The Bloomfield ones are ancient as it is. The Extreme Gulftown desktop processors included a tower cooler from Intel. Now they are not even going to supply one. AMD is considering water cooling as well for Bulldozer.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Mac Pro 2011 "Sandy Bridge E" everything we know
- Thread starter Amethyst
- Start date
- Sort by reaction score
Yup, and you can over clock too.
I don't know why people are worrying about this tbh.
If anything, we'll have to cope with slightly higher fan speeds if all things are equal.
Well, this is assuming the form factor remains the same. If Apple starts mucking with case design this will obviously affect things.
Exactly. CPU bump and GPU bump are the main things. TB is essentially useless for desktop users, save a sparse few. Such as those that need to share TB peripherals for laptops used for field work as a means of reducing equipment costs for example.
The main issue to me, is if they release the GPU cards with TB only due to the necessitation of adapters to work with 3rd party monitors.
Imagine needing a TB to MDP adapter, then MDP to whatever port the user's monitor uses. It could get expensive, particularly for those that exceed 1920x1200 resolution (i.e. need Dual Link DVI, as a MDP to DL DVI adapter is $100 last I checked). Not exactly cheap (it's an active adapter = has circuits to it, not just wire and connector ends). But at least this approach would allow those that already have MDP to whatever adapter versions to use their existing monitors can recycle what they have (just need TB to MDP).
But it's also possible that new adapters would be TB to the interconnect needed by the monitor, such as DVI. Which could force those that bought the existing MDP to DVI or MDP to DL-DVI to buy another adapter if there isn't a 3rd party adapter made available by companies such as Monoprice. Some would be rather upset if a situation like this will exist, and I wouldn't blame them one bit.
See above.
There's a potential for a real mess for those that need 3rd party monitors, or don't want to replace their current MDP equipped Apple Displays IMO.
Exactly, and I'd be shocked to see them keep both models running simultaneously (single line reduces costs on their end by increasing the economy of scale). So IMO, they will EOL the current models once the TB versions are shipping.
Yes.
And the SB-E5's will be able to be air cooled as well, even if the cooler needs to be designed to dissipate 200W. Adequate airflow would be needed, but it's possible to do. Thin and small would be dangerous for this much heat, but a normal sized workstation case could allow for enough airflow to accommodate an air cooling solution. Good baffle design goes a long way in such situations, and Apple has done a decent job with that so far.
Take a look at existing 6 pipe based coolers to get an idea of physical size of the coolers that would be needed to handle it. Which means for the general PC market at least, some current performance oriented coolers by makers such as Nocuta, Thermaltake,... should be able to handle it, thus only needing a new base plate to fit the new socket. I'm not saying they'd be able to do much overclocking, but they'd certainly be able to handle the LGA2011's in their stock clocks IMO.
