Between a PCI-X and a PCI Express x4 NIC... a gigabit NIC to be more precise.
would the PCI express x4 perfome lot better ( throughput (spelling ;/ )) than a PCI-X NIC ?
NIC's dont require anymore bandwidth than PCI can supply so you'll not see any increase over either. The PCI Express may show a small Nanosecond less latency but, latency being measured in Milliseconds, you'll not notice it so save that slots for a part needing low latency.
so you think, with a pci express card i could maybe a ms or 2 if i am lucky?
i have sensitive data being transfered over a 100mbps dedicated line. i have a ping of 36ms in average with the host. if i could improve this by 1 or 2.. that would be amazing. the data has to reach asap the destination
you think it could be better with a pci express nic?
The PCI bus moves 32 bits @ 33 MHz = 1,056 Megabits per second.
We have a Linksys Gigabit NIC installed in a standard PCI slot
on an ASUS P4SGX-MX micro-ATX PGA478 motherboard, and it keeps up with the on-board GHz LAN port on our P4C800-E Deluxe i875 motherboard.
The only feature a PCI GHz NIC might be lacking
is "jumbo frames", but this may also be true
of on-board GHz LAN ports.
ya, pci-x, and x4, because it's all i got available in the Dell Workstation. pos motherboards they have...
and the PCI-X... i would like to avoid it.. just because it's not that clear...
but a pci-x nic gigabit starts @ ~155$ cnd and the only nic pci express x4 i have seen, is from Intel and it costs ~250$ cnd
so was just trying to find out if i could gain a ms or 2 with one of these cards.
You have a 100Mb/s line, so that's around 10MB/s. You don't need a gigabit nic at all, but lets just say it will make a difference for the sake of argument. You will only be using 10 percent of the bandwidth of the gigabit nic, so I really can't see it making any different what bus you put it on.
you might concider moving to fibre channel. Fiber optics would be the best bet at this point. The more dependant you are on the connection, the more important it would be to move to fiber.
it's a fiber connection, 100mbps full duplex line. from montreal to chicago.
ping of 36ms from Mtl <-> Chicago
right now, all those dell workstations are using the onboard gigabit nic atm. so basically, i was just wondering if the interface could help get a lil packet get to the town a lil faster.
and the 2 interfaces i have available is a pci-x and a pci express x4 slots
it's a fiber connection, 100mbps full duplex line. from montreal to chicago.
ping of 36ms from Mtl <-> Chicago
right now, all those dell workstations are using the onboard gigabit nic atm. so basically, i was just wondering if the interface could help get a lil packet get to the town a lil faster.
and the 2 interfaces i have available is a pci-x and a pci express x4 slots
and 1 ms would be great.
i think you're good with your current setup...
how's traffic over that line?
right, traffic is low. 3-4mb/sec
but soon, it should go over 17-19 approx.
but what matters, mainly for us, is the speed of transmission. if using a standalone card instead of onboard and using a faster interface could gain us a ms or 2. then it's great. it's the detail im trying to figure out.
right, traffic is low. 3-4mb/sec
but soon, it should go over 17-19 approx.
but what matters, mainly for us, is the speed of transmission. if using a standalone card instead of onboard and using a faster interface could gain us a ms or 2. then it's great. it's the detail im trying to figure out.
i dont think it will show any improvement, as the onboard controller actually is a pci controller. Even if you use a dedicated x4 pci-e card, it will probably be connected to the southbridge, the same bus that your onboard card, then, it will not reduce latency or something.
Like someone said, it COULD show some nanoseconds improvement, which could low your time from ~35 to lets say ~34.5.
You can contact your link company and ask them if they can reduce that latency. Is it fiber optics? I've seen 25ms latency with fiber optics links
FYI: I just read something about dual-Gigabit LAN ports
that are "double-teamed" to effect 2 Gb/s, but
I don't know much more about it and can't remember
exactly where I saw it -- maybe the ASUS website.
The PCI bus moves 32 bits @ 33 MHz = 1,056 Megabits per second.
We have a Linksys Gigabit NIC installed in a standard PCI slot
on an ASUS P4SGX-MX micro-ATX PGA478 motherboard, and it keeps up with the on-board GHz LAN port on our P4C800-E Deluxe i875 motherboard.
The only feature a PCI GHz NIC might be lacking
is "jumbo frames", but this may also be true
of on-board GHz LAN ports.
The PCI bus moves 32 bits @ 33 MHz = 532 Megabits per second. The PCI-X bus is a revision to the PCI standard that doubles the clock speed (from 66 MHz to 133 MHz) and hence the amount of data exchanged between the computer processor and peripherals. The theoretical maximum amount of data exchanged between the processor and peripherals PCI-X is 1.06 Gb/s, compared to 532 Mb/s with standard PCI.
