PCI-e x16 speeds

[Snipe343]

whats benifet from runningat x16 than x8 i have mine in the 8x slot because the case is not in line with the mobo all the way

[kwchang007]

whats benifet from runningat x16 than x8 i have mine in the 8x slot because the case is not in line with the mobo all the way

What the case isn't in line with the mobo? Well depends on the card....for a 8600..probably little to no difference.

[ccleorina]

whats benifet from runningat x16 than x8 i have mine in the 8x slot because the case is not in line with the mobo all the way

2% or 3% difference from 16x to 8x

[Snipe343]

ok i dont why but the 1st pci e slot is out of line with the slot on the case

[kwchang007]

ok i dont why but the 1st pci e slot is out of line with the slot on the case

Defected case, or improper install? Can you post some pics, I'd like to see this, lol. But if all else fails, try bending the metal out of your way or something.

[ccleorina]

ok i dont why but the 1st pci e slot is out of line with the slot on the case

Post some pic... we like to see it...

[JrRacinFan]

Full ATX motherboard in a ATX case with only 6 expansion slot openings? Hmmmm .....

[Snipe343]

dont have a camera right now ill try to post picks later, i hate the mobo anyway its terrible i would love another excuse to get rid of it

[AddSub]

2% or 3% difference from 16x to 8x

Hmm, aren't differences only noticable in SLI/Crossfire setups (with AA/AF maxed) and not in single card ones?

[imperialreign]

a x8 slot is a bit slower than a x16 slot, so one might notice a FPS difference between the two, depending on other hardware.

but, quite a few mobo's out there drop the x16 speeds to x8 in xFire or SLI, even without AA/AF set to max. There are also a mobo's that don't drop the bandwidth, either . . .

[Kursah]

I think some higher end PCI-e cards may notice a bigger fall when going from x16 to x8 link speed. But most mid-range and lower end cards don't seem to use the full x16 bandwidth, so maybe 8x link is fine, but really what would the point be if there is no real change aside from messing around? You won't gain anything by dropping speed, and odds are you won't lose much, but I just don't see why you would want to run it at 8x all the time.

Sure in some X-fire setups the second card will drop speed, take my P5B Deluxe, the first card runs at full x16, the second runs at a cut-at-the-knees 4x. But I'm not a fan of dual cards, but I can usually get one card, OC it, and make it run as fast as I need for my games. You should run some benches at 8x and 16x and see what the difference is! Maybe try each at 100Mhz PCI-e and 105/109Mhz PCI-e...I wonder if there is anything there worth looking at...probably not. Just a tired thought...and after that I'm off to bed.

:toast:

[AddSub]

most mid-range and lower end cards don't seem to use the full x16 bandwidth

yeah, I believe that is the case.

[Kursah]

Too bad there wasn't a way to measure how many of the x16 links current cards are using. I wonder if increasing PCI-e Bus speed to the USED serial links of the PCI-e bus could improve performance. I mean this is just in theory, and I'm at work so I can't test it w/my x1950pro, but what if the link speed is reduced to 8x and PCI-e is set to 109 or 115+ Mhz if there could be performance there.

I know on newer NV chipsets and newer NV cards (8800's) that increasing the x16 PCI-e bus speed can improve performance...but if a lower multi, higher speed PEG/PCI-e link was established for let's say 8600's, 2600's, pro's/gt's and such if there could be some extracted extra performance...probably not but I will try when I get home in about 7 hours.

If anyone else is interested in testing this theory that'd be cool, I won't assume it will hinder or help performance until I test it and see some results. It can't hurt to try and see what happens w/the general synthetic benches and see if there's anything there.

Anyone interested in trying? I may be able to check back in a little while, depends on how busy I get.

:toast:

[kwchang007]

Hmm, aren't differences only noticable in SLI/Crossfire setups (with AA/AF maxed) and not in single card ones?

Yeah, but same theory applies if this guy can't get his card all the way into the x16 slot and only makes contact with 8 lanes...how that happens I have no clue.

[imperialreign]

Anyone interested in trying? I may be able to check back in a little while, depends on how busy I get.

I'm willing to give it a go with my 1950 later when I have time, I know my card is defi a mid-range, but any possible attempt to get some extra juice out of it . . . I haven't really messed with the PCIe frequency setting in BIOS yet either.

