4X4 News:

[Hard Ball]

Just when you thought that this 'QuadFather' effort couldn't get more haphazards and poorly executed, there is this in the news today:

http://www.theinquirer.net/default.aspx?article=35843

Chimpzilla's other partners did not receive a single set of 4x4 CPUs and they're unable to tell us if the 4x4 is going to work on their existing Opteron mobos or not.

An un-named product manager company accused AMD of, "acting like Intel". "We won't help them in future", he said.

... ...

Nvidia employed Asustek to manufacture all of its first-generation 8800 boards and 680a motherboards. Now AMD is doing the same thing - and miffing its far-eastern partners into the bargain.

On the other hand, that aforementioned Asus board has apparently arrived in some reviewer's hand:
http://www.gizmodo.com/gadgets/pcs/first-amd-4x4-mobo-spotted-216206.php

 

[aka1nas]

Was it even supposed to work in non 4x4 boards anyway? Why you would you even bother?

 

[lopri]

I'm curious what kind of HSF will fit the sockets? I don't know if existing Socket 939/AM2/775 HSFs will work. (The mounting holes don't look aligned square, but rectangular) Also, can you imagine TWO heavy-weight HSFs (such as Tuniq) hanging off the board.. The board is eATX form factor, so that alone will exclude many users with standard ATX cases. The board will likely command ~$400 price tag and two 2.6GHz A64 dual-core for $1,000.. This just doesn't seem practical. Will make a very colorful showroom rig, though.

 

[myocardia]

I don't see what difference it would make, as to how many board manufacturers make 4x4 boards. None of them are going to sell that many, anyway. The profit is in the $50-100 motherboards. My guess though, is that AMD could only talk Asus into building 4x4 boards if they built them exclusively, since they also realized that not very many will be sold.

 

[aka1nas]

Have they stated any details of the 680a chipset for 4x4 yet? I.E. Is this board a regular 680a chipset or are they doing something weird like linking two 680a's together for the extra PCI-E lanes? I would be more interested in a single socket board with 4 x16 slots sometime down the road.

 

[Skott]

Well, some may feel ASUS was a poor choice to partner with but I know there are far more worse choices than ASUS. It'll be interesting to see how it works out.

 

[SunnyD]

Why... WHY in the name of all things sane did they put PS/2 ports and a Parallel port on the damn thing?

As for the chipset, I'm thinking it has to have something new. If you look at the heat spreader, there are two sets of traces on the top layer of the board going to two of the PCIe slots. I would assume one of the mid or lower layers has another route to the other two PCIe slots (16x). My guess is that it will have a new chip capable of 64 PCIe lanes, give or take, which would mean it's not just a 5xx + 2200 combo. Also, the chips under the spreader are probably the same size, if I recall the 2200 was smaller than a 5xx significantly. The heat spreader looks a bit too small to fit on something like a 5xx chip.

And finally - there could be good reason Asus is getting the nod here. After all... NVIDIA picked Asus to be the sole supplier of the 8800 cards to world+dog. NVIDIA is making this chipset for AMD. Therefore, it's very very logical to extrapolate that NVIDIA wants to keep reigns on the 4x4 reference design, and as such insisted the contract go to Asus. I'm guessing that the NVIDIA-Asus connection has more to do with this situation than anything AMD wanted.

 

[RichUK]

Originally posted by: SunnyD
Why... WHY in the name of all things sane did they put PS/2 ports and a Parallel port on the damn thing?

It?s obviously been implemented for legacy support. I don?t see anything wrong with either the PS2 ports or the parallel port. I would however, be confused if they implemented COM ports, as their requirement has pretty much gone the way of the dodo.

Other than the obvious, it just broadens the target audience. Some people might still use standalone printers that communicate via the parallel port, instead of using USB. Also, I for one still require the use of the keyboard PS2 port, and I can imagine a lot of people still use both PS2 ports too.

 

[SunnyD]

Originally posted by: RichUK
Other than the obvious, it just broadens the target audience. Some people might still use standalone printers that communicate via the parallel port, instead of using USB. Also, I for one still require the use of the keyboard PS2 port, and I can imagine a lot of people still use both PS2 ports too.

I suppose... I do know most non-knowledgeable folk don't understand they can take the USB-to-PS/2 adapter off the end of their keyboard/mouse.

Honestly, if you're getting a modern system, you should have no need for legacy ports. That space would be much better utilized housing 2 more USB hub blocks. I've got 6 USB ports on the back of my system, one of which is occupied with the cable for the hub in my monitor, and frankly, I use all of those ports. I do not use either PS/2 port on my current motherboard. And finally, parallel printing in Windows 2000/XP is so ungodly slow, it's no wonder why most printers these days are not coming with a parallel port on them.

There really is no justification these days for the legacy ports.

 

[zsdersw]

Oh who cares.. it's a keyboard and/or a mouse that uses PS/2. It's not like there's an ISA slot on the motherboard.

 

[Hard Ball]

Originally posted by: lopri
Also, can you imagine TWO heavy-weight HSFs (such as Tuniq) hanging off the board.. The board is eATX form factor, so that alone will exclude many users with standard ATX cases.

Actually, that board looks to be 12' X 10.5', while technically EATX, should fit into almost all full-ATX boxes and most of the mid-towers. This is similar to K8N-DL, which has fit into a lot of people's standard ATX cases. It's actually the reason why they went with only 4X DIMM slots for this board, so that it may fit into most enthusiasts' cases today.

