VOGONS


pci-x speed

Topic actions

First post, by emosun

User metadata
Rank Oldbie
Rank
Oldbie

quick question that I'm having a hard time finding a simplified answer on. How much faster is pci-x over pci? specifically pci-x 1.0 which (i think) is what my server uses.

I tried reading the pci-x wiki page but it seems to say the maximum speed is either 1gb or 2-4gb in two different sections so it's somewhat confusing what it's speed really is. it seems to indicate the speed of pci-x 2.0 but not 1.0.

Reply 1 of 28, by Ampera

User metadata
Rank Oldbie
Rank
Oldbie

It ultimately depends on the card and slot. All PCI-X cards/slots are 64-bit, meaning for each clock cycle of the bus, 64 bits of data are transmitted/received.

Then there is the clock speed. Most slots for PCI-X 1.0 are 66mhz or 133mhz. My IBM EServer 235 has 133mhz slots. So at max bus speed, assuming full attention to the bus, you will be getting 64 bits of data 133,000,000 times every second. The calculation to get the speed in bytes per second would be 64/8 x 133,000,000 (for the simplest one) for the advertised 1,064,000,000 bytes per second or 1,064 megabytes per second (SI units). Some servers (at least mine) have the speed of the slot written somewhere on the case, if not in the manual or on the board. If you happened to have an older server with a 66mhz slot, then you would have a different calculation (but in essence it would be around half the speed because 66.66mhz is 133.33mhz/2).

Keep in mind that while this is the maximum theoretical bus speed, things like system interrupts, other cards, and other things may slow it down depending on motherboard construction, chipset, how many CPUs you have, and other factors.

Reply 2 of 28, by dexvx

User metadata
Rank Oldbie
Rank
Oldbie

I think a simpler calculation is to base it as a multiple of the standard 32bit/33MHz PCI (133MB/sec).

The slowest PCI-X is 64-bit/66MHz, so its basically 4x 133 = 533MB/sec
The most common high end PCI-X is 64-bit/133MHz, so its basically 8x 133 = 1066MB/sec

PCI-X 2.0 extended this to 266/533 Mhz, but those are quite uncommon due to proximity to PCI-Express. Fun fact is that by the time consumers could buy PCI-X devices, Intel already announced PCI-Express

Reply 3 of 28, by emosun

User metadata
Rank Oldbie
Rank
Oldbie
Ampera wrote:

It ultimately depends on the card and slot. All PCI-X cards/slots are 64-bit, meaning for each clock cycle of the bus, 64 bits of data are transmitted/received.

Then there is the clock speed. Most slots for PCI-X 1.0 are 66mhz or 133mhz. My IBM EServer 235 has 133mhz slots. So at max bus speed, assuming full attention to the bus, you will be getting 64 bits of data 133,000,000 times every second. The calculation to get the speed in bytes per second would be 64/8 x 133,000,000 (for the simplest one) for the advertised 1,064,000,000 bytes per second or 1,064 megabytes per second (SI units). Some servers (at least mine) have the speed of the slot written somewhere on the case, if not in the manual or on the board. If you happened to have an older server with a 66mhz slot, then you would have a different calculation (but in essence it would be around half the speed because 66.66mhz is 133.33mhz/2).

Keep in mind that while this is the maximum theoretical bus speed, things like system interrupts, other cards, and other things may slow it down depending on motherboard construction, chipset, how many CPUs you have, and other factors.

hmm interesting.

the machine specifically is a hp dl380. I was guessing it was pci-x 1.0 based on the age of the cpu's it has which are two prestonia 2.4ghz cpus. idk if that will help in determining the speed of the slots

Reply 4 of 28, by emosun

User metadata
Rank Oldbie
Rank
Oldbie
dexvx wrote:

The slowest PCI-X is 64-bit/66MHz, so its basically 4x 133 = 533MB/sec
The most common high end PCI-X is 64-bit/133MHz, so its basically 8x 133 = 1066MB/sec

so would it be fair to say that provided you are using a native pci-x card , that pci-x is roughly 4 times the speed of standard pci at it's slowest?

