Interpreting Gigabyte LGA775 mobo specs

  • Thread starter Thread starter docsavage20
  • Start date Start date
D

docsavage20

I'm looking for a Gigabyte LGA775 board that has among other things a PCIe 2.0 or higher slot for the video card.

I've been looking at specs on their boards and notice there are some that in the overview specify PCIe 2.0 but some for example this GA-EP43T-UD3L

http://www.gigabyte.com/products/product-page.aspx?pid=3597&dl=#ov

that say "Ultimate graphics performance with PCI-E x16 interface" but doesn't specify PCIe 2.0 but in the specifications area it says:

1.1 x PCI Express x16 slot, running at x16 (The PCI Express x16 slot conforms to PCI Express 2.0 standard.)

Is this the same as saying it has a PCIe 2.0 slot? This particular board supports DDR3 Ram, is there such a thing as a board with DDR3 that doesn't have a PCIe 2.0 slot? There are various like this where 2.0 isn't explicitly stated in the overview. Wondering if it's just an editorial oversight.

Thanks
 
I'm looking for a Gigabyte LGA775 board that has among other things a
PCIe 2.0 or higher slot for the video card.

I've been looking at specs on their boards and notice there are some that
in the overview specify PCIe 2.0 but some for example this GA-EP43T-UD3L

http://www.gigabyte.com/products/product-page.aspx?pid=3597&dl=#ov

that say "Ultimate graphics performance with PCI-E x16 interface" but
doesn't specify PCIe 2.0 but in the specifications area it says:

1.1 x PCI Express x16 slot, running at x16 (The PCI Express x16 slot
conforms to PCI Express 2.0 standard.)

Is this the same as saying it has a PCIe 2.0 slot? This particular board
supports DDR3 Ram, is there such a thing as a board with DDR3 that doesn't
have a PCIe 2.0 slot? There are various like this where 2.0 isn't explicitly
stated in the overview. Wondering if it's just an editorial oversight.

Thanks
Click to expand...

First, some irrelevant trivia.

On modern systems, there is no Northbridge. The processor has PCI Express
directly on it, the processor also has a memory interface on it. The processor
in that case, determines some of the characteristics of the video slot (due
to the video slot interface being on the processor itself). On a modern
system, some processors support PCI Express 2.0, and some support 3.0.
Not that this is important, for all but the most extreme cases (probably
not a factor for gamer video cards). Maybe a RAMDisk card would care
about this.

*******

The LGA775 architecture, is "old school". There is a Northbridge (P43).
The Northbridge has the memory interface, the Northbridge has PCI Express x16.
And it's the characteristics of P43 you're interested in. Doesn't matter
about the processor, as all it does is "process". No doodads hanging off
the sides, like more modern systems. My system looks like this one
(LGA775 architecture).

CPU
|
| quad rate FSB
|
Memory_Channels ---- P43 ---- x16 video slot
|
| DMI (limited bandwidth)
|
SATA --- Southbridge --- PCI Express x1
|
PCI slots

There is also some irrelevant trivia about the Southbridge, but
we won't get into that.

*******

You missed a space character in the first line here. That's not "1.1".
It is "bullet point 1." followed by "one PCI Express x16 slot..." :-)

http://www.gigabyte.com/products/product-page.aspx?pid=3597&dl=#ov

"1. 1 x PCI Express x16 slot, running at x16
(The PCI Express x16 slot conforms to PCI Express 2.0 standard.)

2. 4 x PCI Express x1 slots <--- these are likely to be 1.1 mode
"

The reference to 2.0 standard, means 500MB/sec per lane in one direction,
or 8GB/sec for the whole x16 interface. A quick analysis says that
shouldn't be a problem for the chipset - it's in the right
ballpark bandwidth-wise.

Note that, in practice, you cannot get 8GB/sec. The chipset uses
"tiny buffers" for PCI Express packets. On Intel, this typically
cuts the practical bandwidth in half. More details here.
Half of 8GB/sec roughly, gives you 4GB/sec practical transfer
rate, and anything over 1GB/sec is good enough (doesn't harm
video performance too much). (Tomshardware did tests years
ago, and back then, anything over 1GB/sec was plenty.)

http://www.plxtech.com/files/pdf/technical/expresslane/Choosing_PCIe_Packet_Payload_Size.pdf

Paul
 
Note that, in practice, you cannot get 8GB/sec. The chipset uses

"tiny buffers" for PCI Express packets. On Intel, this typically

cuts the practical bandwidth in half. More details here.

Half of 8GB/sec roughly, gives you 4GB/sec practical transfer

rate, and anything over 1GB/sec is good enough (doesn't harm

video performance too much). (Tomshardware did tests years

ago, and back then, anything over 1GB/sec was plenty.)
Click to expand...


What I want to ensure is tat the slot won't rob any performance from a video card for gaming purposes. I.e. - if I get a faster card than what I have - a GTX 460 Cyclone 1gb that I'll see all the performance the card is capable of with some room left over for OC'ing.

From what I've been reading a lot of the hoopla about the PCIe generations is marketing hype, that many 2.0 labeled cards don't actually perform any better in a 2.0 slot than they do in a 1.0 slot. I've read that the GTX 460 is such a card. It's labeled 2.0 but supposedly it doesn't even saturate a 1.0 slot. I wonder how one finds out whether a given card actually would beimpeded by a 1.0 slot.

I assume the same is true going from 2.0 to 3.0.
 
What I want to ensure is tat the slot won't rob any performance from a video card for gaming purposes. I.e. - if I get a faster card than what I have - a GTX 460 Cyclone 1gb that I'll see all the performance the card is capable of with some room left over for OC'ing.

From what I've been reading a lot of the hoopla about the PCIe generations is marketing hype, that many 2.0 labeled cards don't actually perform any better in a 2.0 slot than they do in a 1.0 slot. I've read that the GTX 460 is such a card. It's labeled 2.0 but supposedly it doesn't even saturate a 1.0 slot. I wonder how one finds out whether a given card actually would be impeded by a 1.0 slot.

I assume the same is true going from 2.0 to 3.0.
Click to expand...

I don't think this test measures bandwidth.

Instead, they did a benchmark comparison between PCI Express 2
and 3 operating modes (500MB/sec per lane, versus 1GB/sec per lane).

http://www.anandtech.com/show/5264/sandy-bridge-e-x79-pcie-30-it-works

And the claim is, it doesn't help for gaming. And helps a bit
for the GPGPU benchmark they were using.

There are some gaming results here. For a GTX 680.

http://www.hardocp.com/article/2012/07/18/pci_express_20_vs_30_gpu_gaming_performance_review/2

http://www.techpowerup.com/reviews/Intel/Ivy_Bridge_PCI-Express_Scaling/24.html

There is also a PCIe-Speed-Test, but I can't be sure what
it's testing. The danger with tests like this, is when
they draw their results from some cache, rather than
traveling across a system bus. At least one line in
the test is anomalous (seems too high).

http://www.xtremesystems.org/forums...peed-Test-v0.1-CPU-GPU-GPU-CPU-bandwidth-test

These are the kind of things, you want a test instrument to
verify the software test is consistent with a hardware
capture.

Here, a NIC card company does some testing. This is not fast
enough for the purpose of testing. You really need a card
dedicated to such testing. You can't get that from FPGAs,
so it would take some other kind of chip (like a test mode
in a PLX chip perhaps).

https://www.myricom.com/software/my...e-dma-performance-on-my-pcie-motherboard.html

Paul
 
Back
Top