Edit: K8N-DL pics;
http://www.newegg.com/Product/ShowImage...xtended+ATX+Server+Motherboard+%2D+OEM

 

[Viditor]

Originally posted by: Hard Ball

Originally posted by: lopri
Also, can you imagine TWO heavy-weight HSFs (such as Tuniq) hanging off the board.. The board is eATX form factor, so that alone will exclude many users with standard ATX cases.

Actually, that board looks to be 12' X 10.5', while technically EATX, should fit into almost all full-ATX boxes and most of the mid-towers. This is similar to K8N-DL, which has fit into a lot of people's standard ATX cases. It's actually the reason why they went with only 4X DIMM slots for this board, so that it may fit into most enthusiasts' cases today.

Edit: K8N-DL pics;
http://www.newegg.com/Product/ShowImage...xtended+ATX+Server+Motherboard+%2D+OEM

Excellent point, and thanks for the data HB...that makes a good deal of sense!

 

[Arkaign]

Originally posted by: SunnyD

Originally posted by: RichUK
Other than the obvious, it just broadens the target audience. Some people might still use standalone printers that communicate via the parallel port, instead of using USB. Also, I for one still require the use of the keyboard PS2 port, and I can imagine a lot of people still use both PS2 ports too.

I suppose... I do know most non-knowledgeable folk don't understand they can take the USB-to-PS/2 adapter off the end of their keyboard/mouse.

Honestly, if you're getting a modern system, you should have no need for legacy ports. That space would be much better utilized housing 2 more USB hub blocks. I've got 6 USB ports on the back of my system, one of which is occupied with the cable for the hub in my monitor, and frankly, I use all of those ports. I do not use either PS/2 port on my current motherboard. And finally, parallel printing in Windows 2000/XP is so ungodly slow, it's no wonder why most printers these days are not coming with a parallel port on them.

There really is no justification these days for the legacy ports.

Fwiw, I work in corporate outsourced IT, and constantly run into oddlball devices that plug into serial/parallel ports (less so on the serial side). I think that for now, mobo makers take for granted that someone who needs more than 10 (6x rear, 4x aux/front) USB devices might be using a powered-usb hub. Better to have USB+Legacy and reach more potential customers than cutoff even a few percent.

Also, some old parallel laser printers still rock I have a 16 year old IBM Laserprinter E that I got for free that prints great B&W for invoices, etc.

 

[myocardia]

Originally posted by: SunnyD
Why... WHY in the name of all things sane did they put PS/2 ports and a Parallel port on the damn thing?

Hmm, maybe they didn't want to alienate all of the power users that already have more than one PC, and have them linked with PS/2 KVM's?

 

[lopri]

Overclocking this beast won't be easy. (My guess) I don't remember where I've seen it but the diagram clearly showed only one CPU accessing NB. That is,

CPU1/Memory <----> CPU2/Memory <----> NB <----> SB

*<---->: HyperTransport

Excuse my bad drawing. According to AMD PowerPoint slide, it looked as if each CPU has exclusive memory pool, and CPU1-CPU2 are connected via HyperTransport. CPU2 behaves as if a gateway to chipset for CPU1 in that CPU2 only has access to NB, of course via HyperTransport.

Again this is my translation but the diagram I saw in the presentation is correct. If my translation is correct, the platform can't be called 'NUMA' (Non-Uniform Memory Access) because a CPU will only have an access to half the memory. (Assuming a user is filling all 4 slots) As I mentioned at the beginning, overclocking such configuration will be very interesting. I'm guessing there will be one common clock generator for both CPUs?

 

[Hard Ball]

Originally posted by: lopri
Overclocking this beast won't be easy. (My guess) I don't remember where I've seen it but the diagram clearly showed only one CPU accessing NB.

Keep in mind, these are FX chips, with the multipliers unlocked; so there usually shouldn't be the need for a large HTT freq increase to OC; not that there is really that much more head room, since these babies are already bumping 125W. For some serious OC, you better get some exotic cooling.

That is,

CPU1/Memory <----> CPU2/Memory <----> NB <----> SB

*<---->: HyperTransport

That diagram is incorrect. According to NV, there are 2 X MCP on the platform, and each is individually linked to CPU in Socket0 through its own HTT link. And technically, there is no NB on an AMD64 platform, since most of the NB functions (SRQ, cross-bar, mem control) now reside on CPU die.

Excuse my bad drawing. According to AMD PowerPoint slide, it looked as if each CPU has exclusive memory pool, and CPU1-CPU2 are connected via HyperTransport. CPU2 behaves as if a gateway to chipset for CPU1 in that CPU2 only has access to NB, of course via HyperTransport.

That is basically correct, that each CPU has its own physical memory accessed by its on die components; this is the same with any DP or MP Opteron setup as well.

Again this is my translation but the diagram I saw in the presentation is correct. If my translation is correct, the platform can't be called 'NUMA' (Non-Uniform Memory Access) because a CPU will only have an access to half the memory. (Assuming a user is filling all 4 slots) As I mentioned at the beginning, overclocking such configuration will be very interesting. I'm guessing there will be one common clock generator for both CPUs?

However, there seems to be a fundamental misunderstanding of what NUMA is. The non-uniformity refers to the amount of latency caused by the number of hops over the system interconnect; as the number of sockets in the system scales, so does the max number of hops of HTT from a core to the physical location that corresponds to the location in address space; so access different locations may result in varying amount of latencies for ld/st ops, from the CPU's local memory to an on-die NB that may be as far as 3 hops away on the interconnect. As long as the connection between the sockets are ccHTT, and provided that the OS is NUMA aware, then the system is NUMA capable.