Yeah, but same theory applies if this guy can't get his card all the way into the x16 slot and only makes contact with 8 lanes...how that happens I have no clue.

y'know . . . I didn't think you could fit an x16 card into an x8 slot . . . like how an x8 won't fit in an x1 . . . seems odd, IMO

[kwchang007]

y'know . . . I didn't think you could fit an x16 card into an x8 slot . . . like how an x8 won't fit in an x1 . . . seems odd, IMO

Neither do I. He says it's a x16 slot, but since the case doesn't line up he doesn't get full contact :confused:. Extremely confusing imo.

[Darknova]

There was an article on Tom's Hardware about PCI Express Scaling (The link is now dead..seems like they removed the article). Anyway they used the 7 series nVidias and 1950 series ATi cards, and by taping off parts of the connector they found they could reduce the number of lanes. So they tested this, and nVidias were the worst for PCI-E scaling. At 8x most of them lost nearly 40% performance. Where as ATi's were hardly hit, it wasn't until they got to 4x that performance halved, by this stage the nVidia's were at something like 20%.

[Kursah]

Yeah I remember that article, kinda fuzzy, but I do remember skimming over it. If I have some time I'm still gonna see what happens at 8x/109+MHz, probably, little to nothing, but it's worth a shot.

:toast:

[AddSub]

If I have some time I'm still gonna see what happens at 8x/109+MHz

Excellent. I would be really interested in your results.

[imperialreign]

played with the PCI-e frequency setting in BIOS some, not much, though, as it's getting late here. I'll mess with it some more tomorrow, as I'm interested in seeing how high that setting can go before I start getting errors.

Anyhow, there does seem to be a performance gain for a x1950. I used 3m06 for testing purposes, and a minimum of 15 minutes of cool down to idle temps before each change/run. Tested at the stock clock speeds of 575/1390, with ATi Tray Tools 3D profile set to "Auto Balanced". 3m06 at defaults. All other PCIe settings in BIOS (labeled as Peg Root controls) were all left to auto.

PCIe @ 100 (the most commonly *recommended* setting) = 4192 marks
PCIe @ 105 = 4193 marks
PCIe @ 110 = 4241 marks
PCIe @ 120 = 4306 marks

so, it does seem to help. Although, I want to test higher settings, and I also want to test having it set to [AUTO]. From there, I want to see how much of a difference it makes running the card 5% OC (the highest 100% stable I can manage :ohwell:), and if there is any change with tweaking the Peg controls (specifically, slot voltage and bandwidth).

[Mussels]

played with the PCI-e frequency setting in BIOS some, not much, though, as it's getting late here. I'll mess with it some more tomorrow, as I'm interested in seeing how high that setting can go before I start getting errors.

Anyhow, there does seem to be a performance gain for a x1950. I used 3m06 for testing purposes, and a minimum of 15 minutes of cool down to idle temps before each change/run. Tested at the stock clock speeds of 575/1390, with ATi Tray Tools 3D profile set to "Auto Balanced". 3m06 at defaults. All other PCIe settings in BIOS (labeled as Peg Root controls) were all left to auto.

PCIe @ 100 (the most commonly *recommended* setting) = 4192 marks
PCIe @ 105 = 4193 marks
PCIe @ 110 = 4241 marks
PCIe @ 120 = 4306 marks

so, it does seem to help. Although, I want to test higher settings, and I also want to test having it set to [AUTO]. From there, I want to see how much of a difference it makes running the card 5% OC (the highest 100% stable I can manage :ohwell:), and if there is any change with tweaking the Peg controls (specifically, slot voltage and bandwidth).

Please test this some more and compile your results into a single post. Once you get some good results, i'll test out the most interesting choices (if you hit a wall, or one result is faster than the others) and we'll see how it compares % wise to an 8800GTX.

[Tommithy]

There was an article on Tom's Hardware about PCI Express Scaling (The link is now dead..seems like they removed the article). Anyway they used the 7 series nVidias and 1950 series ATi cards, and by taping off parts of the connector they found they could reduce the number of lanes. So they tested this, and nVidias were the worst for PCI-E scaling. At 8x most of them lost nearly 40% performance. Where as ATi's were hardly hit, it wasn't until they got to 4x that performance halved, by this stage the nVidia's were at something like 20%.


