the bandwidth used never approached the x8 lane's limit. Obviously, the person writing the review did not take this into consideration when performing his tests.

I just wanted to share this with the group. I am a big HardOCP fan and typically have no issues with the reviews done there. But I simply have no idea why this review ever made it on the web. Don't drink the foolaid on this one.

In short, if you don't plan on Overclocking your system to above 4 Ghz then you shouldn't invest in multiple graphics cards. It's the equivalent of putting twin turbos on a bicycle.

EDIT: Due to ND40oz's astounding ability to derail a thread and completely ignore the intent of this thread's original post I hereby ask that anyone else reading this thread IGNORE the summarization. It does not do a good job of summarizing the intent of the OP which, in turn, confuses many folks who have no idea how to read a post in its entirety.

P.S. Do not worry about all the sand in ND40oz's mangina. One of the grains of sand is actually his penis and as a result he gets confused and frustrated. He is very sad that all his boyfriends giggle at the little pecker. Please be kind and not point and laugh, it will give him a complex and undoubtedly cause him to flame other threads like this one...

[Baleful]

In short, if you don't plan on Overclocking your system to above 4 Ghz then you shouldn't invest in multiple graphics cards. It's the equivalent of putting twin turbos on a bicycle.

This has got to be one of the dumbest, most fucking uneducated statements I've ever read in my life.

Multi GPU scaling is affected by more than just CPU speed. The review done at [H] is fine, throwing an extra 400Mhz on the processor to bump it up to 4.0 would only bump up the frames MAYBE 1-2% IF THAT.

Multi GPU's DON'T NEED a massive OC. Sure they could benefit from it, but saying that running multi GPU's on anything less than a 4.0 OC isn't worth it, is completely fucking retarded :bird:

[ND40oz]

Yeah, OP, you need to go back and re-read the article, you're clearly missing something. It wasn't testing single vs multicard setups, it was testing multicard setups with different PCIe 2.0 bandwidths.

[Baleful]

Yeah, OP, you need to go back and re-read the article, you're clearly missing something. It wasn't testing single vs multicard setups, it was testing multicard setups with different PCIe 2.0 bandwidths.

I think he got that, but he's saying that if [H] had that 920 OCed to 4.0+ then the x16 tests would have been leaps and bounds better than the x8 tests.

Still, completely uneducated post :bird:

[KaptCrunch]

Have you wanted to space your SLI or CFX video cards farther apart on your motherboard to allow for better airflow? Do you have a motherboard that will not support a x16/x16 PCIe configuration? We put x16/x16 and x16/x8 PCIe SLI and CFX configurations head to head and show you what sacrifice there is to be made, All of our real world gaming tests on the following pages are performed apples-to-apples at the exact same in-game settings for the specific resolution. We are comparing x16/x16 versus x16/x8 on each video card combination in three games, Aliens vs. Predator, Bad Company 2 and Metro 2033. The more we learn about PCIe Gen. 2.0 the more we have to love about it. While PCIe Gen. 3.0 should be upon us in the coming year, we have found that we have a lot of usable PCIe 2.0 bandwidth now, at least as far as we are concerned with multi-GPU video card configurations.


You can find slight framerate differences between x16/x16 and x16/x8 SLI and CrossFireX video card configurations if you go looking for them while pushing games to super high end resolutions and quality settings, but nothing that negatively impacts a real world gaming experience. If you are wanting to use SLI or CrossFireX on that motherboard that does not support x16/x16 PCIe lanes, have no fear. If you do have a x16/x16 motherboard and want to space those hot video cards out a bit to let those breath a bit easier, well have at it. (Next week, x16/x16 vs. x8/x8 in real world gaming.)
.

what going on is the pci lanes is lagging compared to multi gpu's iThinks it will show up on thier next test

[LowBrowser]

I think what you should have found from the article is that the PCIe 2.0 bandwidth does not seem to be limiting two cards when they both run as 8x, when the CPU is clocked at 3.6GHz. At 4GHz, the bandwidth of an 8x channel does seem to begin to limit the system, and at that stage, a 16x channel (logical not physical) seems to mediate this.

