Announcement

Collapse
No announcement yet.

PCI Express 6.0 Reaches Version 0.5 Ahead Of Finalization Next Year

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • brouhaha
    replied
    Originally posted by StefanBruens View Post
    For wireless communications, PAM-2 is called BPSK (Binary Phase Shift Keying), two points on a unit circle spaced 180 degrees apart, which happen two lie on a straight line through the circle center.
    I may be nitpicking a bit here, but that's not quite correct, because PAM is amplitude modulation, not phase modulation (or phase keying). The radio equivalent of PAM-2 is OOK (on-off keying), which is used by many remote controls but not for high-data-rate communication. There is no commonly used radio modulation equivalent to PAM-3 or higher; high-data-rate wireless uses QAM, which modulates both amplitude and phase, or OFDM, which is multiple QAM carriers, as used by e.g. WiFi and LTE.

    Leave a comment:


  • tg--
    replied
    Originally posted by bug77 View Post
    Yup, FEC is mentioned in there, too.

    I'm not sure what "reasonably cheap" means, looking at what PCIe4 did to motherboards. But who knows, maybe enterprise will take care of costs this time around and it will be affordable once we need it in a PC.
    Oh, PCIe 6.0 will be MUCH more expensive than 5.0 hardware, PAM-4 coding costs realestate in the chipsets, and FEC even more.
    However, there the usual gamble mostly works: silicon will get cheaper, and sooner or later this isn't a problem.

    In contrast to doubling the clock, which needs board-realestate instead of (or in addition to) silicon, which won't come down in price, since the limit here is already physics.

    Leave a comment:


  • ms178
    replied
    Speaking of being cheap, I hope they found a way to decrease costs, I am not a fan of inflating price levels due to higher costs from build materials and using retimers. The engineers should have come up with something more clever than that.

    Leave a comment:


  • bug77
    replied
    Originally posted by tg-- View Post
    There is always a tradeoff to be made, nothing comes for free.
    Usually higher-order modulation works better than doubling the clock, which comes with its own set of downsides.
    To achieve acceptable robustness, PCIe 6.0 will very likely also add Forward-Error-Correction, which will decrease the net-throughput.
    Yup, FEC is mentioned in there, too.
    Originally posted by tg-- View Post
    Most importantly, PAM-4 is a very very simple modulation, which makes it reasonably cheap. Doubling the clock would require quite a bit more expensive hardware.
    I'm not sure what "reasonably cheap" means, looking at what PCIe4 did to motherboards. But who knows, maybe enterprise will take care of costs this time around and it will be affordable once we need it in a PC.

    Leave a comment:


  • tg--
    replied
    Originally posted by bug77 View Post

    And dividing the acceptable noise levels by 3. This is going to work wonders for signal integrity/reach.
    Though I'm sure the engineers working on the spec are more aware of this than I am.
    There is always a tradeoff to be made, nothing comes for free.
    Usually higher-order modulation works better than doubling the clock, which comes with its own set of downsides.
    To achieve acceptable robustness, PCIe 6.0 will very likely also add Forward-Error-Correction, which will decrease the net-throughput.

    Most importantly, PAM-4 is a very very simple modulation, which makes it reasonably cheap. Doubling the clock would require quite a bit more expensive hardware.

    Leave a comment:


  • bug77
    replied
    Originally posted by tg-- View Post

    To provide a slightly simpler explaination than StefanBruens, in addition:
    The signal lines to PCIe ports are not purely digital (on, off/1, 0), they carry a defined voltage.
    What PAM does is add the possibility to decode partial voltages in steps (PAM-4 is 4 steps), which increases the datarate.
    In PAM-4 the voltage between 0 and 1 is divided in 4 steps, so every cycle encodes 2 bits (2^2 is 4 steps), thus doubling the data rate.
    And dividing the acceptable noise levels by 3. This is going to work wonders for signal integrity/reach.
    Though I'm sure the engineers working on the spec are more aware of this than I am.

    Leave a comment:


  • tg--
    replied
    Originally posted by FireBurn View Post
    Michael what's PAM-4 encoding?
    To provide a slightly simpler explaination than StefanBruens, in addition:
    The signal lines to PCIe ports are not purely digital (on, off/1, 0), they carry a defined voltage.
    What PAM does is add the possibility to decode partial voltages in steps (PAM-4 is 4 steps), which increases the datarate.
    In PAM-4 the voltage between 0 and 1 is divided in 4 steps, so every cycle encodes 2 bits (2^2 is 4 steps), thus doubling the data rate.

    Leave a comment:


  • StefanBruens
    replied
    Originally posted by FireBurn View Post
    Michael what's PAM-4 encoding?
    PAM - Pulse Amplitude Modulation

    Binary would be PAM-2, two different voltages or currents. Current ethernet (twisted pair, 1G) uses PAM-5. Current SSDs use multple levels as well, so you could call it PAM-4 ("MLC", multi level), PAM-8 ("TLC", triple, more correctly 2^3) or even PAM-16, though the PAM term is typically only used for transmissions.

    For wireless communications, PAM-2 is called BPSK (Binary Phase Shift Keying), two points on a unit circle spaced 180 degrees apart, which happen two lie on a straight line through the circle center. For higher bitrates the first step is two add a second, same frequency but orthogonal carrier (sine and cosine), which gives you QAM. If each of the carriers has two levels, you have QAM-4 (though thats called QPSK), with 4 levels each ("PAM-4") you have QAM-16. Most current WIFI uses up to QAM-1024, i.e. up to 32 different levels per carrier.

    The win of using higher modulation is higher data rates without using more spectrum. The downside is each valid level has less distance to its other valid neighbours, so it can't tolerate as much noise. Also the hardware implementation becomes more complex - the receiver now essentially needs an ADC running at several GSample/s, though only with a few bits.

    Leave a comment:


  • nomadewolf
    replied
    Originally posted by Danny3 View Post
    Does the chipset still comes with a fan?
    I don't want any moving parts on the motherboard which makes it less reliable.
    And I don't want any noise coming from the motherboard. I want to make the computer as silent as possible.
    I assume the power consumption is even higher which I don't like also.
    So, with all these anti-features, I think I will say again no thanks and buy an older motherboard instead.
    There are motherboard options with passive cooling.

    Leave a comment:


  • jo-erlend
    replied
    Originally posted by Danny3 View Post
    Does the chipset still comes with a fan?
    I don't want any moving parts on the motherboard which makes it less reliable.
    And I don't want any noise coming from the motherboard. I want to make the computer as silent as possible.
    I assume the power consumption is even higher which I don't like also.
    So, with all these anti-features, I think I will say again no thanks and buy an older motherboard instead.
    You can always use passive cooling, you know.

    Leave a comment:

Working...
X