is pcie2.0 worth it

D

drjonas0

Weaksauce
Joined
Sep 7, 2004
Messages
118
i was just about to buy my new machine and realized that the 4870 is pcie 2.0 x16 and the motherboard i was going to buy doesnt have 2.0 I know that that it is backwards compatible but wasnt sure if the computer was going to bottleneck because its not 2.0
 
It's not a big deal at all for the high end cards. Surprisingly, it's only the low end ATI cards that benefit from PCI-E 2.0.
 
no, not enough to care at any rate. I have yet to find a really compelling reason to have PCIe 2 let alone the new version coming out.

If you were crossfiring the cards that might be different as there seems to be a small performance hit if you run 16X by 4X, but 8X by 8X seems to be all right. I am hoping H does a review of this and settles the argument.
 
Valset said:
actually you just supported that it doesn't, there isn't much difference between PCIe 16 and 8, let alone PCIe 2.0 and PCIe 1.1. in theory PCIe 2.0 8X += PCIe 1.1 16X. I doubt it works that neat though.
'
Click to expand...
well the CoD 4 benchmark for the 8800 shows a pretty significant difference
 
cannondale06 said:
well the CoD 4 benchmark for the 8800 shows a pretty significant difference
Click to expand...

actually the argument still holds even with COD4, the worst case.
Scaling as %
16X PCIe 1.1 100%
8x PCIe 1.1 83.2%
4X PCIe 1.1 56.3%
1X PCIe 1.1 25.3%

If you look at this way halving the bandwidth of the PCIe 1.1 gives you 83% of the performance, a quarter of the bandwidth gives you 56% of the performance, what do you think 200% of the bandwidth is going to do for you?

I can't see it being worth a new motherboard or a higher dollar on for that matter. If its there, great, should give you a few % performance increase, being able to have 4 8X lanes is nice if you need them. outside of that...
 
Pff... ISA is where it's at.

But in all seriousness, you can have some pretty good benefits by using PCI-E 2.0 over 1.0, depending on your situation.
 
Valset said:
actually the argument still holds even with COD4, the worst case.
Scaling as %
16X PCIe 1.1 100%
8x PCIe 1.1 83.2%
4X PCIe 1.1 56.3%
1X PCIe 1.1 25.3%

If you look at this way halving the bandwidth of the PCIe 1.1 gives you 83% of the performance, a quarter of the bandwidth gives you 56% of the performance, what do you think 200% of the bandwidth is going to do for you?
Click to expand...

Guys... it was CoD2 and using PCIe 1.0a, not 1.1.
FFS, that article was from March 27, 2007
 
Valset said:
actually the argument still holds even with COD4, the worst case.
Scaling as %
16X PCIe 1.1 100%
8x PCIe 1.1 83.2%
4X PCIe 1.1 56.3%
1X PCIe 1.1 25.3%

If you look at this way halving the bandwidth of the PCIe 1.1 gives you 83% of the performance, a quarter of the bandwidth gives you 56% of the performance, what do you think 200% of the bandwidth is going to do for you?
Click to expand...
I understand that but losing 17% just because you didnt have a pci-e 16 slot is still fairly significant to me. its odd too that only certain games would make a difference.
 
cannondale06 said:
I understand that but losing 17% just because you didnt have a pci-e 16 slot is still fairly significant to me. its odd too that only certain games would make a difference.
Click to expand...

the point was not that you lost performance but how the scaling worked, after a point adding bandwidth is important, put it this way, doubling the bandwidth beyond the point that the card is fully feed produces a point diminishing returns. for every increment of speed you go up you get a smaller and smaller gain.

don't get me wrong, everything else being equal get PCIe 2.0. if nothing else its better future proofing.
 
I think it's worth it, yet I am going to run 2 4850's. I still find it all worth it. That was the whole reason I purchased my A8N32-SLI was so I could have 16x, 16x. Though I don't believe it was PCX 2.0
 
for SLI or crossfire I would def not run less then 16X PCIe or 8XPCIe 2.0. Again I wish H would do a review on this, would be interesting to see
 
AMD says:

Source: http://www.techtree.com/India/News/EXCLUSIVE_--_An_Interview_with_AMD/551-90984-579.html
TT: In CrossFire, most of the motherboards, almost 95 per cent of them, support 8X mode instead of 16X each (PCI-E bus speed). Is 8X good enough?

