The GPU keeps being leveraged in more areas, but you're correct in that integrated would most likely be fine for things that aren't heavily reliant on it. I haven't checked if Intel came through on OpenCL support there.
OpenCL 1.1 come with Ivy Bridge's HD4000. ( OpenCL 1.2 is most recent version. We will have to wait for Haswell update but probably on track). It will probably work OK for very limited size problems. The "Last Level" cache is 8MB, but will be easy to overwhelm the two memory controllers in the package if the x86 are also concurrently making large requests of the same RAM. It is shared but the adresses aren't really shared so there will be copying.
They are OpenGL 3.0 which brings them into line with OS X's advancement along that line too.
Wasn't it switching to PCIe 3.0? They should be able to run it over fewer lanes like that, or am I missing something in the way it works?
Conceptually yes. if some users just need to "run in place" in terms of bandwidth usage. Two GPU cards in two 8x v3.0 slots will see the same bandwidth as two GPU cards in two 16x v2.0 slots. This runs in conflict with the first point above though. If transferring a relatively large amount of single/double float data to GPU card for number crunching then this doubles transfer times.
If chasing gamers (and perhaps those with very high SAN I/O cards ) then a configuration that splits the 16 lanes is a better fit. There is already evidence that games are partially tweaked to deal with split 8x v2.0 bandwidth constraints now (i.e., are optimized to this lowest common denominator class of machines). If looking to exploit GP-GPU then it doesn't. Again it is a design priority issue.
It wouldn't be so bad if they could get one of the cheapest 6 cores into the base model, but I don't think that will happen as it costs more than the one currently in use there.
The cheapest appropriate six core is the E5-1650 ($583). The trade-off here is that for the "limited core, GHz chasers" it is more expensive and slower than the 1620 ($294). The other disconnect though is that currently the 6 core is $3699. By introducing a 6 core middle ("better") offering 6 cores would drop to the mid price; $2899 . [ assuming they stick with same pricing increments.] That is a $800 drop and below the $3,000 price barrier. More than a few customers are going to see that as a value proposition increase.
I highly doubt Apple is going shave $289 out of its margins to push the 1650 at 1620 prices. The other huge problem is that there are only 3 E5 1600 offerings. If Apple is committed to doing a "good, better, best" line up then there aren't three 6 core offerings to peddle. You will probably have to wait for Ivy BRidge E5 1600's which may get an 8 core at the top, so the bottom entry may also bump to 6 without loosing too much base clock speed (e.g., slides back to 3.2GHz but still tops out at 3.8 or 3.9 in Turbo).
I suspect you have the same 4,6,6 offering but substantively speed bumped.
We had a discussion about this a while back. You responded with an essay about dongle farms hanging off the back of macs
.
Substantially increasing the TB devices hanging outside of a Mac Pro (or Mac Pro lite ) is what would increase the dongle farms. My point was that most likely many Mac Pro users would probably leverage PCI-e cards and connectivity they already have and
reduce the amount of external boxes hanging outside their box that would incur with more TB usage.
In many cases, the TB device is just an intermediary to a set of legacy plug(s). The Mac Pro already has cards with connect to those plugs. For example, here are six products ATTO is rolling out:
Thunderlink
http://www.attotech.com/products/category.php?id=15&catid=16
ThunderStream
http://www.attotech.com/products/category.php?id=15&catid=17
Every single one of those involves plugging in non-Thunderbolt cable(s) which Mac Pro's take now (either on Mac Pro's edge or on a PCI-e card edge). For those who already have those connections with PCI-e cards, they can likely just move that card to a new Mac Pro even if it doesn't have Thunderbolt. [ A "Mac Pro lite" would be a different story since slots were being dramatically cut back. ]
Of the TB devices on the market the large majority of them have sockets to other older cables on them. It is actually the minority of devices that solely have one or two TB connectors (and possible power connector ) on them.
If toss the "corner case" of RAID card + SATA drives integrated TB boxes the count drops to practically zero. Even Blackmagic's heavily TB weighted Cinema Camera doesn't fit that "is only connector" model:
http://blackmagic-design.com/products/blackmagiccinemacamera/design/
It has SDI out also. And the storage drive is removable.
It would allow recent thunderbolt type peripherals to be shared between mac pro when in the office and macbook pro when away. It's most likely a small market, but obviously the major criticism of the mac pro on here is that the whole line represents a small market. The only thing that annoys me is the speculative sales figures.
5% (double up crowd) of even 10% (over estimation of Mac Pro share) isn't even 1%.
For more sense IMHO would be something like
Mac Pro lite w/Thunderbolt $1900-2500
Mac Pro single package (no Thunderbolt) $2600-3600
Mac Pro dual package (with Thunderbolt) $3700-6000
than to saddle the whole line-up with TB. The "lite" version would have no 4x slots ( only the 1-2 slots discussed above) so would "need" TB for smaller card like expansion. The dual package version has 80 PCI-e lanes should could plug a embedded GPU (like the daughter card that goes into a iMac) into the model with enough lanes.
If Intel eventually (Haswell timeframe ?) adds a small iGPU to the E5 class offerings then the others would join the Thunderbolt club. Similarly if Haswell slightly cranked up the PCI-e line count to 48 then a embedded GPU could be added to motherboards with minimal problems to align with TB design expectations.
Frankly more folks who need to buy a "Mac Pro + MacBook Pro" combo are going to go with the "lite" version anyway because it is more cost effective. They will use the money shaved off the full Mac Pro price to help buy (or pay off) the laptop.
