Light Peak is holding things up I think.
Actually, it's not, as LP parts aren't even due to ship until Q4 2010, which means if on time, it won't appear in systems until 2011.
The new CPU's are meant to work in existing LGA1366 boards with a microcode update. It's cheaper for Intel and vendors to take this approach (chipset and resulting boards can run for 2 years, with only the CPU changing to improve system performance). In most cases, new tech, such as LP, USB 3.0, FW1600, SATA 6.0 Gb/s,... are expected to be added to the non-compliant boards via add-on cards.
Newer board designs will incorporate them when determined financially feasible (i.e. is it a budget board that relys solely on bare minimum part count = fewer features, or a high-end unit that additional semis are added for features prior to their inclusion by Intel in the chipset). A good example is USB 3.0 not being added to Intel's chipsets until 2012 (formally announced not too long ago).
I don't know if it will really catch on and be able to replace everything that it potentially could, but I expect that that will be the intention.
It's definitely intended to be capable of replacing a fair few interfaces, but we'll have to see what "shakes out" (most will result over cost).
The cost of high end network infrastructures is a large percentage of the overall cost of the potential customer base of the Pro machines and typically limits the usefulness and scalability of the product.
I'm not so sure on this, as I've the strong impression most are owned/operated by small shops (i.e. graphics design), and independents more than larger entities.
In the former group, they're less likely to be on such a high speed network, and run more as an independent (stand alone) system. The network might be there for a backup solution, if there's not a dedicated means per system. But comparing price for small scale solutions, the independent means is going to be cheaper.
For the latter group, yes, there are significant investments in networking equipment for the desired requirement. But this is shrinking for workstations from what I've seen. In these cases, it's more common to use clusters and then attach the workstation via say FC or Infiniband.
LP would be desirable here though, and is a big selling point IMO (makes small clusters a practical solution for SMB or independents with a high computing requirement).
It is expensive to even support Gigabit speeds for any reasonable number of machines and I can picture Intel really looking hard at getting into the networking game by bypassing the current status quo of Ethernet.
I agree. When I saw the video of LP's first demonstration, that's one of the things that immediately crossed my mind. This would give Intel a serious "leg up" over the competition.
Assuming LP hits their target pricing, it will be cheaper too per system (i.e. card and cable ~$100 per, not including the switch; not sure what the switch costs will be yet). But I suspect they're well aware that it needs to come in under 10G Ethernet to make it attractive. Assuming this happens, the enterprise community will bite.
...(duplex fiber? Seriously?)...
LP is actually Duplex as well (Up and Down are each 10Gb/s). It's just a lot thinner = smaller cable.
And as for need, of course they need it, it's just too expensive for consumers right now (artificially so if you ask me.) You want to switch back to 100 megabit Ethernet? How about 10 megabit? You can copy the same stuff, it just takes longer right? You might argue that the difference between a second and ten is not very great, but given the choice I'm sure most will go with what is faster, and that's what LP is promising. Remember it's starting at 10GB, and it will scale from there. Moving around terabytes of data, especially locally is becoming really common even for 'normal' people. Using cheap fiber cables (yes there is such a thing!) and low cost high volume components LP could potentially change how data is moved.
It's not necessary for consumers to actually need 10G yet, but that doesn't mean they're not interested either.
For consumers, the data rate is usually governed by their ISP throughput anyway (i.e. ISP signal distributed via a router to the other systems in the household). That's not to say they don't transfer large files between systems, but it's not critical that it be done within a fixed period (fast enough to transfer say 1TB/s, as you might need in an enterprise environment for nightly runs such as backup).
Also, you may recall that LP was developed in partnership with Apple (and may have actually been a 100% Apple project.) And it was demoed on Apple hardware by Intel, so I'm willing to bet we will see it there first. And as I said before probably on the next generation Pro's.
The hardware used in the demo wasn't Apple's, but appear to be Intel's own products and commodity components (i.e. PSU, CPU cooler with an LED fan,...). Apple's boards use proprietary connectors, even for the PSU (think of how HDD's and graphics cards get power on an Apple board).
Apple may have played a part (and I think they did, but their contribution is with
software, which is why the system was running OS X), Intel is the driving force behind LP in conjunction with other partners. Apple would benefit from this type of arrangement, as they get OS X developed in time, with most of the bugs worked out at the time of release.
and how much do you think LP will be to start off with? I'm not so sure apple would be keen to include it, it could add $500+ to the machine costs - does that seem a reasonable number?
It won't be that expensive, as the parts are only expected to be ~$50USD per LP connection. I'm not sure of switch costs yet, as fewer units sold mean a higher R&D figure per unit sold in order to recoup those costs.
But it would be attractive for Apple as users would be better able to set up small clusters for render farms. As it's a cheaper way to go, that means more independents and small shops can implement such a solution. That means more system sales for Apple (rather than just one workstation per user). Those additional systems may be XServes (assuming they're willing to have a rack rather than just pedestal systems). Either way, more systems sold would increase Apple's sales figures.
of course i don't want to go backwards - but i disagree that even "normal" people have a need for 10GBoE. will "normal" people have a RAID cluster that is fast enough to support that? no. all they have are external USB or internal HDDs that cant go any faster then 1GBoE at this point in time.
Good point, as that much throughput means there must be another part of the system that can push data at such a rate. Ultimately, there's additional costs over the network, and it's all expensive. This means it's out of reach for the average consumer at this time.
Consumer products are all made with a single driving compromise;
low cost.
Seriously, there are couple Macs in need of an update, Mac Pro and MBA. There are always some updates in autumn so lets keep our thumbs up for that! All Macs except Mini and maybe MacBook can be updated some time during the autumn.
