Obvious Question: Why is AGP faster than PCI in results?

Ok .. i'm just starting to look at Dave's X800 423 Review but having seen some other reviews and seeing that in most cases the AGP8x is still faster than PCI-E. Why is this? I would've have thought PCI-E should've been faster than AGP or at least equal. But alot of benchmarks I saw had the PCI-E slower(about to check Dave's).

US
 
On what particular results are you basing your statement on? Have you accounted for the fact that the AGP X800 XT PE is clocked higher than the PCI-E X800 XT?
 
oops ye .. I think I must have missed that on the other reviews. After reading what Dave has done on his specs I kinda realised that. :oops:

I intend to go check those reviews again though .. just to be sure.

US
 
Sorry, I’m afraid to say that my initial PCIe reviews are pretty horrific in terms of comparing the actual differences between PCI and AGP in a gaming situation due to the platform differences – it would be nice if we were able to get 3.4 or 3.6GHz Prescotts for both Socket 478 and LGA 775, but we don’t command similar sway with Intel as some, such as Anand or Toms do (although if anyone reading could assist that situation that would be rather nice). However, even if that were the situation we’re still not able to get 100% like for like platforms so there will always be some difference – we’ll have to see if the AGP and PCIe capable northbridge from VIA is released with both AGP and PEG x16 slots on a motherboard to be able to get try like for like equivalence (and even then we are working on the assumption that these will both be good implementations).

In general terms the AGP X800 XT PE clocked to XT speeds (500/500) is more or less a little behind the PCIe XT in all cases, but this is largely due to the CPU differences – even when things begin to get fill-rate constrained the PCIe board appears to be faster still (but remember, the AGP board is a PE running at XT speeds, so its memory timing may not be set up for an XT configuration). We can certainly see that when bus bandwidth is an issue the PCIe boards have a significant advantage.
 
Ok I've found the review.

http://www.driverheaven.net/reviews/atifreeforall/

DriverHeaven said:
The cards im playing with today are:
X800XT PE AGP 520/560
X800XT PE PCIe 520/560

X800XT PCIe 500/500

Yet the results have the X800XT PE AGP 520/560 outclassing the X800XT PE PCIe 520/560 in the Halo Benchmark and the Serious Sam 2 Benchmark.

Even in the Farcry benchmark the AGP version outguns the PCI-E version at high res. In UT2004 though the cards run about equal with the AGP just topping the PCI-E at high Res again. In RTCW the AGP wins most test vs the PCI-E.

*editted* They also using the same Mobo by the looks of it.

US
 
K thx Dave. So when can we expect PCI-E to really take over from AGP8x? I would've thought that with all the hype of PCI-E that it would've had better results. Isn't it one of the reasons why most manufacturers are said to be going the PCI-E way?

I know we'll start seeing differences within the next year, with textures and stuff requiring more memory and bandwidth(also media should take advantage) .. but does it really make that huge of a difference for games?

US
 
A couple of things on that one.

I've just checked with ATI and the board they use in that review isn't actually a PCIe X800 XT Platinum Edition, its an overclocked XT.

The second on, and an issue in general, is that the 925X platform is using DDR2 533 memory, which can neither match the bandwidth of 875 and DDR400 and, is suspect, the latency issues of DDR2 are bad for the type of transfers 3D ops require.
 
Ahh.. ok thx Dave. :)

Guess we'll have to wait for proper chipsets to come out of Intel or VIA.

BTW. Dave. Have you asked AMD for a PCI-E mobo? I'm rather looking to get a AMD64 and would like a PCI-E mobo and Graphics Card review. I'm was just rather dissapointed at what I saw though.

US
 
AMD don't make chipsets so its up to the chipset vendors when they decide to move to PCIe. Intel were first off the blocks, the others will follow and thats when they will be coming out with their AMD versions.

As for benefits - I suspect we'll see multimedia implementations and media center type devices take advantage of PCIe before games will, as the benefits are non-obvious. There will be some cases where the texture demands will see some benefits, especially if lots of offscreen buffers / render to textures are used and the user is selecting high res and FSAA, but I suspect that dev's will still code to the contemporary levels of onboard RAM.
 
Do you have any idea, how much agp drivers of the motherboards' chipsets have improved over the years? Then compare it to how much time has there been to improve pci-e drivers... then ask yourself why isn't agp more faster than its now.