Xeon E5-2600 Listed
Dual Core
Xeon E5-2637 / 2 Core (4 Threads) / 3.0 GHz / L3 5 MB / DDR3-1600 / TDP 80 Watt
Quad Core
Xeon E5-2603 / 4 Core (4 Threads) / 1.8 GHz / L3 10 MB / DDR3-1066 / TDP 80 Watt
Xeon E5-2609 / 4 Core (4 Threads) / 2.4 GHz / L3 10 MB / DDR3-1066 / TDP 80 Watt
Xeon E5-2643 / 4 Core (8 Threads) / 3.3 GHz / L3 10 MB / DDR3-1600 / TDP 130 Watt
6 Core
Xeon E5-2620 / 6 Core (12 Threads) / 2.0 GHz / L3 15 MB / DDR3-1333 / TDP 95 Watt
Xeon E5-2630 / 6 Core (12 Threads) / 2.3 GHz / L3 15 MB / DDR3-1333 / TDP 95 Watt
Xeon E5-2640 / 6 Core (12 Threads) / 2.5 GHz / L3 15 MB / DDR3-1333 / TDP 95 Watt
Xeon E5-2667 / 6 Core (12 Threads) / 2.9 GHz / L3 15 MB / DDR3-1600 / TDP 130 Watt
Xeon E5-2630L / 6 Core (12 Threads) / 2.0 GHz / L3 15 MB / DDR3-1333 / TDP 60 Watt
8 Core
Xeon E5-2650 / 8 Core (16 Threads) / 2.0 GHz / L3 20 MB / DDR3-1600 / TDP 95 Watt
Xeon E5-2660 / 8 Core (16 Threads) / 2.2 GHz / L3 20 MB / DDR3-1600 / TDP 95 Watt
Xeon E5-2665 / 8 Core (16 Threads) / 2.4 GHz / L3 20 MB / DDR3-1600 / TDP 115 Watt
Xeon E5-2670 / 8 Core (16 Threads) / 2.6 GHz / L3 20 MB / DDR3-1600 / TDP 115 Watt
Xeon E5-2680 / 8 Core (16 Threads) / 2.7 GHz / L3 20 MB / DDR3-1600 / TDP 130 Watt
Xeon E5-2690 / 8 Core (16 Threads) / 2.9 GHz / L3 20 MB / DDR3-1600 / TDP 135 Watt
Xeon E5-2687W / 8 Core (16 Threads) / 3.1 GHz / L3 20 MB / DDR3-1600 / TDP 150 Watt
Xeon E5-2650L / 8 Core (16 Threads) / 1.8 GHz / L3 20 MB / DDR3-1600 / TDP 70 Watt
Source: http://www.tomshardware.com/news/sandy-bridge-e-xeon-cpu-servers,13308.html
Dual Core
Xeon E5-2637 / 2 Core (4 Threads) / 3.0 GHz / L3 5 MB / DDR3-1600 / TDP 80 Watt
Quad Core
Xeon E5-2603 / 4 Core (4 Threads) / 1.8 GHz / L3 10 MB / DDR3-1066 / TDP 80 Watt
Xeon E5-2609 / 4 Core (4 Threads) / 2.4 GHz / L3 10 MB / DDR3-1066 / TDP 80 Watt
Xeon E5-2643 / 4 Core (8 Threads) / 3.3 GHz / L3 10 MB / DDR3-1600 / TDP 130 Watt
6 Core
Xeon E5-2620 / 6 Core (12 Threads) / 2.0 GHz / L3 15 MB / DDR3-1333 / TDP 95 Watt
Xeon E5-2630 / 6 Core (12 Threads) / 2.3 GHz / L3 15 MB / DDR3-1333 / TDP 95 Watt
Xeon E5-2640 / 6 Core (12 Threads) / 2.5 GHz / L3 15 MB / DDR3-1333 / TDP 95 Watt
Xeon E5-2667 / 6 Core (12 Threads) / 2.9 GHz / L3 15 MB / DDR3-1600 / TDP 130 Watt
Xeon E5-2630L / 6 Core (12 Threads) / 2.0 GHz / L3 15 MB / DDR3-1333 / TDP 60 Watt
8 Core
Xeon E5-2650 / 8 Core (16 Threads) / 2.0 GHz / L3 20 MB / DDR3-1600 / TDP 95 Watt
Xeon E5-2660 / 8 Core (16 Threads) / 2.2 GHz / L3 20 MB / DDR3-1600 / TDP 95 Watt
Xeon E5-2665 / 8 Core (16 Threads) / 2.4 GHz / L3 20 MB / DDR3-1600 / TDP 115 Watt
Xeon E5-2670 / 8 Core (16 Threads) / 2.6 GHz / L3 20 MB / DDR3-1600 / TDP 115 Watt
Xeon E5-2680 / 8 Core (16 Threads) / 2.7 GHz / L3 20 MB / DDR3-1600 / TDP 130 Watt
Xeon E5-2690 / 8 Core (16 Threads) / 2.9 GHz / L3 20 MB / DDR3-1600 / TDP 135 Watt
Xeon E5-2687W / 8 Core (16 Threads) / 3.1 GHz / L3 20 MB / DDR3-1600 / TDP 150 Watt
Xeon E5-2650L / 8 Core (16 Threads) / 1.8 GHz / L3 20 MB / DDR3-1600 / TDP 70 Watt
Source: http://www.tomshardware.com/news/sandy-bridge-e-xeon-cpu-servers,13308.html
Last edited:
The Xeon E5-2687W looks to be the new King of the Beasts. Will be interesting to see how it compares to the present leader of the Mac Pro pack.
