What I said was someone (else) wanted to install xyz on their G5, and the software req'd Leopard. Worth is relative. So with G5 2.5DP, 5x moving to the base 4-core in processing I would say. Geekbench
2000 for 2.5DP to
8800 pts. Mac Pro 2.66.
It has usually taken 6 months after Intel debuts a new processor, and the whole tic-toc cycle. Often November, and leading to preliminary protypes and first boot followed by test, testing, new compilers, and getting firmware in place.
Tide over, I don't think buying for 4-5 yrs makes sense. Tide you over? so you aren't hit with whatever comes after 10.6 Snow Leopard AND a new socket, firmware and processor. Something to use, and you think it won't do the job? not fast enough? no resale value in 1-2 yrs?
I build a system or buy when I need it not on rumors but I do watch Intel and others roadmap. And sometimes it is "now or wait a year" on some things. Available parts, RAM prices, Intel moving to X68, or that Core i7 D0 Stepping was finally out. And that PC BIOS development took 6-9 months, and Windows 7 wasn't out so that could change things as well.
When Mac Pro came out originally in 2006, it was amazing that it had so few problems, it shipped with three builds in 3-5 weeks of 10.4.7. And 10.4.8 boosted performance in some apps 20%. A full software and compiler takes TIME to master and put in place and figure out, nothing instant, but 20-40% is reasonable to expect - over 12-18 months.
Given the added cost of $2600 to upgrade processors, today, and Gulftowns - take a look at prices - $1795 each. Guess what the BTO will be?
And I still see apps that don't properly deal with hyper-threading and other features.
Some people saw real improvements from a couple quality SSDs than their (2.93 8-core processor). Sure it sounds great to have. And a single socket 6-core Gulttown is faster, over-clocks easily -- but not on Macs, and might make sense, with its 20% improvement, clock for clock, and not need a dual processor.
My money is that if it came out - maybe in the fall, and when Nvidia GTX 470s (and would require modification for more aux power with 8-pin) but then have to redo another motherboard soon after. Plausible. But, rule of thumb, wait for second silicon and 'stepping' not the first errata and revision, and for bugs to get worked out. In 3-4 months extra. And for 10.6 to advance at least one or two points. Those are learned rules of thumb.
Want to have the fastest GTO on the block for awhile?
The low end model you are laughing or scoffing at?
you haven't test driven it or your apps.
CS5 users seem happy.
http://macperformanceguide.com/OptimizingPhotoshopCS5-Issues.html
In fact, DDR3 helps with much better bandwidth.
Gulftown needs triple channel or quad to help feed data to memory, and from larger L3 caches.
I don't see people buying into large fast scratch arrays as much, and fewer need 32GB.
People use to spend $2000 and up on SCSI arrays for scratch.
8GB DIMMs help, memory matters. And an 8GB DIMM for $380. People use to buy duals JUST for the 8 DIMM slots, just so they could upgrade to 16-32-48GB RAM, in the past. Put money into RAM and SSDs. And sacrifice some going with slower processor (but what could you do? $1200 extra to make up, $2400 + RAM
And you could still build or have 3 base models for your one dual Gulftown, or dual 2.93 today. And those that bought dual 2.93s? they don't think they are even pushing to the limits, and are held back by software. 100% on a core is NOT being efficent, it is code that doesn't work properly.
The audio bug 2009 caused lost processor performance, the temps to hit 85C and above with little work being done, a real real mess and signs someone was sleeping and never tested for real world conditions. Got fixed but not until November/Dec long after shipping, and not for those that weren't ready to move to 10.6. Because not everything ever is. SoftRAID 4 just came out with their 32-bit/64-bit kernel mode driver.
A 64-bit kernel mode does help CS5, enough that the cost of hardware to get the same.
Mainframes use to consider a 20% boost in processing to justify an upgrade, and mainframes unlike other servers, tend to run above 80-85% utilization at the least. Few servers and fewer desktop users can say that.
The experience of many is that it takes a couple months to upgrade and get settled in and have RAM, hard drives, RAID, backup strategy and everything installed and customized and transferred over once they make the move. Not over-night thing.
Graphics. Takes time to get good drivers. And even Windows ain't always pretty and definitely takes a long time, and room for improvement. And in that area, and OpenCL look for real returns in what, a year. Same with old technology with new names to mature, like Grand Central.
I think DLloyd says it best:
In the meantime, if a new or refurbished Mac Pro saves you time and/or increases reliability now, consider an MPG Photo Workstation, taking into account the quad-core vs 8-core shootout. My top recommendation remains the 2.66Ghz quad-core “Burly” MPG Photo Workstation, with the 3.33GHz model a smart move for increased interactive responsiveness.
http://macperformanceguide.com/blog/2010/20100607_MacPro--macpro-news.html
http://macperformanceguide.com/Shootout-MacPro-Intro.html