When I say PCI its really both in that PCI-X is generally backward compatible with PCI, meaning that you can, for example, install a PCI-X card in a standard PCI slot but it will be limited to the standard PCI bus speed.
so you think, with a pci express card i could maybe a ms or 2 if i am lucky?
i have sensitive data being transfered over a 100mbps dedicated line. i have a ping of 36ms in average with the host. if i could improve this by 1 or 2.. that would be amazing. the data has to reach asap the destination
you think it could be better with a pci express nic?
The PCI Express 4X wouldn't even make 1ms differance as the difference in PCI-X and PCI Express is in nanoseconds. The PCI Express only has a few nanoseconds advantage and the 36ms is made up almost completely of the distance form your NIC to where the NIC your sending the data.
Now to lower your 36ms you could invest in fiber optics as copper twisted pair line from your NIC to your router, switcher, or DSL box could reduce it as much as 2 or 3ms. The whole package will set you back about $500 so I advice against it but if money is no object it will lower your access time.
If your connection from your PC to the data's destination is fiber optic line then you'll have to find a way of sending data faster than light.
> The PCI bus moves 32 bits @ 33 MHz = 532 Megabits per second.
I think your arithmetic is off:
32 bits @ 33 MHz = 1,056 megabits per second
divided by 8 bits per byte = 133 megabytes per second
(got a calculator? roughly, 30x30=900
so 32x33 would be larger than 900)
This upper limit to the PCI bus is the same number
as ATA/133 = 133 MB/second, and that was one
of the main reasons for the initial speed of SATA-I
of ATA/150 = 150 MB/second.
You are correct about the different flavors
of PCI-X:
one increased the signal rate to 66 MHz, and the second increased the signal rate to 133 MHz.
Also, the PCI-X bus transmits 64 bits per cycle,
instead of 32.
In order to transmit 64 bits per cycle, a larger slot was needed, hence the term
"PCI-Xtended" ). See the ASUS P5WDG2-WS
motherboard, for an example of both PCI
and PCI-X slots:
Thus, a PCI-X adapter will not fit into a standard PCI slot, but standard PCI adapters
will fit into a standard PCI-X slot. Whether
the latter will work correctly depends on
several things, like chipset support etc.
p.s. I wouldn't trust anything I find at Wikipedia,
particularly if it's technical data you are seeking.
The higher bandwidth quoted by Intel
(1.58 megabytes per second for PCI)
results from factoring in the signals used
for parity and synchronization control,
because PCI is a parallel bus architecture.
133 MB/second is EFFECTIVE data bandwidth.
Also, use Google http://www.google.com to locate Intel's .pdf "White Paper" on the subject:
Each PCI-Express lane consists of two signal pairs
in a dual-simplex channel oscillating at 2.5 GHz.
Thus, raw bandwidth in one direction is 2.5 Gb/s,
and 5.0 Gb/s bi-directionally, because each lane
is bi-directional ("dual-simplex" ). See Intel's
illustrations above.
When computing bandwidth on Serial ATA devices,
use 10 bits per byte, because this serial protocol adds a start bit and a stop bit, just like serial modem
communications. That's why Intel's White Paper
recommends deducting 20% to arrive at
effective data bandwidth.
Thus, a PCI-X adapter will not fit into a standard PCI slot, but standard PCI adapters
will fit into a standard PCI-X slot. Whether
the latter will work correctly depends on
several things, like chipset support etc.
p.s. I wouldn't trust anything I find at Wikipedia,
particularly if it's technical data you are seeking.
> The PCI bus moves 32 bits @ 33 MHz = 532 Megabits per second.
I think your arithmetic is off:
32 bits @ 33 MHz = 1,056 megabits per second
divided by 8 bits per byte = 133 megabytes per second
(got a calculator? roughly, 30x30=900
so 32x33 would be larger than 900)
This upper limit to the PCI bus is the same number
as ATA/133 = 133 MB/second, and that was one
of the main reasons for the initial speed of SATA-I
of ATA/150 = 150 MB/second.
You are correct about the different flavors
of PCI-X:
one increased the signal rate to 66 MHz, and the second increased the signal rate to 133 MHz.
Also, the PCI-X bus transmits 64 bits per cycle,
instead of 32.
In order to transmit 64 bits per cycle, a larger slot was needed, hence the term
"PCI-Xtended" ). See the ASUS P5WDG2-WS
motherboard, for an example of both PCI
and PCI-X slots:
Thus, a PCI-X adapter will not fit into a standard PCI slot, but standard PCI adapters
will fit into a standard PCI-X slot. Whether
the latter will work correctly depends on
several things, like chipset support etc.
p.s. I wouldn't trust anything I find at Wikipedia,
particularly if it's technical data you are seeking.
In theory your correct but In traditional PCI protocol, devices often add extra clock cycles, or wait states, into their transactions to hold the bus if the target PCI device is not ready to proceed with the transaction. Simply Multipling the 2 is nothing like you'll see in the real world. 532 Megabits is a accurate reflections of real world transfer.
PCI-X eliminates use of wait states, except for initial target latency.