Reply 5 of 28, by dexvx

User metadata
Rank Oldbie
Rank
Oldbie
emosun wrote:
dexvx wrote:

The slowest PCI-X is 64-bit/66MHz, so its basically 4x 133 = 533MB/sec
The most common high end PCI-X is 64-bit/133MHz, so its basically 8x 133 = 1066MB/sec

so would it be fair to say that provided you are using a native pci-x card , that pci-x is roughly 4 times the speed of standard pci at it's slowest?

Yea, but realize that while most PCI-X cards uses the same physical interconnects as 64bit PCI, they are not the same. There also exists 64bit/33MHz PCI slots (bandwidth is 2x 133 = 266MB/sec). These are not PCI-X, but the slots look like them. Example would be a Tyan Thunder K7 S2462.

Edit:

You said HP DL380 with Prestonia Xeons, so likely DL380 Gen3.

http://h18000.www1.hp.com/products/quicKspecs … /11473_div.HTML

Looks like it has 2x 64-bit/100Mhz and 1x 64-bit/133MHz. Now you have to see what your device is capable of to determine what it will link at.

Reply 6 of 28, by Ampera

User metadata
Rank Oldbie
Rank
Oldbie
dexvx wrote:
I think a simpler calculation is to base it as a multiple of the standard 32bit/33MHz PCI (133MB/sec). […]
Show full quote

I think a simpler calculation is to base it as a multiple of the standard 32bit/33MHz PCI (133MB/sec).

The slowest PCI-X is 64-bit/66MHz, so its basically 4x 133 = 533MB/sec
The most common high end PCI-X is 64-bit/133MHz, so its basically 8x 133 = 1066MB/sec

PCI-X 2.0 extended this to 266/533 Mhz, but those are quite uncommon due to proximity to PCI-Express. Fun fact is that by the time consumers could buy PCI-X devices, Intel already announced PCI-Express

That complicates things and makes rooms for errors. It's really only a matter of simple maths anyways, anybody wielding a calculator can figure it out.

Also watch out. Make sure you match your card's speed to the bus speed, because PCI-X backplanes operate at the speed of the slowest card. Most PCI-X systems have multiple PCI backplanes at different speeds to fix this, but it's just something to pay attention for.

PCI-X is sort of like an even less successful VLB. It's long, not all that practical, has strange issues with it, and at the end of the day was replaced by a better standard that stuck firm.

Reply 7 of 28, by luckybob

User metadata
Rank l33t
Rank
l33t

this image comes in handy:

pci-x-modes.gif

It is a mistake to think you can solve any major problems just with potatoes.

Reply 8 of 28, by luckybob

User metadata
Rank l33t
Rank
l33t
Ampera wrote:

PCI-X is sort of like an even less successful VLB. It's long, not all that practical, has strange issues with it, and at the end of the day was replaced by a better standard that stuck firm.

1dxet4.jpg

pci x was a staple of the server market for a long time. It came into play at the mid to alt p3 era. like 1998. And was in EVERY server board made until the core2duo era (771). It was only then replaced by pci-e. but it was still available on boards at late as socket 1366. like this badass: http://www.supermicro.com/products/motherboar … IPMI=O&LRDIMM=Y

it had a very solid run spanning 7 cpu sockets.

It is a mistake to think you can solve any major problems just with potatoes.

Reply 9 of 28, by emosun

User metadata
Rank Oldbie
Rank
Oldbie
luckybob wrote:

pci x was a staple of the server market for a long time. It came into play at the mid to alt p3 era. like 1998. And was in EVERY server board made until the core2duo era (771). It was only then replaced by pci-e. but it was still available on boards at late as socket 1366. like this badass: http://www.supermicro.com/products/motherboar … IPMI=O&LRDIMM=Y

it had a very solid run spanning 7 cpu sockets.

can you edit that so there isn't a picture of trump?

I get the joke but god I don't want to even look at him even ironically

Reply 10 of 28, by luckybob

User metadata
Rank l33t
Rank
l33t

yes I can.

will I?

No.

It is a mistake to think you can solve any major problems just with potatoes.

Reply 11 of 28, by Ampera

User metadata
Rank Oldbie
Rank
Oldbie
luckybob wrote:
https://i.imgflip.com/1dxet4.jpg […]
Show full quote
Ampera wrote:

PCI-X is sort of like an even less successful VLB. It's long, not all that practical, has strange issues with it, and at the end of the day was replaced by a better standard that stuck firm.

1dxet4.jpg

pci x was a staple of the server market for a long time. It came into play at the mid to alt p3 era. like 1998. And was in EVERY server board made until the core2duo era (771). It was only then replaced by pci-e. but it was still available on boards at late as socket 1366. like this badass: http://www.supermicro.com/products/motherboar … IPMI=O&LRDIMM=Y

it had a very solid run spanning 7 cpu sockets.

VLB had a long running life, starting with presences on 386 board, having it's time in the sun on a LOT of 486 boards, and even making appearances on socket 4 and 5 boards. Maybe not as long ran as the PCI-X standard, it had similar traits. Perhaps to say less successful VLB would be wrong. In the server market it was a runaway success, but it saw 0 presence in the consumer market. With PCI slots still available on modern boards today, and PCI-E popping up around 13 years ago, PCI-X still had a shortish lifespan compared to other slot standards.

EDIT: I would also like to note that this is a way to give me solace for paying over 100 bucks for an old VLB board on E-Bay, even though I do love it.

Reply 12 of 28, by emosun

User metadata
Rank Oldbie
Rank
Oldbie
Ampera wrote:

but it saw 0 presence in the consumer market.

to be totally fair.... pci-x was used on the powermac g3 , g4 , and g5 towers

so it was at least 0.1 presence in the consumer market. and yes I consider the powermac a consumer product more than a "server only" product

not saying you're wrong just trying to give credit due

Reply 13 of 28, by Ampera

User metadata
Rank Oldbie
Rank
Oldbie
emosun wrote:
to be totally fair.... pci-x was used on the powermac g3 , g4 , and g5 towers […]
Show full quote
Ampera wrote:

but it saw 0 presence in the consumer market.

to be totally fair.... pci-x was used on the powermac g3 , g4 , and g5 towers

so it was at least 0.1 presence in the consumer market. and yes I consider the powermac a consumer product more than a "server only" product

not saying you're wrong just trying to give credit due

Eh, that is an exception. Maybe the word you are looking for is prosumer.

Reply 14 of 28, by NJRoadfan

User metadata
Rank Oldbie
Rank
Oldbie
dexvx wrote:

Yea, but realize that while most PCI-X cards uses the same physical interconnects as 64bit PCI, they are not the same. There also exists 64bit/33MHz PCI slots (bandwidth is 2x 133 = 266MB/sec). These are not PCI-X, but the slots look like them. Example would be a Tyan Thunder K7 S2462.

This is easy to spot though since all PCI-X slots are keyed for 3.3V cards. 64-bit 33Mhz PCI slots are almost always keyed for 5V. Later Power Mac G5s were all entirely PCI-X and lacked standard 5V PCI slots, thus leading to the rise in "universal" keyed 3.3V/5V PCI cards on the market. Before that it was fairly rare to see a 3.3V keyed standard PCI card. Many standard 32-bit PCI cards also operated at 66Mhz, but this was rarely advertised.

Reply 15 of 28, by luckybob

User metadata
Rank l33t
Rank
l33t
NJRoadfan wrote:

Many standard 32-bit PCI cards also operated at 66Mhz, but this was rarely advertised.

notable cards:

Voodoo 5 pci
this particular ethernet card: https://www.newegg.com/Product/Product.aspx?I … N82E16833166002

It is a mistake to think you can solve any major problems just with potatoes.

Reply 16 of 28, by dexvx

User metadata
Rank Oldbie
Rank
Oldbie
NJRoadfan wrote:

This is easy to spot though since all PCI-X slots are keyed for 3.3V cards. 64-bit 33Mhz PCI slots are almost always keyed for 5V.

I think there are 3.3V 64-bit PCI slots such as those found on Supermicro i860 boards (see links below).

PCI Expansion • 2x 64-bit 66MHz PCI (3.3V) slots

luckybob wrote:

pci x was a staple of the server market for a long time. It came into play at the mid to alt p3 era. like 1998. And was in EVERY server board made until the core2duo era (771). It was only then replaced by pci-e. but it was still available on boards at late as socket 1366. like this badass: http://www.supermicro.com/products/motherboar … IPMI=O&LRDIMM=Y

it had a very solid run spanning 7 cpu sockets.

@luckybob, I think you are confusing PCI-X with 64-bit PCI. PCI-X did not come into existence until the E7500 chipset:

https://www.supermicro.com/products/motherboa … E7500/P4DP6.cfm

Even the i860 (Xeon 603) only had 64-bit PCI.

https://www.supermicro.com/products/motherboa … n/860/P4DCE.cfm
https://www.supermicro.com/products/motherboa … 60/P4DC6_II.cfm

The Pentium 3's also only had 64-bit PCI.
https://www.supermicro.com/products/motherboa … SL/P3TDE6-G.cfm
https://www.supermicro.com/products/motherboa … SL/370DE6-G.cfm

Ampera wrote:

In the server market it was a runaway success, but it saw 0 presence in the consumer market. With PCI slots still available on modern boards today, and PCI-E popping up around 13 years ago, PCI-X still had a shortish lifespan compared to other slot standards.

PCI-X lifespan was quite short. But it still had an impact on the workstation and high-end consumer space. For U160 SCSI adapters, you had to get PCI-X (or 64-bit PCI) to get the best usage. Same with Gigabit Ethernet. Especially true if you combine both.

Reply 17 of 28, by Jade Falcon

User metadata
Rank BANNED
Rank
BANNED
emosun wrote:
luckybob wrote:

pci x was a staple of the server market for a long time. It came into play at the mid to alt p3 era. like 1998. And was in EVERY server board made until the core2duo era (771). It was only then replaced by pci-e. but it was still available on boards at late as socket 1366. like this badass: http://www.supermicro.com/products/motherboar … IPMI=O&LRDIMM=Y

it had a very solid run spanning 7 cpu sockets.

can you edit that so there isn't a picture of trump?

I get the joke but god I don't want to even look at him even ironically

Can you remove your post. Some trump supporters may find it offensive.

Reply 18 of 28, by Jade Falcon

User metadata
Rank BANNED
Rank
BANNED
Ampera wrote:
luckybob wrote:
https://i.imgflip.com/1dxet4.jpg […]
Show full quote
Ampera wrote:

PCI-X is sort of like an even less successful VLB. It's long, not all that practical, has strange issues with it, and at the end of the day was replaced by a better standard that stuck firm.

1dxet4.jpg

pci x was a staple of the server market for a long time. It came into play at the mid to alt p3 era. like 1998. And was in EVERY server board made until the core2duo era (771). It was only then replaced by pci-e. but it was still available on boards at late as socket 1366. like this badass: http://www.supermicro.com/products/motherboar … IPMI=O&LRDIMM=Y

it had a very solid run spanning 7 cpu sockets.

VLB had a long running life, starting with presences on 386 board, having it's time in the sun on a LOT of 486 boards, and even making appearances on socket 4 and 5 boards. Maybe not as long ran as the PCI-X standard, it had similar traits. Perhaps to say less successful VLB would be wrong. In the server market it was a runaway success, but it saw 0 presence in the consumer market. With PCI slots still available on modern boards today, and PCI-E popping up around 13 years ago, PCI-X still had a shortish lifespan compared to other slot standards.

EDIT: I would also like to note that this is a way to give me solace for paying over 100 bucks for an old VLB board on E-Bay, even though I do love it.

Asus had a lot of high end desktop boards up to the x58 platform with pci-x slots. I believe gigabyte did too.

885054651_o.jpg

Pcix was around for over 10 years in both high end desktops and servers and can still be seen in use today.
Heck the dell gx240s even had pcix. Pcix was a stable of oem system in the early 2000s and made its way into higher end systems.

Vlb had a very short window of use wile techicly it was around for quite some time it was really only in use for a few short years

Reply 19 of 28, by luckybob

User metadata
Rank l33t
Rank
l33t

to me, they are the same thing. it's kinda like the difference between pci-e generations.

It is a mistake to think you can solve any major problems just with potatoes.