R: If I answer this from a forward looking viewpoint -- yes more bandwidth is always good, but from an existing games stand point, it is good enough, that's what we find.


TT: Good enough even for the 4-series?

R: Yes, because our drivers and the games, at least the last generation games, they've been optimized to avoid traffic between GPUs. But my prediction is that moving forward it will be important that there's more bandwidth between GPUs. So if you want to use this configuration for a long time, then you'll have to go for 16X.
Click to expand...
 
To answer those who want an explanation on my earlier post

http://www.hardforum.com/showpost.php?p=1032759148&postcount=2

It's because in the ATI drivers, only the lowest end cards (3450/3470 to be specific) are optimized to take advantage of the extra bandwidth of PCI-E 2.0. The difference is pronounced when you run tests that require a lot of framebuffer, and the card has slow memory or not enough memory. Even then, not all applications show an improvement. 3DMark06 with 4xAA is one of the tests that benefit from this. PCI-E 2.0's extra bandwidth is controlled via the driver, not the motherboard. The driver is able to dynamically switch on and off the extra bandwidth (2.5 GT/s vs 5 GT/s), and how the driver uses this extra bandwidth depends on the driver itself. So far, only the 3400-series will have a tangible performance boost with PCI-E 2.0.

If you want actual performance data to confirm this, run it yourself.
 
lol_wut said:
To answer those who want an explanation on my earlier post

http://www.hardforum.com/showpost.php?p=1032759148&postcount=2

It's because in the ATI drivers, only the lowest end cards (3450/3470 to be specific) are optimized to take advantage of the extra bandwidth of PCI-E 2.0. The difference is pronounced when you run tests that require a lot of framebuffer, and the card has slow memory or not enough memory. Even then, not all applications show an improvement. 3DMark06 with 4xAA is one of the tests that benefit from this. PCI-E 2.0's extra bandwidth is controlled via the driver, not the motherboard. The driver is able to dynamically switch on and off the extra bandwidth (2.5 GT/s vs 5 GT/s), and how the driver uses this extra bandwidth depends on the driver itself. So far, only the 3400-series will have a tangible performance boost with PCI-E 2.0.

If you want actual performance data to confirm this, run it yourself.
Click to expand...
why would they make low end cards take advantage of bandwidth that most low card users dont even have in the first place?? I mean most of the users of those cards dont even have pci-e 2.0 boards. :confused:
 
Don't ask me why. I'm not the one coding the drivers. To go Beyond3D and ask Dave Baumann if you want more technical details on why this is. This is as far as I'm willing to go in terms of details.
 
Here's some motherboard testing from Legion Hardware.

Link

By simply firing up 3Dmark Vantage we found that the new X48 chipset was able to provide 17% more performance than the older P35 chipset when using two Radeon HD 4850 graphics cards in Crossfire mode. More over, the X48 was less than a percent faster than the P35, which was just a percent faster than the X38 chipset. These numbers would not exactly have X38 owners rushing out to pick up an Intel P45 or X48 motherboard.

However, as we suspected this did not turn out to be the full story, as we began testing with some real games. When testing with games such as Crysis, we quickly discovered that the X48 chipset was able to deliver 10% more performance when compared to the X38, while it was slightly faster than the P45. While this was impressive, Crossfire is not all that well suited to Crysis, so we continued on and hit Devil May Cry 4, which works a treat.

Now when testing with the new hit title Devil May Cry 4 we found that a typical P45 setup is going to be 36% faster than a P35 system at 1920x1200 when running Crossfire Radeon HD 4850 graphics cards for example. The X48 was also roughly 6% faster than the P45 chipset, and I would suggest that this is a best case scenario for the high-end X48 chipset. That said, this did make the X48 around 15% faster than the older X38 chipset.

For the most part the X48 is just 5% faster than the cheaper P45 chipset, while this chipset held roughly the same margin over the older X38 chipset. Therefore as we see it, X38 owners really would need to purchase an X48 motherboard to see any real performance gains, while the P45 is a massive step up from the P35 chipset when talking Crossfire performance.

