gigabyte pci ethernet card = impossible?

[capybara]

if a pci slot is rated 33 mhz, how can that work with a 100 mhz ethernet card?
and of course a gigabyte ethernet card would be still limited by the pci slot speed.
right?

 

[Lynx516]

Well with a GIGABIT NIC the maximun throughput is 125mpbs each way. (which is the speed of the PCI bus). However due to the nature of hte PCI bus. The Card is efffectivly limited to 75mbps. also the frequency on the line is 100mhz not the speed of the card. It will have its own clock generator to provide the signal along the cable.

 

[CTho9305]

yes. that is why there are 66mhz and 64 bit PCI slots, and new standards under development. I don't know why you think the ethernet card has to talk at 100mhz though.
edit: I'm replying to the gigaBYTE question, not the assumed gigabit question.

 

[Lynx516]

There is no Gigabyte Ethernet standard. There is ether 1GBit or 10Gbit. the 1GBit needs 64bit PCI bus to operate at full speed anyway

 

[Mday]

33MHz, 32 bits ~ 1060 Mbits per second. roughly 1 Gbit max. obviously that is never going to happen for a single pci device.

dont confuse bits and bytes, and bits and Hz. and dont confuse m and M. one stands for 1/1000, and the other stands for 1000000 (in decimal)

 

[yak8998]

the PCI bus (except on new mobos being released) also contains all your drive controllers, correct?

 

[Lynx516]

For a long time it did but pretty much any Chipset from the last 2 years has it off loaded. Raid has just been ofloaded by intels new chipsets (dont quote me on that). Before if you wanted to get the 1Gigabit max out of the Gigabit card it was impossible. This is due to 2 things 1 you have to be able to shift 125mbps out of your HDDs for this you need RAID 0. For this untill recently you had to have it on the PCI bus. THis creates another probelm. ON the PCI bus to transfer from 1 device to another you cannot do a direct link you have to go via the bus controller to the data is effectivly sent twice halving the maximum bandwidth. Though if you had a 33Mhz/32bit PCI bus connected to a 66Mhz/64bit bus you could get the max through put aslong as the raid card was either on a different bus to the Gigabit card r they are both on the 66mhz bus

 

[Peter]

Standard PCI bus does have peer-to-peer transfer capabilities from one PCI device to another alright. However this isn't useful for disk-to-LAN, since you need to pack the raw file data into ethernet packets first.
So no matter what the system architecture is like, the HDD controller feeds the data into main system RAM, the CPU makes ethernet packets out of it, and then makes the LAN unit fetch these for transfer over the network connection.

 

[tdowning]

FWIW, I do believe that I have seen something that lead me to believe that there are more than one PCI Bus in Compaq Servers. In fact I believe that there are usually three, in modern ones, (Not counting ones with PCI-X, I haven't seen any of those) a 64 bit 66 Mhz, usually 2x, one 33Mhz, 32 bit and all the rest 64 bit 33 Mhz.

I think there may be a fourth for internal SCSI drives, as the server in question had an add in board, that looked more like a Mini PCI board than anything else, and added RAID functionality to the onboard SCSI controllers, (wether that was a fourth PCI, or something else I couldn't say though.

 

[Tab]

What is the speed of the PCI Bus anyway? 133 isnt it?

 

[Lynx516]

Peter the PCI bus does NOT have a driect link to each devices. Everything has to go through the bus controller.

tdowning

I doubt they have 4 PCI busses.

The chipsets tht they are mostlikly to have (IF thses are x86) will be teh Serverworks chipsets or an INtel chipset.

These chipsets have multiple PCI-X channels and 1 66mhz,64bit PCI bus and a 33mhz,64bit PCI bus .

Most server boards not using these expencive chipsets will have 1 66,64bit PCI bus and a 33mhz,64bit PCI bus.

Where do you get the idea that there are 3 if not 4 PCI busses?


Tabb the speed of the PCI bus is either 33mhz (in desktops) and 66mhz (workstations servers)

 

[Peter]

Lynx, I suggest you read up a bit on how PCI works.

Every (bus master capable) PCI device pumps its data where they belong, without involving the bus controller in the actual transfer. All devices on a PCI bus are created equal, including the host bridge chip, with the sole exception being that the host bridge chip does the arbitration on who gets to own the bus next. But the actual transfer is always under control of the transfer initiating device - just as well as taking the data off the bus is completely at the decision of the target device. Every PCI target has been programmed by BIOS to decode to a certain range of memory and/or I/O addresses, and when it sees a transfer address it "owns" it claims the cycle and takes the data. The host bridge is no exception here either, as a target it's claiming cycles aimed at the memory locations of the main system RAM.

E.g. when your TV card feeds the video stream onto the graphics card's overlay buffer, and they're both on the same PCI bus, then it's a direct peer-to-peer transfer. Even when the TV card is PCI and the graphics card is AGP, it's a peer-to-peer transfer; the difference being that the chipset's integrated PCI-to-AGP bus bridge forwards the cycles from PCI to AGP in behalf of the AGP card behind it. Again, the transfer is not going through the host system bridge and is not going through system RAM either.

 

[Peter]

tdowning, PCI bus count is usually one wimpy 32-bit 33 MHz on commodity chipsets, two on entry-level server chipsets like the lower ServerWorks models or AMD's 760MPX (typically one slow and one fast 64-bit 66 MHz).
The PCI bus count doesn't get massive until the chipsets do - e.g. with Intel's 750x series you can pile up to three P64H2 bus bridges on, for a maximum of seven fully concurrent PCI busses (six fast PCI-X, one legacy 32-bit 33 MHz).