PCI Express And SLI Scaling: How Many Lanes Do You Need?

[7amood]

The articles just don't stop... I love it...
keep them coming...

 

[demonhorde665]

very interesting article , really enlightening on my optiuons to reduce cost on my next comp build going with x8 x x8 doesnt sound so bad any more. on another note , i'd like to toms do some bench work on thermal paste , because for years evwery one has hailed artic 5 as the "bomb" of thermal paste , but in more reccent times i have heard that there are actually much better alternatives out there now , so an article testing different thermal past on the (exact) same cpu/heat sink combo would be awsome though i suppse sucha test might take abit more time , since many thermal paste solutions (such as arctic 5) require a month or 2 of settign time before they reach peak colling performance .. still it would be an article to die for if you ask me.

 

[Petey1013]

[citation][nom]sambadagio[/nom]For all your fps hunters, I bet you only have a screen at home with 50 or 60Hz. So just for your information, everything above 50 or 60fps is just useless... In this aspect, a PCIx 4x is actually enough... ;-)[/citation]

This is completely false. Even on a 60hz screen there is a noticeable difference above 60 fps, especially if your playing First Person Shooters. Don't spread false info.

 

[ares1214]

[citation][nom]Petey1013[/nom]This is completely false. Even on a 60hz screen there is a noticeable difference above 60 fps, especially if your playing First Person Shooters. Don't spread false info.[/citation]

Hmmmm, sorry to burst your bubble, but hes right... more or less, Hz=fps, 60Hz=60fps, anything above that the screen cant react fast enough to show it, nor would the human eye be able to detect it. The human eye can detect studdering going from black to white at around 30 Hz, where as all it sees is grey is about 60 Hz. That applies to video games. The only thing holding back the GPU is the monitor, maxing at 60 hz/fps, and the only thing holding back the monitor is the human eye, as it can in a way only move fast enough for 60HZ, however higher does make things look better in rare occasions, but it has nothing to do with it being blurry or studdering. Almost all LCD monitors have 50-70 HZ anyway.

 

[Crashman]

[citation][nom]ares1214[/nom]Hmmmm, sorry to burst your bubble, but hes right... more or less, Hz=fps, 60Hz=60fps, anything above that the screen cant react fast enough to show it, nor would the human eye be able to detect it. The human eye can detect studdering going from black to white at around 30 Hz, where as all it sees is grey is about 60 Hz. That applies to video games. The only thing holding back the GPU is the monitor, maxing at 60 hz/fps, and the only thing holding back the monitor is the human eye, as it can in a way only move fast enough for 60HZ, however higher does make things look better in rare occasions, but it has nothing to do with it being blurry or studdering. Almost all LCD monitors have 50-70 HZ anyway.[/citation]
Actually, it's more complicated.

60FPS is an average. If you have a 60FPS average, you might have a 40FPS minimum (below your screen refresh) and an 80FPS max. Or, you might even have a 30 FPS minimum and a 90 FPS max. And it doesn't have to be centered like that, you could have a 118 FPS max that happened only on 1 frame, and a 25 FPS minimum that happened for three consecutive frames.

Obviously you're going to notice something if your minimum stays below 25 FPS for multiple frames, so he's at least partly right.

 

[ares1214]

Of course these are averages but he stated "Theres a noticable difference above 60 fps..." If that is one number, then no, there isnt. Of course that number changes from area to area, scene to scene, and so on, but at or around 60 fps as a minimum there is no noticable gain.

 

[wolfram23]

Great article! Makes me feel good about my i5 750, P55 mobo and overclocked, crossfire 5850s. It also makes me feel good for not shelling out a few hundred more on an i7/X58 set up.

 

[ares1214]

BTW Tom's, please show the minimum FPS in addition to the average, or maybe even instead of it. They are far more useful than averages, for the exact reason Crashman stated.

 

[warezme]

[citation][nom]sambadagio[/nom]For all your fps hunters, I bet you only have a screen at home with 50 or 60Hz. So just for your information, everything above 50 or 60fps is just useless... In this aspect, a PCIx 4x is actually enough... ;-)[/citation]
For one if you value your fps, like I do, you already know of the limits of the refresh rate of your monitor. That is why I have 120Hz monitors. Everything above 60fps is not useless and is what folks with weak systems think to make themselves feel better. 60fps should be your minimum fps at the highest resolution and game settings your card can handle. Thus, the need for lots of bandwidth, fast CPU and video cards exist. Anything above that is cake.

 

[MadAdmiral]

A good resource for more proof for the reason to not build a X58 system just for gaming. I also found it interesting how little difference there was between the 8x and 4x lanes as well. While significant, it certainly adds a lot of value to the boards with a second slot at 4x that cost as much as single slot boards.

 

[Stardude82]

Ditto about showing FPS minimums...