That Tom's Hardware article is Here (http://www.tomshardware.com/2007/03/27/pci_express_scaling_analysis/)

[Tatty_One]

There is plenty enuff bandwidth available at 8x for an 8600.....you only get the minor slowdowns when you use up all the bandwidth......an 8600 is very unlikely to do that, damn in one test I read.....2 x 8800GTX's at 16 x 16 against the same at 8 x 8 (the 680 V 650 boards showdown) only showed 2-3FPS difference at very high res in most games.

[Tatty_One]

played with the PCI-e frequency setting in BIOS some, not much, though, as it's getting late here. I'll mess with it some more tomorrow, as I'm interested in seeing how high that setting can go before I start getting errors.

Anyhow, there does seem to be a performance gain for a x1950. I used 3m06 for testing purposes, and a minimum of 15 minutes of cool down to idle temps before each change/run. Tested at the stock clock speeds of 575/1390, with ATi Tray Tools 3D profile set to "Auto Balanced". 3m06 at defaults. All other PCIe settings in BIOS (labeled as Peg Root controls) were all left to auto.

PCIe @ 100 (the most commonly *recommended* setting) = 4192 marks
PCIe @ 105 = 4193 marks
PCIe @ 110 = 4241 marks
PCIe @ 120 = 4306 marks

so, it does seem to help. Although, I want to test higher settings, and I also want to test having it set to [AUTO]. From there, I want to see how much of a difference it makes running the card 5% OC (the highest 100% stable I can manage :ohwell:), and if there is any change with tweaking the Peg controls (specifically, slot voltage and bandwidth).

Is that not frequency your testing there as opposed to bandwidth? As in speed of port as opposed to size of port.

[kwchang007]

Is that not frequency your testing there as opposed to bandwidth? As in speed of port as opposed to size of port.

Well bandwidth is a factor of speed multiplied by width, isn't it?

[imperialreign]

Is that not frequency your testing there as opposed to bandwidth? As in speed of port as opposed to size of port.

well, to the best of my knowledge, the PCIe frequency is directly related to the slot bandwidth . . . someone correct me if I'm wrong, but . . . the frequency is like it's BUS speed, how fast it can transfer and recieve data, whereas the bandwidth is how much data it can transfer and recieve.

in my system BIOS, though, I don't have any settings for the PCIe bandwidth - I believe bandwidth is automatically determined by BIOS. But, aside from the slot frequency, I do have settings for buffer length, link latency, root control, link mode and slot power.

I'm sure there will be difference between mobos, though

Please test this some more and compile your results into a single post. Once you get some good results, i'll test out the most interesting choices (if you hit a wall, or one result is faster than the others) and we'll see how it compares % wise to an 8800GTX

sure thing, I prob won't have a post on these test until sometime this weekend, though.

Question for anyone, seeing as how I've heard/read LOTS of different things about PCIE slot frequency settings - opinions on: how liable is it too high a setting could torch the GPU/mobo? Is the freq setting really only for keeping up with an OC CPU? Does this setting function similar to a standard PCI BUS frequency setting?

My own thoughts to those questions: I'm sure too high a setting for too long would damage something as with all components, but I think you'd run into some form of comm error with the card long before an actual serious concern arose. I think the setting has a lot to do with keeping up with an OC CPU, but I also think that it can help remove comm bottlenecks with some procs. And yeah, I think also that the setting is very similar to PCI BUS.
What are y'alls opinions?

[Tatty_One]

Well bandwidth is a factor of speed multiplied by width, isn't it?

Yes of course it is, but my point, perhaps explained poorly is that if a card is not using all the available bandwidth in any case then increasing it is not goint to bring any benefits?

[kwchang007]

Yes of course it is, but my point, perhaps explained poorly is that if a card is not using all the available bandwidth in any case then increasing it is not goint to bring any benefits?

No it shouldn't, in all realism results within x% of the other should be considered the same. Idk what the percentage usually is though....

[imperialreign]

still compiling more results. I hit a wall at 140. I tried increasing slot power - no dice; increasing buffer length, slowing link latency . . . I even tried upping the voltage to the SSB and NSB. 140 slot frequency just won't pass POST. I'm thinking this has more to do with the cards architecture than anything else. It's also possible it might work later, if I OC the CPU and BUS speeds some more - but I'm not getting into that just at the moment.

