B
Bill
[email protected] says... said:
Just what do you have populating all those PCI slots, if I may ask?
Bill
Just what do you have populating all those PCI slots, if I may ask?
Bill
A
Arthur Hagen
Bill said:
Let's see...
- HiFi multi-channel audio card (and no, you can't do the same with USB, as
the latency is way too high).
- 2nd RAID controller (raid 1+0) (in addition to the onboard one).
- SCSI controller for external tape station.
- TV card (again, USB 2.0 is too slow to get full quality without stuttering
if you actually do something).
- Network card with encryption and CPU offloading.
I'm currently watching time-shifted TV in full quality on my second monitor
and downloading a several hundred megabyte file, and CPU use is less than
10%. And this is not the fastest system in the world either -- a more than
two year old P4T533 with 512 MB RDRAM and a 3.06MHz P4.
It does have two PCI buses, though, which helps a lot.
Regards,
A
Arthur Hagen
Tom Lake said:
USB is a serial protocol, with the advantages *and* disadvantages that
incurs. The problems include high and non-predictable latency and packet
loss during saturation. It's no replacement for PCI (and neither is PCIe --
PCI-X, on the other hand would be nice to see more of).
Regards,
B
Bill
Yup, thats a lot of stuff. Would a pci bus extender be worth looking into, or are they more
trouble than they're worth?
Bill
D
Doug
This is exactly the response I was looking for. I remember studying the PCI
standard in a microprocessors course in college and there were 64-bit and 66
Mhz standards in place for PCI even back then (and I've heard they were even
implemented). So PCI Express is more of a marketing move to obsolete
motherboards than a huge leap forward in bus design?
standard in a microprocessors course in college and there were 64-bit and 66
Mhz standards in place for PCI even back then (and I've heard they were even
implemented). So PCI Express is more of a marketing move to obsolete
motherboards than a huge leap forward in bus design?
D
Doug
Huh? Are you saying differential data signalling takes less power? Um I know
for a fact RS-422 uses MORE power than RS-232.
for a fact RS-422 uses MORE power than RS-232.
J
J. Clarke
Doug said:
It pretty much seems that way to me.
Server boards are currently available with 64-bit 133 MHz slots, and the
current revision of the spec provides for 266 and 533 although I don't know
of anybody shipping hardware yet.
P
Phoenix AG
True, as of now, 4x and 8x have no difference.
I am with you on this one
A person tends to need a lot of slots. Ihave 3 PCI and 2 PCIe 1x slots in my board, with a PCIe 16x for the
gfx card. The placement is such that one of the PCIe 1x slots is
useless as it is too close to the graphics card.
But, I have found that you tend to use those slots less if the
motherboard is full featured.
In my old comp, I had 2 network cards and 1 sound card attached to the
PCI slots. Now a lot of that has become redundant. My new MB already
has a gigabit ethernet port and 7.1 surround sound. So I don't need
the ethernet card. I am still testing the sound but as of now, its
quite satisfactory. The other ethernet card, I really don't need now
as I have recently purchased a wireless router.
So really, I used to need a lot of slots earlier. Now, all of my 3 PCI
slots and 2 PCIe 1x slots are free on the board. I just don't know
what to put in them. Earlier, I even had a PCI to USB add in card
which added 2 more USB ports to my computer. I don't even need that
cos now I have 8 USB 2.0 ports. More than I need
***
....the Phoenix shall rise...
P
Phoenix AG
As this is a video card newsgroup, I focused on the video card utility
of the PCIe architecture. Let's try to keep things on topic.
Or the speed of PCI-XYZ. Or ABC. The thing is, it can be anything. And
change is good. I am glad they are making things faster. Ok, so it may
not be PCIe, but PCI-X. You will still need the speed sometime or the
other. What really surprises me about your argument is your question
"Where is the need?" That reminds me of that old statement by Bill,
640k should be enough for everyone.