The major issue is that a long time ago, a logical 8x PCIe 1.0 channel did seem to limit frame rates when compared to a 16x channel, but this is not as much of a problem with PCIe 2.0, which has double the bandwidth of the original specification.

Well that's what I got from the article (which admittedly I read a long time ago).
Edit:The article I read was from HardOCP, but with older cards. Basically the same thing applies though.

[0xygenthief]

This has got to be one of the dumbest, most fucking uneducated statements I've ever read in my life.

Multi GPU scaling is affected by more than just CPU speed. The review done at [H] is fine, throwing an extra 400Mhz on the processor to bump it up to 4.0 would only bump up the frames MAYBE 1-2% IF THAT.

Multi GPU's DON'T NEED a massive OC. Sure they could benefit from it, but saying that running multi GPU's on anything less than a 4.0 OC isn't worth it, is completely fucking retarded :bird:

Ok, so let me get this straight, Baleful. You ADVOCATE the use of intel celeron cpu's clocked at 1.8 ghz or lower to push SLI/Crossfire? Oh yeah! And I'm fucking retarded by your standards? Eat shit. :bird:

The point of my entire post was not to say that you need a heavy OC to push SLI/Crossfire at all. The point was to call out the fact that without an effective OC to push the GPUs that the bandwidth of the x8 pcie lane would never be utilized. How the fuck can you perform any such comparison without determining when the limit will be reached? I'd insert another (possibly more effective) metaphor to explain but I am sure you would simply focus in on the metaphor rather than the real issue.

In any case, maybe my example wasn't a good one, for that I am really fucking sorry. However, I say who fucking cares, you read the rest of my post and instead of weighing in on the real issue you nit-picked on my metaphor. Idiot... :htfu:

[KaptCrunch]

head room on the buss lol x8 + x8 = x16 but if using x16 + x16 = x32 thats the issue

you cant get both cards to run @ x16 not till pci 3.0 comes

[0xygenthief]

I think what you should have found from the article is that the PCIe 2.0 bandwidth does not seem to be limiting two cards when they both run as 8x, when the CPU is clocked at 3.6GHz. At 4GHz, the bandwidth of an 8x channel does seem to begin to limit the system, and at that stage, a 16x channel (logical not physical) seems to mediate this.

My point exactly LowBrowser. It's not that the article was wrong, in fact it was a decent article. It just didn't include this one important bit of information. Essentially, any noob that has no fucking clue would interpret the article to say there is absolutely zero difference. Yet those same noobs would sit and whine when their Evga x8/x8/x8 motherboard doesn't get high scores similar to other x16/x16/x16 boards during benchmarks.

Thanks for reading my entire post and not bitching about my metaphor. I promise I won't call you retarded for agreeing with me. ;)

[Baleful]

You fucking dumb bitch, when the fuck did I say anything about a god damned Intel Celeron? I didn't you fucking hick.


From your original post, you are comparing the CPU freq. that [H] used (3.6) to some other testing you've found on the net that are using a higher CPU freq. (4.0+). Your LAST statement is "In short, if you don't plan on Overclocking your system to above 4 Ghz then you shouldn't invest in multiple graphics cards." - This is completely fucking WRONG. I'm not going by [H]'s testing, though I don't see anything wrong with it, I'm going by my OWN experience from testing hardware. In multi GPU solutions where you experience a x16/x8 PCIE 2.0, having a high OC isn't going to make much of a fucking difference. PERIOD.

Yeah sure, if you're using a god damned Intel Celeron in a multi GPU solution, you're just fucking yourself. But if you aren't a complete fucking tool and have ATLEAST a quadcore clocked at 3.2, then moving up to 4.0 isn't going to push that x8 lane much further.

Now, why don't you move the fuck along and actually speak of some shit you have ACTUAL experience with, not just some bullshit you're reading off the net.

[0xygenthief]

You fucking dumb bitch, when the fuck did I say anything about a god damned Intel Celeron? I didn't you fucking hick.