If Apple takes the same model numbers because similar in price (and mostly the same in TDP) then .... the dual package line up would be:
Apple didn't pick the "king of beasts" in the 5600 series ( 5690 ) and even more doubtful now. Two 150W beasts are 300W all by themselves. Just not worth the drama in power/heat. Second, just not worth the drama in pocketbook.... That's likely a $1,700+ processor. Throw Apple 30+% mark up on top and that around $4,000 in sticker price just in CPUs.
It is going to be a challege with the 115W E5 2670. Some of that is a swap of power from the support chip (since PCI-e v3.0 controller onboard now). So it already had to be dissapated from the CPU/RAM daughtercard anyway. Likewise if the "second" E5 isn't doing any PCI-e dutiies should be to fly under that TDP. However, there seems to be a net increase as well.
If the price isn't too far off may see E5 2630 since that is only a small step back from the 2.4 that the dual systems uses now. While shouldn't at this level, some folks buy on "GHz", but they do so I'd suspect Intel will price that out of contention. So around $3,400 for a 12 core 2.0GHz (baseline) box.
Ironically, the king of baseline GHz may end up being the entry level single package model with a E5-1620 ( at 3.6GHz). Probably would make the non commercial gamer crowd happier since they are usually priced capped in that entry zone anyway. Actually, all of the E5 1620-1660 models would clock higher than the E5 2670, but be capped under 8 cores. Folks who don't need throughput/torque would get the single package models. Folks with the parallelized workloads will drift into the Dual package models. So they are more natually segmented markets. Can still share the same basic case to lower development costs (and help ensure survivability) for both sub-markets.
GHz doesn't equal aggregate throughput. Adding 2-4 more cores will make a differnce in most workloads that can be parallelized. [ and crufty software that can't chop up the work .... may be time to chuck it. yeah software lags leveraging bleeding edge hardware, but sometimes it is just bad developers or dead software that can't keep up. ]
Apple didn't pick the "king of beasts" in the 5600 series ( 5690 ) and even more doubtful now. Two 150W beasts are 300W all by themselves. Just not worth the drama in power/heat. Second, just not worth the drama in pocketbook.... That's likely a $1,700+ processor. Throw Apple 30+% mark up on top and that around $4,000 in sticker price just in CPUs.
It is going to be a challege with the 115W E5 2670. Some of that is a swap of power from the support chip (since PCI-e v3.0 controller onboard now). So it already had to be dissapated from the CPU/RAM daughtercard anyway. Likewise if the "second" E5 isn't doing any PCI-e dutiies should be to fly under that TDP. However, there seems to be a net increase as well.
If the price isn't too far off may see E5 2630 since that is only a small step back from the 2.4 that the dual systems uses now. While shouldn't at this level, some folks buy on "GHz", but they do so I'd suspect Intel will price that out of contention. So around $3,400 for a 12 core 2.0GHz (baseline) box.
Ironically, the king of baseline GHz may end up being the entry level single package model with a E5-1620 ( at 3.6GHz). Probably would make the non commercial gamer crowd happier since they are usually priced capped in that entry zone anyway. Actually, all of the E5 1620-1660 models would clock higher than the E5 2670, but be capped under 8 cores. Folks who don't need throughput/torque would get the single package models. Folks with the parallelized workloads will drift into the Dual package models. So they are more natually segmented markets. Can still share the same basic case to lower development costs (and help ensure survivability) for both sub-markets.