You can argue that with Intel
Very simply put, PCIe is the natural progressor and successor of AGP
in the market. As you yourself put it, you would be hard pressed to
find a new board without PCIe on it.
Faster bandwidth? Allowing for some new gfx cards utilizing that
faster bandwidth.
I am aware that the current generation cards don't even use the full
power of AGP 8x. But does that mean that we should be lazy and be
happy with 8x for the rest of our lives?
Yes, those made for AGP buses are the 5th generation of Geforce cards
and the 9000 series by ATI. Those made for PCIe buses are the NEW ones
like the 6th generaton of Geforce and the X series by ATI.
Actually, you can but I won't argue this.
http://www.asus.com.tw/products2.aspx?l1=3&l2=11
Look at P5P800 and below. All AGP boards supporting LGA775 processors.
Come on, you should at least research your arguments
And that's just one manufacturer's website. Many more out there.
My motherboard supports the newer Intel processors with 64bit
extensions. I bought it on a guarantee that within 6 months, I would
be able to exchange my old CPU with one of the newer 64bit ones.
Anyway, as of now, there is no real *need*, as you say of PCIe, of 64
bit processors.
No need for a 64bit processor too.
As you may be aware, the technology that has the strongest market
backing wins. You've heard of RDRam, right? That was backed by Intel.
But Intel did not win the fight with DDR.
Probably RD was as fast or would have been faster in its later
iterations than DDR. But it was killed off from the market by DDR. So
really, maybe PCI-X is faster than PCIe. But the way things are going,
it might just find itself killed off.
Not really. I have given you the link above. You can still buy a new
machine and keep the obsolete AGP architecture, if you want.
Really? Replying to a person's sig is not the highest display of
maturity. And I don't even understand what you're trying to say with
that.
***
....the Phoenix shall rise...
T
Tom Lake
So? One is not _forced_ to accept a P4 3.0+ GHz processor. If one buys a
You can buy a new machine but NOT with the current chipset. The latest
chipset they sell with AGP is the 865PE in their P5P800 board. The Intel
800 series chipset is obsoleted by the 900 series and won't be available
much longer. It looks like the only video bus available with the 915 and
925 chipset is the PCIe.
Tom Lake
You can buy a new machine but NOT with the current chipset. The latest
chipset they sell with AGP is the 865PE in their P5P800 board. The Intel
800 series chipset is obsoleted by the 900 series and won't be available
much longer. It looks like the only video bus available with the 915 and
925 chipset is the PCIe.
Tom Lake
A
Arthur Hagen
Phoenix AG said:
The problem with the on-board features is that they're usually most limited
in features. Your sound card, for example, most likely won't allow you to
listen to a CD and have system sounds at the same time as you capture a TV
show, without messing up the recording. Nor use MIDI. And the network card
almost certainly doesn't have its own processor and RAM for CPU offloading
and IPSec encryption. The difference can be quite amazing -- much like a
few years ago before broadband became common, when people used the bundled
Winmodem that came with the computer instead of a real modem, and wondered
why everything slowed down to a crawl.
Then there's all the internal ports for the 7.1 sound which takes up back
plate space. Meaning you most likely won't have a serial port, and would
have to take up a PCI slot if you need one. And why would you? Well,
there's still plenty of equipment that wants a serial port, including older
PDAs and printers, external modems and IR readers, but more relevant: any
high grade router is set up through a serial line, and so are servers (which
come without a video card and keyboard input).
So really, I used to need a lot of slots earlier. Now, all of my 3 PCI
slots and 2 PCIe 1x slots are free on the board. I just don't know
what to put in them. Earlier, I even had a PCI to USB add in card
which added 2 more USB ports to my computer. I don't even need that
cos now I have 8 USB 2.0 ports. More than I need
It's all about having choices. Your system would not be usable for me, as
it doesn't give me enough choices. It almost certainly fits the 80/20 rule,
though, where they concentrate on the 80% of customers who will only ever
need 20% of the features.