From your original post, you are comparing the CPU freq. that [H] used (3.6) to some other testing you've found on the net that are using a higher CPU freq. (4.0+). Your LAST statement is "In short, if you don't plan on Overclocking your system to above 4 Ghz then you shouldn't invest in multiple graphics cards." - This is completely fucking WRONG. I'm not going by [H]'s testing, though I don't see anything wrong with it, I'm going by my OWN experience from testing hardware. In multi GPU solutions where you experience a x16/x8 PCIE 2.0, having a high OC isn't going to make much of a fucking difference. PERIOD.

Yeah sure, if you're using a god damned Intel Celeron in a multi GPU solution, you're just fucking yourself. But if you aren't a complete fucking tool and have ATLEAST a quadcore clocked at 3.2, then moving up to 4.0 isn't going to push that x8 lane much further.

Now, why don't you move the fuck along and actually speak of some shit you have ACTUAL experience with, not just some bullshit you're reading off the net.

I am really happy to see you didn't appreciate my celeron statement Baleful. It had just as much to do with my original post as your responses to my origional post. How does it feel bitch? :bird:

Now that I am done pushing your buttons how about you answer one question straight out?

Do you or don't you agree that a higher OC @ 4Ghz or higher would likely show a point at which an x8 pcie lane would begin to bottleneck the system and affect the benchmark results?

Agree and I will forgive all statements suggesting I have no idea what I am talking about. Disagree and you can feel free to hop back onto the short bus. Choice is yours dildo baggins...

[ND40oz]

I am really happy to see you didn't appreciate my celeron statement Baleful. It had just as much to do with my original post as your responses to my origional post. How does it feel bitch? :bird:

Now that I am done pushing your buttons how about you answer one question straight out?

Do you or don't you agree that a higher OC @ 4Ghz or higher would likely show a point at which an x8 pcie lane would begin to bottleneck the system and affect the benchmark results?

Agree and I will forgive all statements suggesting I have no idea what I am talking about. Disagree and you can feel free to hop back onto the short bus. Choice is yours dildo baggins...

Dude you didn't say any of that in your first post, instead you said:

In short, if you don't plan on Overclocking your system to above 4 Ghz then you shouldn't invest in multiple graphics cards. It's the equivalent of putting twin turbos on a bicycle.

Now who's on the short bus?

[0xygenthief]

Dude you didn't say any of that in your first post, instead you said:



Now who's on the short bus?

From my original post:

"In the HardOCP review the test system was OC'd to a meager 3.6 Ghz. While other reviews on the same subject were at the 4.0 or higher mark. Essentially the CPU speed was the bottleneck in the tests. It is likely that the bandwidth used never approached the x8 lane's limit. Obviously, the person writing the review did not take this into consideration when performing his tests."

So to answer your question, Baleful is driving the bus and you are in the back drooling on the pleather...

Thanks again for not reading the OP, btw. :bird:

[Baleful]

I am really happy to see you didn't appreciate my celeron statement Baleful. It had just as much to do with my original post as your responses to my origional post. How does it feel bitch? :bird:

Now that I am done pushing your buttons how about you answer one question straight out?

Do you or don't you agree that a higher OC @ 4Ghz or higher would likely show a point at which an x8 pcie lane would begin to bottleneck the system and affect the benchmark results?

Agree and I will forgive all statements suggesting I have no idea what I am talking about. Disagree and you can feel free to hop back onto the short bus. Choice is yours dildo baggins...

LMFAO, you're a fucking trip dude.

I'm not saying your entire post was uneducated, but this is fucking stupid "In short, if you don't plan on Overclocking your system to above 4 Ghz then you shouldn't invest in multiple graphics cards."

Now for your "direct" question - In a system with a high OC (4.0+) with Multi GPU's, the x8 lane WOULD hurt FPS by 1-2%. So YES, if you are within that "perfect" scenario, you have a 4.0+ OC, Multi GPU's, 2560x1600 res and run games at full tilt, then yeah, you'll definitely notice that x8 lane hurting your FPS by a good 1-2%.