GHz doesn't equal aggregate throughput. Adding 2-4 more cores will make a differnce in most workloads that can be parallelized. [ and crufty software that can't chop up the work .... may be time to chuck it. yeah software lags leveraging bleeding edge hardware, but sometimes it is just bad developers or dead software that can't keep up. ]
Last edited:
The dual-link dvi adapter is a real problem when you look at the Apple Store reviews.
I am hoping for a real solution, not some adapter-ism.
If you are serios about image quality you need a NEC, Eizo, HP Dreamcolor.
When Apple only offers consumer displays it should provide decent display connectivity (I'd love to see an updated version of the "real" cinema displays, but that's not going to happen anytime soon).
You're right with the model numbers but let's wait and see what the prices are. Hopefully we'll see GHz closer to what we currently have since that still what so many people look at when trying to quantify speed for themselves. A dual 2.6 will still be much faster than the previous dual 2.93 given its 15-20% speed increase clock for clock plus the 4 extra cores and the huge L3 cache!
That said, I'm not too optimistic about apple choosing the 2690 or the 2687W but you never know! Maybe we'll get the 2680 at 2.7 since they picked the 3rd fastest last time. The 2687E may not be the most expensive after all since it's just the 2680 with a higher wattage, right?
Last edited:
Their avg. gross margin is higher than that (tad over 41% last I checked). Take into account the E5-2687W (x2) @ $1700 per with that much margin, and it's nearly $4800 USD.
Figure in the rest of the odds and ends to build a system, and it will cause nosebleeds for a lot of MP users IMO.Definitely not pretty, so I agree that they'll use lower clocked parts for cost reasons alone (heat too).
In terms of heat in an SP model, the 130W rated part should be doable (still 60W less than current DP systems <95W per>), which I gather you'd agree with. It would also make for a really nice Quad (particularly useful for software that can't utilize multiple cores very well).
I realize the adapter method isn't ideal, but I'm concerned that it's the approach Apple will take as it's additional income, as well as fits their design methodology of late.
Ultimately, users will have to use whatever solution is offered to attach to DVI monitors (such as the professional units you mention).
The E5s don't wow me in any way as far as workstation usage goes. Looks like they will provide the typical Intel tick-tock 15% performance increase - at least for creative types. Don't see anything there that should make 2009 or 2010 owners feel too jealous of. I guess the turboboost capabilities might be really high.
I'm definitely interested to see what Apple do with those speeds though.
2.90GHz 12-core and 2.90GHz 16-core would look nice on their buy page, and with rumours of a new enclosure and having used 120W and 150W TDP CPUs before it is a possibility. Not one I'd put money on, but a possibility none the less.
I'm definitely interested to see what Apple do with those speeds though.
2.90GHz 12-core and 2.90GHz 16-core would look nice on their buy page, and with rumours of a new enclosure and having used 120W and 150W TDP CPUs before it is a possibility. Not one I'd put money on, but a possibility none the less.
What, you expected miracles?!?!
The 15% increase or so is all that usually occurs (same software/usage pattern and storage systems), unless what's being used can truly leverage all of the cores.
The one thing that's overlooked IMO, though has limits right now, is the further push to increase I/O (PCIe 3.0 controller on the die, additional cache, additional memory channel, and doubling of the interconnect bandwidth between the PCH/chipset and CPU, ... sorts of upgrades from previous architecture). Quite welcome changes for high I/O throughput usage (clusters for example would truly reap these rewards).
What, you expected miracles?!?!
The 15% increase or so is all that usually occurs (same software/usage pattern and storage systems), unless what's being used can truly leverage all of the cores.
. I think there was some hope as the LGA 1155 parts are so much better than their predecessors. I suppose if AMD had challenging parts things might be clocked a little higher.As an overall platform I think the improvements are good and are certainly going to be welcomed in enterprise.