Now somebody needs to run a similar test tweaking the latency and bus frequency. Or maybe somebody could link to one, with regards to SLI?

 

[ta152h]

This is an interesting article, but one thing that isn't mentioned is the importance of the card on these tests. I think the card was poorly chosen, because it has too much memory, and is not going to be a popular card for most readers here.

I think what you were trying to show (and correct me if I'm wrong) is whether an x58 motherboard makes sense vis-a-vis a P55 for people who want to game. While this is useful, there are other differences between a P55 and x58, like the P55 having to multiplex the memory bus when the card does memory reads, and, of course, the three memory channels. But, it's still useful information, for sure.

But, my main contention is what I mentioned first. If someone is even considering a P55, they don't have excessive cash, and wouldn't be interested in the 480, in most cases. This card has loads of memory too, and makes it harder to show differences from bandwidth. I think it would be more interesting to use the 460/768, and maybe the 460/1G too. Since the 460/768 has so much less memory, you'd almost certainly see a greater difference. Not only that, it will almost certainly be (with the other 460) far more popular. From an academic standpoint, showing both cards, would be interesting and education, so we'd have some idea (but, only an imprecise one) how memory and bandwidth react to each other.

The problem with these comments is, by focusing on negatives, it gives the false impression the article wasn't good. It was, and I liked it, and did learn something.

 

[insz]

This was a great article, and having a P55 board with x16 and x4 slots, I was interested to see the x4 numbers. Unfortunately there was no x16-x4 numbers posted. Any chance we could see these tested? I'd be interested to see if the small penalty for x4 is the same in an SLI setup, or if it gets magnified.

 

[ZappCatt]

any chance that you can start adding eyefinity/surround gaming setups to these tests?
It is pretty cheap and easy to get a mismatched setup for under $500(unlike 25600x1600)

 

[mattmock]

[citation][nom]sambadagio[/nom]For all your fps hunters, I bet you only have a screen at home with 50 or 60Hz. So just for your information, everything above 50 or 60fps is just useless... In this aspect, a PCIx 4x is actually enough... ;-)[/citation]
I think you might be surprised, a lot of people just held on to their old CRTs until the new faster LCDs came out. Also, at higher performance levels the minimum fps becomes more important than the average. When someone builds a rig to get +100 avg then the dips aren't as noticeable.

 

[redgarl]

It's been near a year since I go against X58 platform for the huge cost at no real benefits. 300$ mobo + triple channel memory + overpriced cpu + the absolute need of SLI or Crossfire with high end cards to be showing real benefits...

I am still not buying it. The LGA1156 is actually now a really affordable platform. Save the money and get better gpus.

 

[mattmock]

If you want to see the difference between 60 and 120 fps, I find it is most noticeable in first person or over-the-shoulder third person games. When walking forward or strafing, busy textures like grass, bricks or dirt blur at 60fps but remain sharp at 120fps. Objects that move across the screen judder more at 60 and appear to move more smoothly at 120. On a LCD the judder at 120 is still a bit more than on a CRT at 120 since the pixels in the screen cant quite make a full change every refresh. (This is also the cause of ghosting in stereoscopic 3D.)

 

[redgarl]

I got a 120hz HDTV hooked as my main monitor... and I can tell you, it does make a huge difference being able to go over 60Hz.

Just go check a Samsung C6300 and compare it to any screen on the market... just watching any HD movies on it boost up the quality by 75% only because of the fluidity.

 

[jsilophi]

another good read from tom's. thanks

 

[capp1688]

I know the i7 > 965 in terms of raw power, but can I expect similar results if I OC mine pII 965 to a stable 4ghz?

 

[bCubed]

Interesting...finally this debate has been put to bed.

 

[mapesdhs]

"Alien Vs. Predator appears to be limited by one of the latter variables, as its performance barely falls even when the graphics card is crippled by a x4 slot."

Instead of 'crippled', surely it's also possible that AVP's 3D
engine is simply operating in a more efficient manner than some
of the other games? Likewise, Crysis may hammer 3D hw because it's
not written very well; how can we know? It's a pity the GPUs don't
have performance counters, etc., which can feed back to the
application to show what the GE loading, texture loading, and
other parameters are at any moment in the pipeline.

FSX is an extreme example, very poor management of data, hit hard
by lower PCIe bandwidth (it keeps reloading data).

Ian.

 

[iqvl]

Since P55 has on-die PCI-E controller, the real difference is even less than the article suggested.

 

[blevsta]

How did you guys manage 4.0ghz overclock with a 160 BCLK? Did you somehow get that multiplier up to 25?

Also, you mention that a dual GPU board isn't effected by the bandwidth limitations. Can the same be said for 3d vision?

 

[terraguy]

Fantastic article. It was like TH was reading my mind, because I was just wondering about this yesterday.