Again, your original post wasn't to far off, but that last statement, not the metaphor, was completely fucking stupid. :bird:

[KaptCrunch]

Dude you didn't say any of that in your first post, instead you said:



Now who's on the short bus?

pci 2.0 is for can't do x16 on both cards under dual cards

another point is tri cards if they really want to show the limits of the BUSS

PCIe 2.0 delivers 5 GT/s, but employs an 8b/10b encoding scheme which results in a 20 percent overhead on the raw bit rate. PCIe 3.0 removes the requirement for 8b/10b encoding and instead uses a technique called "scrambling" in which "a known binary polynomial is applied to a data stream in a feedback topology. Because the scrambling polynomial is known, the data can be recovered by running it through a feedback topology using the inverse polynomial"[17] and does not actually use a 128b/130b encoding scheme as was reported early on in the development. PCIe 3.0's 8 GT/s bit rate effectively delivers double PCIe 2.0 bandwidth. According to an official press release by PCI-SIG on 8 August 2007

[0xygenthief]

LMFAO, you're a fucking trip dude.

I'm not saying your entire post was uneducated, but this is fucking stupid "In short, if you don't plan on Overclocking your system to above 4 Ghz then you shouldn't invest in multiple graphics cards."

Now for your "direct" question - In a system with a high OC (4.0+) with Multi GPU's, the x8 lane WOULD hurt FPS by 1-2%. So YES, if you are within that "perfect" scenario, you have a 4.0+ OC, Multi GPU's, 2560x1600 res and run games at full tilt, then yeah, you'll definitely notice that x8 lane hurting your FPS by a good 1-2%.

Again, your original post wasn't to far off, but that last statement, not the metaphor, was completely fucking stupid. :bird:

Thank you! I agree the my summary statement could use some work. However spot on the statement might be in a general sense, it really didn't summarize my OP well. Sorry for calling you a retard, I just hate it when people ignore the content of an OP and focus on semantics. Would have been even better if someone would have pointed out a mispelled word or two!

Now :htfu: and help me get ND40oz of the short bus!

[Baleful]

Thank you! I agree the my summary statement could use some work. However spot on the statement might be in a general sense, it really didn't summarize my OP well. Sorry for calling you a retard, I just hate it when people ignore the content of an OP and focus on semantics. Would have been even better if someone would have pointed out a mispelled word or two!

Now :htfu: and help me get ND40oz of the short bus!

:rofl: :rofl: :rofl:

Nigga, yur ass is short bus central, you're just to fucking short bus to realize it :rofl:

[0xygenthief]

:rofl: :rofl: :rofl:

Nigga, yur ass is short bus central, you're just to fucking short bus to realize it :rofl:

:moon: My ex would probably agree with you there. :bird:

"A crazy person doesn't lose his/her mind.... It just becomes something more entertaining!"

[ND40oz]

pci 2.0 is for can't do x16 on both cards under dual cards

another point is tri cards if they really want to show the limits of the BUSS

What does this have to do with anything I posted? Now go back and figure out how to type a complete sentence, instead of "pci 2.0 is for can't do x16", what the hell does that mean?

PCIe 2.0 doubled the bandwidth from 1.1, is that what you're trying to get at? As in a PCIe 2.0 8x slot has the same bandwidth as a PCIe 1.1 16x slot?

Thank you! I agree the my summary statement could use some work. However spot on the statement might be in a general sense, it really didn't summarize my OP well. Sorry for calling you a retard, I just hate it when people ignore the content of an OP and focus on semantics. Would have been even better if someone would have pointed out a mispelled word or two!

Now :htfu: and help me get ND40oz of the short bus!

Your summary reaffirms the fact you are a retard. Considering most of us have been running dual video cards all the way back to the nForce 4, 955x/975x platforms, I'd have to say making a blanket statement like:

In short, if you don't plan on Overclocking your system to above 4 Ghz then you shouldn't invest in multiple graphics cards. It's the equivalent of putting twin turbos on a bicycle.

makes you a retard.