I see it as equivalent to the performance difference for LGA1155 and the LGA1156 socket (one gen prior) for the same usage pattern (mostly single threaded). In terms of multi-threaded (not limited to a fixed core count), the performance difference vs. LGA1366 will be more significant (extra pair of cores per die).
As an overall platform I think the improvements are good and are certainly going to be welcomed in enterprise.
Unfortunately, they're not going to make what amounts to totally new architecture from the ground up between one socket to the next, as it's too expensive (takes too long = man hours get crazy). It would also mean longer development cycles, which some are already complaining about the current development times for enterprise parts.
IMO. it hard to consider to buy DP mac pro if apple use same price,tdp chip lineup like this.
sp
3.6 quad
3.2 6core
3.3 6core
dp
2.0 12core
2.0 16core
2.6 16core
at 6core and 12 core yes you can get x2 on core numbers. but at 1.2-1.3 ghz different it hard to tell that 12core will be faster.
it better for apple to use 2.5 ghz 6core in entry dp.
sp
3.6 quad
3.2 6core
3.3 6core
dp
2.0 12core
2.0 16core
2.6 16core
at 6core and 12 core yes you can get x2 on core numbers. but at 1.2-1.3 ghz different it hard to tell that 12core will be faster.
it better for apple to use 2.5 ghz 6core in entry dp.
Last edited:
The 12-core may not be much faster but this is the case with current 8-core as well. It's slower than the 3.33GHz 6-core in most tasks, even in heavily multithreaded ones. 16-core, on the other hand, should be quite a lot faster than the 6-core (in multithreaded tasks of course, the 6-core is going to get the crown for single-threaded anyway).
DP system supports more RAM too.
Wait to compare 6 cores to 6 turbo'ed core ( with 6-10 sleeping). In the Dual Package system, you could have two packages each running with just 3 active ( or a 2 / 4 split , i.e., OS + fork-join work controller / worker cores split . ).
For example, it would not be surprising if the E5 2620 turbo'ed up to around 2.1-2.2GHz when just 4 cores were running ( making it a bit faster than a 4 core 5620 running full tilt at 2.4GHz. There is better internal arch and I/O in Sandy Bridge so don't have to match cycles exactly. ). But yes, a 5620 running a single thread 90% of the time and a E5 2620 90% is probably a loss for the E5. However, in the larger system viewpoint, buying any Dual Package system when you are going to run single thread 90% of the time is buying the wrong system product. That user either needs some software that has made it into the 21st century in design or a single package system with fewer cores than Intel and AMD generally sell. If you really only need one core than being capped at 4 is plenty.
If what folks really need is a single package system there is nothing wrong with selling them the right ( aligned with their needs) system at a lower price with the same margin percentages. Apple (nor Intel) doesn't need to "pull" people into the Dual Package systems. That just leads to need/solution mismatches.
Not if it drives up the price of the entry Dual Package system. $3,400 is already a bit high. Benchmarks on with software like FCP X are going to highlight the advantages of having more cores at your disposal. FCP X lends itself to taking on multiple tasks for the user ( edit right now and background transform material ). Even if there are single threaded blocks in a program, the issue that really matters is working on multiple "to do list" items on behalf of the user. The software either takes one "item" and breaks it up into multiple pieces or task mulitple "items" and works on them all that the same time. There are lots of other applications out there that have extensive parallizable blocks of work in them. Having twice as many cores will make a difference as long as don't starve the I/O requests to the packages.
Last edited:
They could even do a single E5 2687W ( 150W is still less than 190W ). However, other than a "pimp my ride" exercise not sure what bang-per-buck Apple is going to get out of it. (some user couple years down road could take used/cheap one and get a 8-core model as an off spec upgrade, but it still be 4-8 cores behind a real dual package system. ) Unlike, the 3500/5600 daughterboards, I suspect the E5 2600's will work in the single socket boards. ( and the 1600s probably might work in double socket ones if leave one socket empty. ). Part of the difference was high bandwidth PCI-e chipsets, but that is all on die now. A firmware sanity check test probably will keep them apart.
the methodology is really to push the standards that Apple likes as much, if not more, than additional income. Similar to banning Flash to help push HTML5+Javascript into maturity and adaptation faster. The adapters just mean there is now big downside when that push doesn't work as well as they hope.