Remember when agp was new slot and pci (non express) voodoo cards were in the market? Agp seemed like oem only solution, about which gamer didn't need to care, 3dfx those voodoo 2 cards were years ahead of for example the poor ati rage agp solutions at the time...

Remember how much faster is agp 8x compared to agp 4x? If not, here's my old benchmark: (edit: aquamark3)

radeon 9700, p4 2200:

agp 4x: 31.874
agp 8x: 31.929

the difference fits nicely in the margin of error, with another run the results might even reverse.

Why should pci-e x16 then bring any considerable increase? Why should it in fact bring anything other than latency, bugs and trouble that any new technology is bound to bring?

overall: The interface does not matter at all. The card's chipset and memory does.
 
I would wager that the serial point to point nature is better for the types of operations required from 3D than the parallel, multi-date per cycle, nature of AGP, so I wouldn't have expected anything in the ay of extra latencies or anything from the bus itself.

What is the paradigm shift is the upstream bandwidth and that what will take a little while for people to work out what they can do with it in non-multimedia apps.
 
I wager pci-e fails to impress much in the same way as serial ata... no big deal really. It shouldn't fail anywhere near as radically as rdram though. Ultimately the problem lies within the system ram that will always be too slow compared to the card's onboard ram. Therefore the interface to that too slow memory doesn't matter. Does the upstream matter now? No.
later... remains to be seen.

So what's all the fuss about then?

Marketting loves getting rid of old stuff and getting new stuff, no matter if there's performance differences or not. Away from legacy, so to speak!
 
I think this is more of a software issue than a hardware one.

Remember when AGP first came out? PCI cards were faster and we all wondered why. But once they tweaked the software AGP did what it was supposed to.

I think its just a matter of PCIe being new and un-optimised and AGP having been around awhile.
 
AMD don't make chipsets so its up to the chipset vendors when they decide to move to PCIe. Intel were first off the blocks, the others will follow and thats when they will be coming out with their AMD versions.

Ummm, Yes they do...

I wager pci-e fails to impress much in the same way as serial ata... no big deal really.

Or the person simply had unrealistic expectations? SATA impresses me if for anything the more simplified wiring/connector topology. The bandwidth gains don't really come into play until you start RAID'ing and there's still benefits to be seen when more chipsets and drives support native command queuing...
 
archie4oz said:
Ummm, Yes they do...

AMD and chipsets? Where...?

Or the person simply had unrealistic expectations? SATA impresses me if for anything the more simplified wiring/connector topology.

And the same applies to PCIe - look at the size of a x1 connector to PCI Conventional, and it has twice the bandwidth.

The bandwidth gains don't really come into play until you start RAID'ing and there's still benefits to be seen when more chipsets and drives support native command queuing...

And remember, most of the drives that have been seen until recently have in fact been bridged ATA drivers, rather than native SATA.

Anyway, the point about PCIe is a greater level of freeing the traditional boundries we've had in the PC more and ultimately that will be used to a greater extent. The more the PC ends up in a PVR/HTPC role the more this bandwidth will come in handy.
 
Dave they do make chipsets .


The dual athlon xp board is amd and at the launch of the old solt athlons it was an amd chipset .

But they generaly don't anymore unless its something they really want out right away .
 
AMD have always made chipsets, mainly to show case the new processors before Taiwan gets up to speed, though in the past the dual AMD 760 chipset was a mighty thing used by MSI, Tyan etc.

Remember the Opteron K8 chipset that had the musical notes of the Intel jingle on the motherboard on review samples ? That is still the best in joke that a major manufacturer has had at the expense of it's competitor in my view. If I remember correctly nobody really thought about it until some muscian/ geek looked at it and automatically played it over in their head.

To be honest the Intel PCIe chipset has no extra speed over the 875, whether the video card does or does not is a mute point because you will get better performance going AMD64 and AGP for the forseeable future IMO.
 
Well, they aren't making chipsets for Athlon 64 platform. With the memory channel in the CPU they can control the quality of the interface and the RAM types used, regardless of the rest of the chipset, which is one of the main performance elements - evidently they CPU's also have mulitprocessor control embedded in the processors as well, which also alleviates the need for a chipset.
 
Back
Top