That link had no Adobe CS tests in it, do you know of any overclocked I7's that do?
Barefeats shows these results for the 2.93 xeons:
http://www.barefeats.com/mbpp23.html
You'd have to search that out. Keep in mind however, most, if not all that you'll find are the PC versions of the software (unless someone that uses CS5 on a hacked OC'ed PC has posted results for OS X versions).
Why? it is 2010. I was looking across the 2010 Mac line up.
And the current systems are 2009.
The point is, that the current models shifted from previous Intel based systems (% MSRP spent on CPU/s).
Given they're using cheaper processors, and the prices either remained the same (Quad) or increased (Octad), Apple's offering less value than the previous models. As a result, some OS X users are having a fit (valid for independent pros from their perspective, and somewhat justifiably so IMO).
The biggest differences are with systems using the W3520 (PC vendors), as they can be had for ~$1000 or so less last I checked. The cost difference is smaller on the Octad systems (2x E5520), and if you push the clock speeds in their respective models (SP and DP systems).
The ark site at Intel says $797 but that isn't the point.
That's the current price, which has dropped from it's initial release. But as you say, it's not really an issue.
I rather doubt they got them for much under the $800 mark though.
I understand what you're trying to get at with leverage in negotiations, but I'm sure Intel was well aware that the PPC days had ended. Where I do expect they had negotation leverage, was because Apple used Intel as their ODM for the system board as well in the '06 - '08 systems, and bought enough quantity that it made discounts possible.
The more I think about this, the more I'm thinking they shifted to Hon Hai Precision as the ODM for the current models' boards. Which resulted in paying closer to the published Quantity pricing for the existing CPU's used.
No the "can't/won't possibly pay more than $1,900 for a computer " and the "I only need a $1,200 mini-tower computer" crowds.
Ah, I see where you're going then, and I agree. The enthusiast users have been left out in the cold with the cost increase, as Apple isn't willing to release an expandable consumer system. The previous MP's just happened to be cheap enough that they could better afford them. With the current line-up, they'd have to deal with another of Apple's systems, bite the bullet and buy the MP, or go back to PC's (hacked or otherwise).
IMHO I think some folks are mind warped by PCs. If just periodically have
batch rendering work to do it is easy to ship that off to some farm ( either local if have steady stream of work to do) or in the cloud. Timesharing (and cost sharing) on a bigger machine ( bigger virtually or physically) often is more cost effective than buying fastest possible machine now that stays underutilized for large segments of the day ( and/or has "future power" that is underutilized for the first couple of years. ) That too may mean a software migration to some extent.
Farming is valid for some, but not those on tigher budgets. In those cases (again, the independent surfaces), they're budget sticks them to a single system.
They just don't have the funds on hand to pay for sufficient bandwidth, let alone timesharing. So building a farm of their own would be completely out of the question. Heck, the licensing costs might scare them to death.
For a growing production house (i.e. successful SMB), it may be another story, assuming there's a sufficient workload, and the funds are available.
No you don't if there is a significant fraction of memory and I/O accesses. Unless you also clock up the memory those remain at the exact same effective speeds. It primarily just works on problems which can get sucked into the in chip package cache. Oracle DB or some large finite element mesh ( weather , crash model, etc.) isn't going to run significantly faster except at defacto idling tasks. The results are application specific.
Most of the OC capable boards do in fact OC the RAM as well as the CPU clock. Particularly with the newer parts (LGA1366), as voltages, multipliers, and BCLK frequencies are accessible in these boards.
Likewise if you are running 8 threads full blast with significant memory streaming you are only going to increase memory I/O pressure by overclocking.
If it's not OC'ed, and the application can utilize the existing memory configuration, then Yes. But as the RAM is usually OC'ed as well, this pressure is alleviated to some extent at least (depending on the specifics, as the scale of RAM increases may or may not be linear with that of the CPU).
If it's balanced, it can benefit SMP. If not, there's going to be issues with data streams (processor uses what it has, then waits for data to continue working; wash, rinse, repeat). What I usually think of as an I/O stutter.