There are Ezio's with MiniDPs and even more with DP. Likewise with the NECs and HP's. It is only up in the 29+" range or displays targeted to legacy or TV environments where the they trade DP for HMDI or older connectors.
Over time the DVI ports are going to disappear. However, Apple could leave a single dual-link DVI on the video cards even if went to the "hocus pocus" to put TB on. It would just have to be one of the "other" video channels the card was putting out.
Last edited:
No. It is higher wattage and speed. That means fewer are going to pass testing. These are all the same chip design. Just different sub-component parts (cores , cache blocks, QPI links, etc. ) turned on/off because they failed or to meet market segmentation goals. Fewer passes means higher price.
Even if it weren't more expensie making a quiet system is at odds with adding many ten's of Watts of additional power to generate and dissipate.
The could limbo into the power envelope by just using one ( for an 8 core system), but who really needs one of those for $1,000+ more than one with 12 cores ? Or $1,000+ more than one with 6 core but runs singled thread the vast majority of the time?
You can keep noise down by using larger diameter fans ( and more of them) but the Mac Pro is already pretty 'wide'. There is no huge plus for Apple selling workstation that crack the $8,000 range in basic config as a price point. Too expensive and people want you to hold their hand while they do anything because they trying to buy control as much as a product.
The 5690s didn't exist when the 2010 Mac Pro was released.
5680 was the king at the time.
But still, it's not as if Apple has never gone with the top end part at least as a BTO option. They did in the 08s. I have several of them with the x5482 at work. That is a 150w chip and was roughly the same MSRP at the time. If they are refreshing the design they may have some more cooling headroom.
Well if you're right, a 16 core 3.1GHz Sandy Bridge E Mac Pro will totally blow the previous gen out of the water! When the chip prices are revealed we'll have a better idea of which ones Apple might use but we can always remain hopefully/foolishly optimistic
True, but the MSRP on such a system would make it prohibitive for all but those that are very well funded (and can justify the system, or
).I'm not sure what you're referring to, but DP processors work in SP boards for LGA1366 designs (IIRC, I've not seen a user state they placed a DP processor in an Apple SP daughterboard, as the threads that asked about this were always advised against it, due to cost reasons).
Given the block diagrams I've seen, I suspect this may be possible as well (PCH only connects to a single CPU). So even if a designer actually choses to use a pair of PCH's, it appears it would be possible, with cross communication occurring via the QPI links between the processors (data located on a drive attached to PCH A needed by CPU B situations, and vice versa).
MDP isn't that common though compared to DVI for professional monitors. Particularly for slightly smaller monitors (24" or so running 1920x1200).
Of course DVI will vanish at some point, but that's not today (or in the near future IMO; just consider how long VGA has been around, and is still with us).
As per DVI-DL as well as TB on an Apple GPU card, it's possible. But if I were a betting person, I wouldn't bet on such a configuration (maybe MDP + TB, as it's very easy to do, and can be done for very low cost - essentially just the MDP connector and a little extra time in PCB layout).
Their processing technology has been very good (very high yields), and is supported by their increasing margins as well as industry articles.
So it's more of a situation of downgrading parts to meet orders than just a means of selling off those that otherwise didn't pass all tests during the binning process from what I've read.
That was then however. Things have changed since then, particularly the negotiation leverage of Apple with Intel on enterprise grade CPU's and ever increasing gross margins (hint #1: they used to get workstation parts early, but that ended with 2009 MP's, and I suspect that was due to the contract dates; hint #2: Intel no longer does the board work for Apple on the MP's).
So as a result, Apple has to go for less expensive parts in to keep the MSRP within a range consumers will take the plunge.
Last edited by a moderator: