How is RAM frequency controlled in 2009 Mac Pros with QPI?

[nanofrog]

So the W5690 is supposed to top out at 3,66 GHz without turbo boost.

I hope this is the part to go for $1400USD quantity pricing that was mentioned previously (wiki last I checked; amount, but no P/N, clocks,... associated with it). Otherwise the cost on the Gulftowns is really going to hurt.

I'm looking forward to see what Apple will do to castrate this new chip. Continue four DIMMs/socket and keep locking the multiplier at 8 instead of 12?

I'd expect Apple to either increase, or unlock the multiplier (go SPD in the next). I'm basing this on the fact that for the same $$$, Gulftown will have a lower clock. Granted, there's no 3.66GHz part at all right now, but Apple doesn't use the fastest parts anyway.

So if $284 (quantity pricing) gets a Quad @ 2.66, I'm thinking maybe 2.4GHz in the SP Gulftowns for example. Obviously you get an extra pair of cores. But by allowing the memory bandwidth to be better utilized, they can advertise the new systems are faster than the Nehalem based systems, despite the lower clocks per model point (base, mid, high). That's where locking the memory multiplier makes sense in the current models. Otherwise there's not really any incentive for users to upgrade their systems beyond core count (or they'd be looking for Nehalems, just as the '08's are desired now). The performance/price ratio is going to drop in terms of the chips for single threaded operation for many users I feel (or other apps that can't take advantage of 12 cores).

Hopefully you get the idea.

 

[Umbongo]

The last information I spotted didn't mention 1600MHz capability (lots of info was still missing; specifically the actual P/N's, and related clocks and pricing), but I'm not surprised (gives the new parts another edge). It's less of an issue these days for most SP systems, compared to previous architectures, as the boards can already OC quite well (well over 1600MHz for some). DP is another story, and quite important. Unless we finally get an enthusiast DP board, but as it's not out yet (or even an announcement - previous units planned where dropped ), I've a feeling it won't.

I'd expect a similar memory speed breakdown over the part line for Gulftowns as it is now (i.e. bottom parts at x, mid can go to y, and the top end z; where x < y <z). Current parts offer 800, 1066, and 1333MHz over the models. Just sub in larger values in the Gulftowns.

The article over at the inquirer suggests the vendors they were talking to (one was probably ASUS) will have overclockable DP boards, with 6 DIMM slots that can be pushed to 2133MHz. They also suggest no low end with the $562 price point being the starting one, which makes sense looking at the current desktop landscape.

Still too early to tell what the range will look like or where Apple would go though. I'd hope if they move to 6 core only then the prices stay the same and consumers get more for their money.

 

[Umbongo]

I'm also wondering if the rumor of Apple getting processors early might be related to Intel planning on launching dual socket processors before single socket, and Apple needing both at the same time to launch a new product.

 

[gugucom]

I'm also wondering if the rumor of Apple getting processors early might be related to Intel planning on launching dual socket processors before single socket, and Apple needing both at the same time to launch a new product.

That would make a lot of sense. The UP never even had a p/n or name assigned.

What will all this do to the product structure? Realistically Apple will simply take 5500/3500 out and put 5600/3600 in? Anything else would just be a nightmare of complexity that they would never even consider.

 

[nanofrog]

The article over at the inquirer suggests the vendors they were talking to (one was probably ASUS) will have overclockable DP boards, with 6 DIMM slots that can be pushed to 2133MHz. They also suggest no low end with the $562 price point being the starting one, which makes sense looking at the current desktop landscape.

What I saw was only the Intel SkullTrail II and a board by ASUS. It went from both as a Yes, to No on both, then No on the Intel, and unknown, but unlikely on the ASUS (across multiple articles over time). I haven't noticed anything recent though. If you've discovered something recent, please feel free to share.

I'd love to see one though, as that's what I really wanted to begin with. Lack of availability stopped me cold though, and I went with an SP system, as a DP board without the ability to OC wasn't financially attractive. It's sufficing, but if the DP boards show up, I'll take a serious look.

Still too early to tell what the range will look like or where Apple would go though. I'd hope if they move to 6 core only then the prices stay the same and consumers get more for their money.