Right now it is possible to purchase a decent ASUS P45 motherboard, such as the P5Q Pro for example, at just $150 US. Pairing this board with Crossfire Radeon HD 4850 graphics cards would cost around $550 US, making for an extremely lethal combo! The cheapest ASUS X48 board on the other hand, is the P5E Deluxe at $220 US, and we are not sure that at this price it is worth the 5% Crossfire performance gain. Still, if you are looking to get the most out of your Crossfire setup, then the X48 is the weapon of choice.
Click to expand...
 
Comparing Crossfire on X48 to P35 is hardly a fair comparison. Even without any benchmarking, you know that P35 sucks for Crossfire since it runs at 16x/4x.

X48 being that much faster than X38 was a surprise though. It doesn't make a whole lot of sense, since they're the same chip.
 
And ETQW is a good scenario because it actually scales fairly well. In most other tests, the difference should be even smaller.
 
cannondale06 said:
why would they make low end cards take advantage of bandwidth that most low card users dont even have in the first place?? I mean most of the users of those cards dont even have pci-e 2.0 boards. :confused:
Click to expand...

not sure but there might be a grain of truth here, on my last system I had an 6800 256mb (the 4 pipeline version) and I did see some gain in overclocking the PCIe bus. but only on bioshock. I am wondering if low memory cards that have to use system memory for textures would be affected. so on the low end cards they are relying on the PCIe speed as part of their bandwidth. if that was true then there would be a big advantage to PCIe 2.0
 
I may test this tonight if I have time, should be easy enough, I have a PCIe 2.0 board but I have an 8800GTX. I can run a benchmark before and after the OC. Quake 4 would be a good one maybe
 
cannondale06 said:
why would they make low end cards take advantage of bandwidth that most low card users dont even have in the first place?? I mean most of the users of those cards dont even have pci-e 2.0 boards. :confused:
Click to expand...

Low end cards typically have less integrated memory; this means they will:
1) use system memory via some sort of memory mapping
2) swap data in and out more frequently because of less storage capacity.

Both of these reasons are obviously going to increase traffic over the pcie link compared to a high end card with a lot of memory.
 
Valset said:
I may test this tonight if I have time, should be easy enough, I have a PCIe 2.0 board but I have an 8800GTX. I can run a benchmark before and after the OC. Quake 4 would be a good one maybe
Click to expand...
No, Quake 4 is too CPU-bound. Run something more recent like HL2 Episode 2 with AA/AF. Just don't run Crysis because it doesn't scale at all no matter what you do.
 
@ 1920 by 1200 with eye candy it shouldn't be but I don't have HL2, I do have bioshock but no real way to benchmark it. maybe fraps but it tears something terrible without vsync
 
WhiteZero said:
Guy, those are benches comparing PCIe rates (1x - 16x), not PCIe versions...
In fact, the mobo they used, the MSI 975X Platinum, is PCIe SPEC V1.0a:
http://www.msicomputer.com/product/p_spec.asp?model=975X_Platinum_V.2
Not even 1.1... lol
Click to expand...

Doesn't matter, 1.0a and 1.1 have the same bandwidth per lane.

@OP: http://www.tomshardware.com/reviews/pci-express-2-0,1915-10.html

That shows PCI-E 2.0 lane scaling on a 9800GX2 (which is likely to have extra overhead due to the SLI). PCI-E 1.1 x16 ~= PCI-E 2.0 x8. As you can see, in most games it won't matter, in CoD4 it is ~7% FPS loss, and in MS Flight Sim X (the worst case), it is a ~15% FPS loss.

That said, get a 2.0 board anyway, at least in an attempt to future proof your rig (and because it isn't really any more expensive)
 
lol_wut said:
Comparing Crossfire on X48 to P35 is hardly a fair comparison. Even without any benchmarking, you know that P35 sucks for Crossfire since it runs at 16x/4x.

X48 being that much faster than X38 was a surprise though. It doesn't make a whole lot of sense, since they're the same chip.
Click to expand...

I was surprised about the x38/x48 too... when I purchased mine the only difference was supposedly better binned chips and cooling. Can't say I regret it, though :)

Dazed66 said:
on a slightly OT side note, thank you for that link, was an interesting read
Click to expand...

No prob, I've been running into a bunch of random articles lately while researching the 4 series :p
 
Wait a sec - so does this matter for single card configurations then? I don't EVER plan on buying dual setups.
 
You must log in or register to reply here.
Back
Top