[0xygenthief]

What does this have to do with anything I posted? Now go back and figure out how to type a complete sentence, instead of "pci 2.0 is for can't do x16", what the hell does that mean?

PCIe 2.0 doubled the bandwidth from 1.1, is that what you're trying to get at? As in a PCIe 2.0 8x slot has the same bandwidth as a PCIe 1.1 16x slot?



Your summary reaffirms the fact you are a retard. Considering most of us have been running dual video cards all the way back to the nForce 4, 955x/975x platforms, I'd have to say making a blanket statement like:



makes you a retard.

LOL! Just like ignoring the intent of the OP makes you a retard! Gotta love it!
:bird:
I don't suppose you will be jumping off the short bus anytime soon. So all I can say is keep nit-picking at something I already made amends for, it makes you seem super intelligent and offers the rest of us much needed entertainment!
:deadhorse:

:htfu:

[KaptCrunch]

find me a MB that will accept 2 or 3 video cards

all cards running @ x16

the buss is too short lol 32 total on buss
http://www.legitreviews.com/images/reviews/782/nvidia_x58_3.jpg

[Ketzer7]

:popcorn:

BTW, not to split hairs, but NF200s do allow for all full x16 bandwidth slots, albeit at a slight performance hit due to the switching nature of the chip to enable all of those lanes.

Also, didn't TPU do an article on 480 PCIe scaling? I thought the take away was that x8 2.0 slots didn't suffer that much of a performance hit compared to x16 2.0 slots? :shrug:

[Baleful]

Also, didn't TPU do an article on 480 PCIe scaling? I thought the take away was that x8 2.0 slots didn't suffer that much of a performance hit compared to x16 2.0 slots? :shrug:

They don't suffer at all until you throw HUGE resolutions at it and up the eye candy to full. Even then the performance hit is extremely low.

[Ketzer7]

They don't suffer at all until you throw HUGE resolutions at it and up the eye candy to full. Even then the performance hit is extremely low.

Yeah, that's what I thought as well.

I'm hoping to try running 3x 1920x1080 LCDs on a 3x SLI 480 setup when I get done someday here. I figure although the performance increase from 2x -> 3x 480s isn't that great that it was a safer bet using all 3 at a 5760x1080 resolution.. Unless I'm still overkilling it :hmm2:

[ND40oz]

LOL! Just like ignoring the intent of the OP makes you a retard! Gotta love it!
:bird:
I don't suppose you will be jumping off the short bus anytime soon. So all I can say is keep nit-picking at something I already made amends for, it makes you seem super intelligent and offers the rest of us much needed entertainment!
:deadhorse:

:htfu:

Nope, you're still a retard. I'm not ignoring anything in your OP, the content just doesn't matter when you sum it up with a retarded statement saying people shouldn't go multigpu unless they're overclocking any cpu to 4.0 GHz.


find me a MB that will accept 2 or 3 video cards

all cards running @ x16

the buss is too short lol 32 total on buss
http://www.legitreviews.com/images/reviews/782/nvidia_x58_3.jpg

:popcorn:

BTW, not to split hairs, but NF200s do allow for all full x16 bandwidth slots, albeit at a slight performance hit due to the switching nature of the chip to enable all of those lanes.

Also, didn't TPU do an article on 480 PCIe scaling? I thought the take away was that x8 2.0 slots didn't suffer that much of a performance hit compared to x16 2.0 slots? :shrug:

See Ketzer's statement, I have no problem getting full 16x performance out of 3 slots on my P6T6 Revo. You'll see a little extra latency because of the switch chip. There are a few dual tylersburg boards (5520) but I'm not sure if anyones using them for full 16x bandwidth, most of them seem to have 7 8x lanes instead.

[0xygenthief]

I'm not ignoring anything in your OP, the content just doesn't matter when you sum it up with a retarded statement saying people shouldn't go multigpu unless they're overclocking any cpu to 4.0 GHz.