[psychomage343]

what were your scores at 140??

[imperialreign]

couldn't POST at 140, so I backed it down to 135.

at 135 it pulled 4306 marks in 3m06, with stock clocks and drivers set to auto balanced.
5% GPU OC at 135 yielded 4356 with drivers auto balanced.

But, at 130MHz, at stock speeds it pulled 4313 marks, and 4391 with a 5% OC.

so . . . it appears that there is a performance plateau, and beyond that scores start coming back down. I haven't had the chance yet to see if those slot frequencies would allow for a higher OC to the card, but I really don't think it would make a difference as I get the feeling that drop is due to the cards architecture.

I still want to test it one MHz at a time, instead of by 5MHz incremements, to see where exactly performance starts to drop off, and what exact speed fails POST - I already know 140MHz will. Once I find the highest, lowest and the median scores, I'm going to take it a step further and see how the Peg Link Controls function with the slot frequency and GPU clocks. Highest 3m06 score I managed with this card was 4616 (IIRC), and that was at 100MHz slot . . . I'm almost 100% certain I can get this 1950 PRO to break 5000 marks with a P4.

[Kursah]

Nice work! I haven't had time to do anything, but all I've ever messed with is PCI-e MHZ (it goes from like 80-150, the highest I've POST'd at is 119, 120 won't post), and PEG Link Speed, which goes from Auto, Normal, Fast, Faster. I haven't messed with latency, the best score I've attained is 5552 on my current rig.

Your results are most certainly interesting. Plus I am pretty sure I can't pass 120 on PCI-e due to using an SATA HDD, so you may have a plus for "still using IDE". :toast:

[Tatty_One]

I have always run mine at 110Mhz I forgot to say!

[kwchang007]

couldn't POST at 140, so I backed it down to 135.

at 135 it pulled 4306 marks in 3m06, with stock clocks and drivers set to auto balanced.
5% GPU OC at 135 yielded 4356 with drivers auto balanced.

But, at 130MHz, at stock speeds it pulled 4313 marks, and 4391 with a 5% OC.

so . . . it appears that there is a performance plateau, and beyond that scores start coming back down. I haven't had the chance yet to see if those slot frequencies would allow for a higher OC to the card, but I really don't think it would make a difference as I get the feeling that drop is due to the cards architecture.

I still want to test it one MHz at a time, instead of by 5MHz incremements, to see where exactly performance starts to drop off, and what exact speed fails POST - I already know 140MHz will. Once I find the highest, lowest and the median scores, I'm going to take it a step further and see how the Peg Link Controls function with the slot frequency and GPU clocks. Highest 3m06 score I managed with this card was 4616 (IIRC), and that was at 100MHz slot . . . I'm almost 100% certain I can get this 1950 PRO to break 5000 marks with a P4.

Well not so much a plateau as 3d mark gives you slightly different results every time you run it.

[imperialreign]

Well not so much a plateau as 3d mark gives you slightly different results every time you run it.

well, I mean, yeah, no matter what the configuration . . . I'd have to run 5+ benchmarks back to back to get a median score, and that could still be different than two days from now - but, I can at least see a noticeable difference between various tests. It's not like 5-10 marks different, it's upwards of 50+.

I just meant that it seems theres a point where performance doesn't go up, and scores are similar across a few differen MHz testing, then they seem to start dropping off again.

I'm going to test things some more later, though . . .

[Tatty_One]

well, I mean, yeah, no matter what the configuration . . . I'd have to run 5+ benchmarks back to back to get a median score, and that could still be different than two days from now - but, I can at least see a noticeable difference between various tests. It's not like 5-10 marks different, it's upwards of 50+.

I just meant that it seems theres a point where performance doesn't go up, and scores are similar across a few differen MHz testing, then they seem to start dropping off again.

I'm going to test things some more later, though . . .

I read somewhere a while ago....the greatest benefits seem to come between 105 -115Mhz. thats kind of why I set mine at 110.

[imperialreign]

I read somewhere a while ago....the greatest benefits seem to come between 105 -115Mhz. thats kind of why I set mine at 110.

I'm sure there's difference from one rig to the next, but . . . as of what testing I've done, that looks to be about right for my rig, too.

