Much better is kind of subjective. If you look at benchmarks for the 6-core for example, Geekbench reports it to be 31% faster (why 32-bit!) than the 6-core from 3.5 years ago!
.....
Many would argue that a 31% increase is a disappointing (nay pathetic) advancement in CPU performance for an equivalent model over that length of time,
3 years , roughtly 10% per year. That's pretty good. If their paycheck went up 10% per year over last 3 years not many folks will be yelling 'pathetic' increase.
The x86 archiecture is old as dirt. It has been optimized by very smart folks for that last 20 years. There are magic potholes to remove to get hypergrowth anymore. There are niche problems like SIMD code that is run through the AVX engine now. The SIMD code improvements actually have been use and the 3 year processors are dog slow in comparison. Likewise, stream AES encryption... 3 years old ... it is a dog.
But general across the board super increases on single cores? Not coming. The large jumps are in more cores.
especially when they're charging $4,000 for it (which correct me if I'm wrong is roughly the same price as a 6-core model from 3.5 years ago)?
It is a bit more. But then that would have been 3GB of RAM versus 12GB now. Similar on storage drive performance. These systems aren't solely or even dominately driven by CPU price at the lower standard config end of the spectrum.
Plus, while 7 TeraFLOPS sounds great, you'll only get that on on the top end 12-core model. We're yet to see how the lesser cards stack up?
Actually not. the 7 TFLOPs is purely the D700 all by themselves. It doesn't necessarily require 12 x86 cores to keep those cards feed with data. In fact, given all three are sharing the same power supply and fan 12 cores is going to draw resources away from the cards also.
Secondly, Only no new information if you haven't bothered to look at Apple's web pages. Apple lists the TFLOPs for each single GPU card they are offering.
http://www.apple.com/mac-pro/specs/
D300 2TFLOPs/card ( 4 TFLOPs )
D500 2.2TFOPs/card ( 4.4 TFLOPs )
D700 3.5TFLOPs/card ( 7 TFLOPs )
All of three of those individual cards are better than even the 12 core Xeon on single precision (non especially AVX ) performance, let alone the fewer core models.
What does remain to be measured though is not these peak rates but the sustained. The dual card numbers are likely more theoretical than real. ( due to power/thermal constraints as well code constraints )
Since Apple has moved to a single-socket design, you'll see dual-socket Xeon workstations that smash these new Pros for roughly the same price soon enough.
Just with Xeons and no/weak GPGPUs??? Only on x86 only code. With same GPUs will give up lots of base clock to keep the price the same.
It would've been good if they could price the 8-core at the 6 and the 6 at the 4 but I understand a lot of R&D went into it and it sure looks good. I appreciate quiet computers too!
The 8 relative 6 pricing is primarily driven by Intel not Apple. There is a substantive chunk added by Apple to both, but nothing like what Intel is adding.
E5 1650 v2 $583 ( 6 cores 12MB L3 )
E5 1680 v2 $1723 ( 8 cores 25MB L3 )
$1,140 difference. So about $500/core. ( there is more L3 cache too ).
The gap here is largely because the E5 1680 is priced like a 1600 series but far more like an E5 2600 processor. In fact there is a 10 core model with similar base clock speeds and not quite as large dynamic Turbo range that is exactly the same price. This is basically that 10 core model with two cores permanently flipped off to support the increased dynamic clock range. So pay for them anyway.
There are more affordable 8 core models but you'll loose very significant base clock speed. The extremely high price is because getting both more cores and relatively high clock for that many cores.
In about 2 more iterations v3 ( Haswell updated arch ) v4 ( Broadwell shrink ), Intel probably will have 8 core priced at 1650 6 core prices. But that is in the future.