You say you aren't ignoring it yet you haven't made one single contribution to either support or not support the OP intent? Some would call that derailing a thread because you are being a little bitch. Others might say you are just being retarded. Though the ones that believe all of your posts are the result of having one too many guys giggle at your smurf sized cock get my vote! :fag:

See, I too can focus in on something besides the OP intent! :bird:

Who here wants to see a picture of ND40oz's frighteningly small penis?:jackoff:

[ND40oz]

You say you aren't ignoring it yet you haven't made one single contribution to either support or not support the OP intent? Some would call that derailing a thread because you are being a little bitch. Others might say you are just being retarded. Though the ones that believe all of your posts are the result of having one too many guys giggle at your smurf sized cock get my vote! :fag:

See, I too can focus in on something besides the OP intent! :bird:

Who here wants to see a picture of ND40oz's frighteningly small penis?:jackoff:

Stop being a little bitch, your OP was poorly written and you chose to sum it up with a blanket statement saying people shouldn't go SLI/Crossfire unless they're overclocking their processors to 4.0 GHz.

If you were really trying to say that the Hard article was flawed because they really couldn't test the bandwidth limits of PCIe 2.0 slots at 8x because they didn't have an fast enough processor, then say that and stop there. Don't throw out a useless statement about not investing in multiple gpus unless you're overclocking to 4.0+ GHz. This isn't rocket science, make your point and stop there. The whole point of Baleful and I saying WTF is that last statement, which is utterly false misinformation. Stick to the facts and people won't give you shit.

[Overclocking101]

fact is no matter what the cpu clock an X8 lane will give damn near a full X16, only noticable differece is in synthetic benches maybe 1-2fps but not likely. todays gpu's exceed the power of pci-e2.0 but nowhere near the bandwidth, hence why pci-3.0 has not launched yet even though it was announced 2 years ago or longer. it wont be launched until a single gpu card is able to exceed 2.0's bandwidth

[0xygenthief]

Edited OP to clarify a few things. Enjoy!

[coolmiester]

Ohhhhhh............i missed this one :popcorn:

Its obviously something that you're quite interested in so surely, it can't be that hard to do some of your own first hand testing to prove/justify your point instead of trying to justify your silly opening gambit on someone else's work :shrug:

[0xygenthief]

Ohhhhhh............i missed this one :popcorn:

Its obviously something that you're quite interested in so surely, it can't be that hard to do some of your own first hand testing to prove/justify your point instead of trying to justify your silly opening gambit on someone else's work :shrug:

I'd like to, and am going to try. Though I am unsure I can with my motherboard. I have 3 480gtx cards to work with and 2 x16 lanes and 1 x8 lane. I am just not sure I can run SLI on the second and third cards while ignoring the first. In other words, taking out the first GPU for x16/x8 testing may not work at all. We will see though.

Oh and I almost forgot :bird: @ silly opening gambit!

[Ketzer7]

2 x16 slots and 1 x8 slot? How is that possible on an X58 board? Those don't add up somehow without an NF200, but then if you had an NF200, you should have 3x full x16 slots instead of 2 and 1 x8. Without an NF200 onboard, the best you can get from X58 is x16/x8/x8 :confused:

[ND40oz]

Lol, I like the edits. :moon:

While you're doing your testing, you may want to test out boards with the NF200 chips as well. Theorectically if you have two cards sharing the NF200 bridge chip, they'll communicate to each other at full 16x speeds through the NF200, negating any slow down for communication done over the chipset (X58)/processor (P55). Although they should be doing most of the card to card communication over the SLI bridge anyway.

Lot of info on [H] about NF200 and PCIe bandwidth, some of us have been down this road before in the name of F@H and other CUDA apps.

[coolmiester]

All i'm saying is, you dug a deep fucking hole with that silly opening gambit, now lets see how well you can dig your way out of it because at this point i only see one person trying to de-rail to save face ;)


:htfu: and get testing to prove your point.