It would be nice, but I'm not that hopeful.

I'm also wondering if the rumor of Apple getting processors early might be related to Intel planning on launching dual socket processors before single socket, and Apple needing both at the same time to launch a new product.

Possible. Intel can actually ship earlier to a single supplier, as they can accumulate stock for the initial orders between the first shipments (Apple's first shipment) to the official release date for the rest of the vendors. The fact is a small group of P/N's helps too. They don't get overwhelmed resulting in supply shortages, and recover funds early. Apple obviously gets to ship faster. Works for both nicely IMO.

Dell and HP for example, don't limit themselves to so few P/N's, so they don't have the purchasing clout. Despite the fact they ship more systems than Apple.

 

[Umbongo]

Sorry nanofrog I did mean to post a link.

http://www.theinquirer.net/inquirer/feature/1562303/intel-core-gulftown-performing

 

[gugucom]

Nice, report. If this is the source for the early MP5,1 rumor I would not be so sure that Apple will actually implement 10G ethernet.

 

[nanofrog]

Sorry nanofrog I did mean to post a link.

http://www.theinquirer.net/inquirer/feature/1562303/intel-core-gulftown-performing

Thanks for posting it.

I hope such boards do ship, and not take forever. I'd like to have one.

Nice, report. If this is the source for the early MP5,1 rumor I would not be so sure that Apple will actually implement 10G ethernet.

Personally, I see 10G on a MP as a waste (added cost for the 10G NIC chip), and especially the cost of the related equipment for most. Definitely not for SOHO use. I'd think there's some interested, and a few that will actually run 10G. But for those few, they can add a card to the system.

I also see it adding difficulties to the PCIe lane configurations or adding cost to increase the lane count on the board. Not a bad thing, but given Apple's pricing and desire to be cheap, it could actually hinder matters (i.e. one of the 4x lane slots becomes an 1x in order to implement the 10G chip without adding lanes). I'd hope not, but possible.

 

[PaulD-UK]

...I see 10G on a MP as a waste... I'd think there's some interested, and a few that will actually run 10G.... ...given Apple's pricing and desire to be cheap...

Hi
Surely 10Gb bandwidth to a peripheral port on the mobo is an essential step on the path to Light Peak?

Lots of Mac users run external kit that will benefit hugely from the (presumable) simplicity of LP.
Also it seems a solution that will fit extremely well with Steve Jobs' perception of Apple's DNA...

 

[nanofrog]

Hi
Surely 10Gb bandwidth to a peripheral port on the mobo is an essential step on the path to Light Peak?

Lots of Mac users run external kit that will benefit hugely from the (presumable) simplicity of LP.
Also it seems a solution that will fit extremely well with Steve Jobs' perception of Apple's DNA...

10G Ethernet /= Light Peak

They each require PCIe lanes to transfer data to the chipset. Eitherway, you really need more lanes than are in the current systems. Adding one or the other (each need 4x Gen 2.0 lanes, as the throughput is 1.25GB/s). But it could allow for one of the slots to become 1x to get that bandwidth for the device. Both would require the addition of more lanes. Period. There's a couple of options, but most likely, they'd have to add a second chipset in a master - slave configuration to do it (the nF200 is also a possiblility, but I seriously doubt Apple would use it, as they wouldn't provide SLI support, and it would result in backlash).

Between the two, Light Peak would benefit more MP users than 10G Ethernet IMO, assuming it's one or the other that will actually show up in the 2010 MP's (assuming the additional chipset is nixed due to cost, preventing the use of both).

I haven't seen solid pricing on the Light Peak chips, but the 10G Ethernet parts are likely to be had for ~$65USD or so in quantity (initially released ~$100USD in 2007 IIRC). Figuring both and the additional chipset, this will add some major costs to the board. Likely to the tune of ~$200USD for the 3 parts. Not inexpensive at all. So pricing could get ugly on the 2010's as a result, and that's before knowing the other aspects of the system (Gulftown pricing Apple obtains from Intel, as it may not line up exactly with the Gainstowns currently used).