Regards,
J
J. Clarke
Phoenix said:
Who appointed you netnanny of the week?
Huh? PCI-X is a technology that was in place and shipping long before PCI
Express.
No, it won't be PCI-X because Intel has decreed that it will be PCI Express.
If it was PCI-X then I would have no complaint--that is a logical
progression of PCI that does not obsolete existing hardware.
(a) Bill Gates did not ever say that. (b) I repeat, where is the need?
What demonstrated need is there for more bandwidth in desktop PCs? Why the
sudden rush to an untested new standard that is not backward-compatible
with existing hardware when there are alternatives available that _are_
backward compatible?
You can argue that with Intel
Does Intel claim that it is a "natural progression"?
Which has little to do with "nature" and much to do with marketing. Or
perhaps your "natural progression" really meant "change" and had nothing to
do with either "nature" or "progression".
Please present evidence that the bandwidth of the bus is the obstacle to
"better graphics".
To do what? And don't say "better graphics". Please demonstrate _how_ that
higher bandwidth will result in "better graphics".
When someone comes up with something that starts pushing that limit then
worry about it.
Check again. Both ATI and nvidia have produced 6th generation chips with
native AGP requiring bridges for PCI Express and they have both gone the
other way around as well. And so far there has not been one iota of
demonstrated difference in performance between the AGP and PCIe versions of
any of those boards.
Using the 865PE chipset, which is previous-generation.
Come on, you should at least research your arguments
You should read more carefully.
So? None of them have boards with 900 series chipsets and AGP buses.
Which are band-aids on a 32-bit processor.
Of course there is. But you don't compile code do you?
Of course there is. There is need for quad-64s. And higher. You think
that companies pay the price for those just to be k3wl?
At the time there were viable alternatives to Intel chipsets. Intel took
care to fix that first.
No, it was killed off by chipsets from via and other companies that offered
the same or better performance at lower price with DDR.
And this is desirable why?
One can buy a new motherboard. Now, how many _machines_ are available on
the shelf at the mass-marketers with AGP?
Why am I not surprised? Humor challenged Intel fanboy who goes gaga over
anything new regardless of its actual utility it looks like.
J
J. Clarke
Arthur said:
I don't seem to have any problem doing that with MCE2K5 and an nforce2.
I've not seen any evidence of a real-world performance gain by offloading
this.
Actually, Winmodems can be quite quick.
Somehow ASUS manages to find room for 6 audio connectors, plus RCA
connectors for digital audio plus 4 USB ports plus 2 network ports plus a
conventional serial and parallel and I'm sure I left something out.
???? _any_? Many of them have in-band management. And some have USB
management ports.
???? Which servers are these?
B
Ben Pope
Arthur said:
Why is PCIe not a replacement for PCI?
What can PCI do that PCIe can't?
Ben
B
Ben Pope
J. Clarke said:
PCI Express does not obsolete existing hardware any more than PCI-X, if
the chipset supports only PCI Express, the motherboard manufacturer can
place a forwarding bridge to a PCI slot. This is COMPLETELY transparent
to any software, including drivers.
PCI Express is not much less backwards compatible with PCI, if
motherboard manufacturers decide that the majority of their customers do
not need PCI (whether thats true or not), then they will remove PCI
support, but until then, it is trivial to support PCI.
So the biggest problem here is that of slot choice of the motherboard
manufacturer. That and wanting to keep all of your old PCI devices for
eternity. Well, things move on, but it's not as bad as you suggest.
PCI Express 8x is about on par with PCI-X QDR 533MHz.
A PCI-X port is 64bit so if you take PCI-E to 64 bits, you get PCI
Express 32x, 4 times faster than PCI-X. Will PCI-X be able to go any
further than QDR? Will 128bit bus be appropriate? Will they be able to
push the clock speed above 133MHz without cross-talk and whilst
retaining compatibility? At some point some incompatibilty will need to
be introduced.