[Baleful]

This thread is completely fucking useless at this point. :shake:

[0xygenthief]

2 x16 slots and 1 x8 slot? How is that possible on an X58 board? Those don't add up somehow without an NF200, but then if you had an NF200, you should have 3x full x16 slots instead of 2 and 1 x8. Without an NF200 onboard, the best you can get from X58 is x16/x8/x8 :confused:

My motherboard has three PCIe slots first two are x16 third is x8. When all three are filled the second lane goes into x8 mode. So to test this I will have to have the second and third filled with nothing in the first slot. Again, I am unsure if this will work.

I am going to see if there is a BIOS setting that will enable me to switch the second slot to x8 with only the first and second slot filled. Going to check this once I get home.

As for testing this against other boards, sorry, no money for that at the moment.

[Ketzer7]

I still don't think it works like that. X58 has a max of 36 PCIe lanes, so true x16/x16/x8 would be 40 lanes total, and more than X58 can do. What board is this? I dunno, but it sounds like they have it mislabeled or something because there's no way with 3 PCIe vid cards you get x16/x16/x8 on X58 without an NF200 or some other bridge chip.

From what you are describing, if you plug cards into the 2nd and 3rd slots, your cards will both be running x8. The slave PCIe slot of two x16's doesn't run at x16 if you don't plug anything into the first one, it's a one way deal, you get x8 only regardless if anything is plugged into the first x16 or not.

[0xygenthief]

Ok, I figured out how to run some benchmarks in both configurations. All I have to do is add a third gpu into the x8 slot. This will put the second slot into x8 mode allowing me to bench using x16/x8. Removing the third card resets the second card to x16 mode for the final tests.

I thought I'd try some Vantage runs but for whatever reason Orb doesn't appear to be up this evening. If I had Crysis installed I'd run through the timedemo on it. Anyone have any other ideas for benchmarks to run for this test? The ones HardOCP ran were all custom benchmarks...

[ND40oz]

You can use this to get a feel for the bandwidth to each card:

http://www.evga.com/forums/tm.aspx?m=253215&mpage=1

You could also try out an encoding app that uses Cuda, such as TMPGEnc and see if bandwith limitations come into play.

http://tmpgenc.pegasys-inc.com/oem/tutorial/en/tutorial.html

[Crow]

When did the Red in Real Red Raider start referring to your tampons? Man you guys bitch a LOT.

[ballz0r]

Ok, I figured out how to run some benchmarks in both configurations. All I have to do is add a third gpu into the x8 slot. This will put the second slot into x8 mode allowing me to bench using x16/x8. Removing the third card resets the second card to x16 mode for the final tests.

I thought I'd try some Vantage runs but for whatever reason Orb doesn't appear to be up this evening. If I had Crysis installed I'd run through the timedemo on it. Anyone have any other ideas for benchmarks to run for this test? The ones HardOCP ran were all custom benchmarks...

im interested in seeing your results... a lot of tests ive seen show only 2 or 3 fps between x16x16x16 and 16x16x8

:popcorn:

[0xygenthief]

I should have some time to run through some benchies this weekend. I plan on only performing Vantage and Crysis benchmarks due to the simplicity of the testing. Plus with only one 30 inch monitor I can't replicate the ultra-high resolutions that HardOCP did in their review. The crysis timedemo, though several years old, is still a demanding demo that should strain my system and perhaps yeild the results I am looking for. The Vantage bench, on the other hand, will be running at the standard resolution, which may not strain the PCIe lanes enough, we shall see. Both are canned benchies but for the purposes for this test, I don't see a need to do any others.

If anyone has some custom bench runs I could port over to my system I'd be happy to use those as well, I simply don't want to waste the time in making my own.

Results to follow.

[0xygenthief]

I found a little light reading on the subject...

Apparently Toms Hardware had done some pcie scaling tests of their own. Granted these tests are X16/x16 vs x8/x8 as opposed to x16/x8 so I will still be performing my own tests to verify.

http://www.tomshardware.com/reviews/pcie-geforce-gtx-480-x16-x8-x4,2696-12.html

The MW2 benchmark reinforces my assertion that a 3.6 Ghz clock on HardOCP's processor would/could end up being the bottleneck during their testing. Not all games are as CPU dependant as MW2 but clearly this IS an issue (and consequently the point of my original post).