[Mussels]

I read somewhere a while ago....the greatest benefits seem to come between 105 -115Mhz. thats kind of why I set mine at 110.

i run 105 out of habit (a board i had once, needed to be 105 or higher to lock the PCI/PCI-E bus for overclocking) so i might change that habit to 110 :)

[imperialreign]

Well, I've finished my first batch of tests . . . kind interesting, I think.

Again, all PCI-E controls in BIOS were left to [AUTO], no changes were made to CPU/SYS configuration. I ran 3m06 with the drivers set to "Auto Balanced", at stock GPU/MEM speeds, and then with a 5% OC. I allowed 15min cool down to idle temps before each run. I tested slot frequency in 5MHz intervals, and what results I have for my system might not necessarily be what someone else will get (did I really need to mention that :D). Anyhow, I assembled my results into a couple of line graphs:

http://img529.imageshack.us/img529/7900/pciecharttz9.jpg


this represents the overall 3DMark06 scores I recieved after each run, at stock speeds and with a 5% OC to the GPU/MEM. I thought it rather interesting how much of a performance drop there was at 100MHz, compared to running below or above that mark. Again, I still couldn't pass POST with the frequency set at 140MHz, and 90 is the lowest option I have in BIOS.

For giggles, I also compiled a line graph of the various test scores . . . SM2.0, HD/SM3.0, CPU:

http://img516.imageshack.us/img516/373/pciechart2vo4.jpg

you can see here how CPU score doesn't seem as affected by the 100MHz performance hit as much as the 1950 is. Also, it's plainly visible how 'inconsistent' scores are across the board. Increasing the slot frequency by 5MHz doesn't necessarily mean that there will be an improvement, as it tends to go up and down - although, staying around the same range.

I also found it interesting that the SM2.0 test results seem to fluctuate more than the SM3.0 tests - I almost wonder if that might be driver related to the card, the only way to know would be to change drivers and re-test . . . although, I'd still like to do more nit-pciky testing this weekend, like, test (at least) from 95MHz to 105MHz in 1MHz increments. I also want to see what affect the other PCI-E controls in BIOS have on performance, also.

[Mussels]

graphs are excellent. please keep up the good work, and try another card if you can.

[hercules71185]

At stock speeds my 8600gt on the ecs elitegroup 6100sm has no difference. But, after 2% overclock my card seemed to plateau. Now I have a full x16 motherboard. and I overclock about 10% and still see noticeable differences. My aquamark3 score went from 96,000 to 110,000 3dmark05 went from 9800 to 11500
I want to vmod still when there is a game that needs it and I can't run. At this point it doesn't seem to have a problem with any games out. but, when a game comes out that it can't handle I'll vmod it till it works flawlessly or blows and get a new one

[imperialreign]

graphs are excellent. please keep up the good work, and try another card if you can.

Sure, although . . . the only other PCI-E cards I have accessible are a VisionTek X1300 512MB GDDR2 and an ATI Radeon X700 PRO 256MB GDDR3 (still highly capable cards by todays standards).

[anticlutch]

Wow that's really interesting... nice work there imperialreign! :D

[kwchang007]

Hmm it peaks at 130mhz...does the pci-e bus employ latency settings like nb to cpu does? Maybe that's why there's an optimal speed..

[imperialreign]

Hmm it peaks at 130mhz...does the pci-e bus employ latency settings like nb to cpu does? Maybe that's why there's an optimal speed..

supposedly - there are some settings in BIOS for the PCI-E slot, but currently, they're all set to [AUTO]. I was wanting to tweak some of those settings and see how the card runs



. . . come to think of it . . . Anyone that uses a non-ASUS mobo - do you have any PCI-E settings in your BIOS?

For ASUS mobo users - what PEG control options do you have?

A thought: especially for us ASUS owners, would anyone be interested in writing a group-effort guide on the ASUS PEG Link Controls? I've been trying to do some research across the internet, and trying to get a better understanding of how PCIE functions, but information on each individual setting can be hard to come by, and even then, most sources aren't too sure how much is affected . . . I just had the thought that a decent guide on these settings, tested across a few different system configurations, might shed some more light on things and possibly give us the ability to write a guide that others haven't yet.

Opinions?