The current trade-off between PCI Express and PCI-X seems to be lower
latency on PCI Express, and generally higher throughput (8x PCI Express
vs PCI-X 533) at lower packet sizes (to about 1024bytes), but you have
to change the socket.
With PCI-E the (minor) incompatibility is now, with some some added
advantages such as lower latency, and more flexibility.
Ben
J
J. Clarke
Ben said:
So let me see you put a PCI board in a PCI Express slot.
It is not transparent to a person trying to stick a board in the machine.
So how do I put a PCI board in a PCI Express slot?
And we need more than this why?
And we need this why?
Does it need to?
Will it be possible to push PCI Express to a trillion exaHz? You're
claiming that compatibility should be sacrifices to gain a small increment
of performance for which there is no demonstrable need.
I'm sorry, but changing the socket just to get NO gain in performance in the
real world smacks of marketing and not engineering.
A
Arthur Hagen
Ben Pope said:
PCIe is a serial protocol striped over multiple channels ('lanes' in PCIe
terminology). It has the advantages of easier board design, theoretical
high speed, but at the cost of overhead (a 1x PCIe card is in theory
transmitting 256 Mb/s, but only 160 Mb/s is usable for data, as the rest is
overhead), being unidirectional (send full speed one way, and nothing can
pass the other way) and high latency.
PCI (and PCI-X) is a semi-parallel system where each device can request
blocks of memory mapped to itself, and anything sent over the PCI bus to
that address space will be routed to the correct device. It allows for DMA,
where devices map to the memory directly, and also allows for one device in
the pool taking advantage of unused bandwidth, and other devices breaking in
as needed. The main disadvantages are lower total bandwidth (as it's
shared), and a more complex card design.
If comparing typical motherboards, keep in mind that PCIe 1x which is what
you find for everything except the graphics card is slower than even normal
PCI, and a step backwards. It's not suitable for adding a RAID controller,
for example, nor any other card that requires low latency or high bandwidth.
PCIe 8x and 16x has the bandwidth, but is generally not available (you need
one PCIe bus for each 16x).
Intel's dream seems to be to simplify things for the consumer, who should be
content with the drive controllers and sound facilities of the motherboard,
and stop adding stuff him/herself. It also splits the market in two (which
is also in Intel's interest), where server boards simply can't work with the
limitations of PCIe, and go PCI-X instead. That creates two different
market segments, and the prices on "server" boards can be kept artificially
high, while the PCIe boards can be sold at the same price as before, despite
being much cheaper to make.
What's the benefit for the consumer with PCIe? None, as of now. A 16x PCIe
card is theoretically faster than a 8xAGP, but due to protocol overhead and
higher latency, not noticably so.
The only thing I can think of is that you can get nVidia SLI with PCIe, but
that's not really a feature of PCIe, but simply that nVidia chose to do that
for PCIe only, as being the future standard as dictated by the tyrant. It
could just as easily be done with two AGP cards. For other cards, it's a
step back for the consumer, as PCIe 1x isn't good for much.
Regards,
A
Arthur Hagen
J. Clarke said:
There is little performance gain. That's not the point. The point is to
offload the CPU, leaving more CPU power for *other* tasks. When the
alternative is to buy a twice as expensive CPU to get 15% performance gain,
it makes good sense to limit the CPU use instead.
Tom's Hardware has an old, but still relevant article/review on this (from
back when THG could still be called independent):
http://www.tomsnetworking.com/network/20010820/index.html
Even without testing the IPsec offloading, the 3Com cards with checksum
offloading and onboard RAM used 15-17% CPU for FTP transfers where an Intel
Pro/100+ card used 31%. That's a *significant* difference.
Note that with gigabit ethernet, this becomes even more important.
Sure. But again, that's not the point. They're quick by using the system
resources, most noticably the CPU. That means you have less CPU time left
for *other* tasks, which slows down everything *else*.
A Winmodem isn't a modem at all, but a simplified sound card with the
input/output going to the RJ jack. The CPU has to do all the AT command set
parsing, converting the data to/from sound data, and pushing/retrieving it
from the modem. While that isn't slower than leaving it to the modem, it
consumes a LOT of CPU resources.
I didn't say "consumer router", even though some of those too do have serial
ports (like the Symantec and Nexland ones). "In-band management"? That
does you no good at all unless you can *reach* the sucker in the first
place.
About any 1U or 2U server you care to buy? A server doesn't need graphics,
keyboard or mouse. You set them up through a standard serial tty terminal
(9600 8N1), at least enough so they can get on your network and you can
continue configuration from there. That's standard procedure.
J
J. Clarke
Arthur said:
I didn't say "consumer router" either. In fact some of the larger Cisco
models have a _separate_ Ethernet port for out of band management via
Ethernet. And some Ciscos have USB ports.
Checking IBM, the 1U xSeries 306, 336 and 2U xSeries 343 & 346 have video,
keyboard, mouse, and USB ports. The same is true for all models I can
identify from Dell, DECPaquard, Intel, SuperMicro, and Tyan. The only
models I can find that do _not_ appear to have onboard display, keyboard,
and mouse are the IBM PowerPC servers and the Sun Sparcs, and even they
have USB ports in addition to serial.
I'd like to see you set up 2K3 using a serial port.
B
Ben Pope
Arthur said:
That depends on packet size, 200MB/s is quite likely too. With PCI
Express being faster than PCI-X, your point is moot.
There is also overhead on PCI-X, if you consider the QDR technology,
only a proportion of the data trasnfers can actually be QDR, the rest
must be SDR and thus relative overhead is increased, the faster you go.
Point in favour of PCI-Express as overhead remains constant with speed.
Err... what? For each lane there is a connection in each direction.
This makes duplex operation possible, something that PCI-X or PCI cannot
do. Point in favour of PCI-Express.
Depending on how you deal with fan-out, the flexibilty of the lanes
system reduces latency, but I don't think latency is ever increased over
PCI-X when dealing with fan-out. Since PCI and PCI-X are contention
based, not only for send and receive, but with other devices as well,
how could it possibly be lower latency than PCI-Express, a
point-to-point protocol?
Point in favour of PCI-Express.
Yes, I asked for differences between PCI-X and PCI-Express. PCI-Express
retains this memory mapped IO area.
I fail to see how 32, and 64 connections for a data/address path is
semi-parallel. Thats quite parallel, really, isn't it?
Yes, this is also the same.
Yes, on PCI and PCI-X there is contention and therefore bandwidth is
reduced and latency increased. With PCI-Express this contention is
removed as it is effectively point-to-point. PCI-Express wins.
PCI on typical motherboards is 32bit, 33MHz thats 133MB/s.
PCI-Express, as previously discussed is 200MB/s per lane, *in each
direction*, *simultanously*. Higher bandwidth and lower latency is a
step backwards?
But since PCI-Express has lower latency (there is no contention) and
higher bandwidth than typical PCI slots on typical Motherboards,
PCI-Express IS faster.
You also forgot to mention that PCI-Express x4 is possible and already
on boards that are currently being sold, and would be ideal if you
wanted say, a bunch of really fast drives in RAID and PCI-Express x1 was
not fast enough.
Eh? Typically a chipset will support at least 32 lanes.
Server has gone PCI-X because it satisfied their requirements sooner,
not because PCI-Express is inferior. In fact, PCI-Express is better in
almost all respects to PCI-X.
Thats just wrong. Typcially, PCI Express 8x is on a par with AGP8x in
terms of bandwidth.
Rubbish.
Go and read:
http://www.rtcmagazine.com/home/article.php?id=100149
or:
http://www.redbooks.ibm.com/abstracts/tips0456.html?Open
You might find it interesting.
Ben