"A serious CPU “bottleneck” is the most likely cause for decreased SLI scaling at lower resolutions. For most games, it doesn’t even make sense to test a pair of GeForce GTX 480 graphics cards at anything less than 2560x1600, and one benchmark was completely crippled by the performance of our 4.00 GHz CPU, even at 1920x1200."

[RedRaider]

Crossfire is for :fag:'s

[Baleful]

I found a little light reading on the subject...

Apparently Toms Hardware had done some pcie scaling tests of their own. Granted these tests are X16/x16 vs x8/x8 as opposed to x16/x8 so I will still be performing my own tests to verify.

http://www.tomshardware.com/reviews/pcie-geforce-gtx-480-x16-x8-x4,2696-12.html

The MW2 benchmark reinforces my assertion that a 3.6 Ghz clock on HardOCP's processor would/could end up being the bottleneck during their testing. Not all games are as CPU dependant as MW2 but clearly this IS an issue (and consequently the point of my original post).

"A serious CPU “bottleneck” is the most likely cause for decreased SLI scaling at lower resolutions. For most games, it doesn’t even make sense to test a pair of GeForce GTX 480 graphics cards at anything less than 2560x1600, and one benchmark was completely crippled by the performance of our 4.00 GHz CPU, even at 1920x1200."

Not a bad review, but you have to take all Tom's Hardware reviews with a truck load of salt.

I like how they state that the CPU was the "bottleneck" for the MW2 tests, yet never test how CPU frequency actually scaled. To me, for that to be a valid statement, I need to see that 920 at stock clocks, then gradually move up to 4.0Ghz. Show us what the difference that CPU frequency makes, don't just blindly state that your 4.0Ghz OC is limiting your bandwidth because honestly, that sounds like a crock of shit. Virtually every game they tested in SLI under x8 mode was only affected by maybe 2%.... except for MW2.

[Ketzer7]

TPU also did a similar article which was the most extensive of its kind that I've seen so far:

http://www.techpowerup.com/reviews/NVIDIA/GTX_480_PCI-Express_Scaling/

[magoo25]

Look, I read HardOCP alot.
They have done a number of articles lately comparing CF and SLi, the 8x vs 16x question and as far as I can see they covered their bases quite well.

I don't think there's much to argue about, you can benchmark until you're blue in the face, but the real world gameplay is not convincing of any issues of x8 vs x16, and it appears that the GTX 460 kicks even the 5870 CFs ass.
(and this comes from a CF owner)

[0xygenthief]

Not a bad review, but you have to take all Tom's Hardware reviews with a truck load of salt.

I like how they state that the CPU was the "bottleneck" for the MW2 tests, yet never test how CPU frequency actually scaled. To me, for that to be a valid statement, I need to see that 920 at stock clocks, then gradually move up to 4.0Ghz. Show us what the difference that CPU frequency makes, don't just blindly state that your 4.0Ghz OC is limiting your bandwidth because honestly, that sounds like a crock of shit. Virtually every game they tested in SLI under x8 mode was only affected by maybe 2%.... except for MW2.

I noticed the lack of CPU testing despite the claim as well. Which is why I still plan on doing my own benchies to verify. Processor speed does have a measurable effect on single card configurations, several reviews out there to show this. I think it makes sense that, though the effect may be smaller, there is still an effect. Who knows, the point at which a x16/x8 configuration will show a performance hit may not come until a CPU is clocked at 6 Ghz? I simply didn't like the fact that CPU speed had zero presence in the HardOCP article. What if, the turning point for a x8 lane is 4.2 or 4.4 Ghz? Definately reachable with watercooled CPUs. It might be enough to influence a buyer to go with a 1366 system as opposed to a 1156 system to get full use of their GPUs. I think the Toms Hardware article tried to make this point as well though they neglected to back it up with any sort of testing.