Please, explain what is this device “optical fiber router” ? Does it route light? BGP or OSPF?
In Australia, Singapore, UK, USA, Japan, and so on ... there are internet connections terminated with optical cable to the home. The 2 routers work together to deliver optical to copper connections.
[doublepost=1521026377][/doublepost]
If Apple intends for the new MP to sell in any kind of volume, it needs to model it after the cheesegrater design. Shrink the case, sure, but provide PCI-e slots, user-configurable memory, storage, and GPUs, etc.
In other words, yes, sell a Mac version of the HP Z6/Z8.
If Apple doesn't do that, countless comments strongly suggest it shouldn't bother. Those few power users still on Macs need Apple to extract its head from its ass or they too will migrate away.
The big pitch idea in the draft spec was the "jet engine" Mac Pro had great control over EMF / EMR for compliance standards. In many respects the design is actually a good solid concept. Heavy inner cooling core with no interference leakage and a solid metal outer shell. This type of sandwich gives you excellent design control over "radio" interference.
The concept was not a lightweight idea. But Apple changed the core from 4 to 3 sides. I think we all see that was one of the reasons for the main limitations in thermal mass and internal storage. When someone comes up with a very carefully made design and someone else makes alteration -> ooops.
[doublepost=1521027237][/doublepost]
Most "PRO" hardware are actually under-clocked version with tantalum capacitors and better cooling, I'm familiar with your concept.
As I see the only way to rise the tcMP design is with AMD's PRO APU, but it barely showed in early benchmarks (a 8 zen core with 24 gpu core) this is aimed to HPC in dual socket configurations, could be the only way to make the trash can form factor competitive or at least somewhat useful as real workstation (for those not requiring dual FULL GPU, and more than 16 cores).
But I dont believe Apple to re-launch the tcMP, as much they will provide an upscaled mac mini and a modular mac pro but not modular in the sense many hopes, but I think they at least should allow one or two 8x PCIe slots for hardware other than GPUs, and sell easy to install often updated new gpu asap are available (the classic minimal redesign of reference designs, not full custom since require much more twaking).
The mMP with derated hardware makes little sense, even the design should allow highr TDP and more PSU power for follow up updates/upgrades.
I agree with you, but I also want to share another thought on this topic. Clock speed can be pushed higher and higher but what we really need is optimal matched hardware.
The logic and reasoning goes like this .... you can obtain vastly superior performance from a system if you match the hardware carefully to the functions intended. This happened already once before in the last few years. A very old model of AMD graphic card was the "king" of bitcoin mining. Even now people are interested to buy in quantity of 100 and even 1000.
When a data process and a hardware system are perfectly matched, the speed can be more than impressive, it can be stunning. This is why some servers are crazy fast and others are not, for the same application. However, I believe that we are no longer moving slow enough in engineering to allow for this type of well planned design. One of the reasons is that chip makers no longer care if you miss the boat with a chip. If your design takes too long to make it to the market, you simply wont have enough parts to build it. You miss out. Start again.
Moore's law has confused itself. Real world performance does not rely on clock speed. It requires some help from data buffering and clock synchronisation. For example. You would be 10x better off installing your ram in banks where only 50% or 33% was addressable by the user. (you install 512GB ram but only get 256GB in the real time). The simple reason is that opening up data channels (DDR4 versus "pseudo DDR16") is faster than making the lanes run wider (8 bit versus 32bit).
But these ideas have died a long time ago except in file servers. When you can build a machine with "standby ram" and "spare ram" and "error correction for the error correction" then you can achieve the scalar effectiveness seen in many middle range servers of being up to 197X faster than a single 4 core server.
Dont get me wrong, I am not married to my DL980. It is just one of those strange machines with crazy specifications. I booted this machine with wrong ram, faulty boards, faulty CPU with 3/8 broken power supply and it still ran.
