Mac Pro 2011 "Sandy Bridge E" everything we know

[wallysb01]

Double it, since there's 16GB DIMM's out. So up to 64GB for the SP systems, and 128GB for the DP units at launch, until 32GB sticks can be had (they were announced some time ago).

Couldn't you just add those 16GB DIMMs to the old Mac Pro too? I guess it might be off-spec, but once Apple Care is gone, that doesn't matter much.

Would be nice to see more slots, but I guess that gets into too many technical details to change easily.

 

[blunti]

Couldn't you just add those 16GB DIMMs to the old Mac Pro too? I guess it might be off-spec, but once Apple Care is gone, that doesn't matter much.

Would be nice to see more slots, but I guess that gets into too many technical details to change easily.

adding 16gig sticks (or any size) will not affect your Apple Care in any way...

 

[wallysb01]

adding 16gig sticks (or any size) will not affect your Apple Care in any way...

Good to know

 

[nanofrog]

Couldn't you just add those 16GB DIMMs to the old Mac Pro too?

Since these are DDR3, they'll only work in 2009, 2010, and newer MP's. So you can add them to certain existing machines.

And as mentioned, this will not affect Apple Care if it's still in effect.

Would be nice to see more slots, but I guess that gets into too many technical details to change easily.

They could address this in the next MP models, but I wouldn't hold my breath (expect them to use 1x DIMM per channel).

Other board makers have been doing this all along, such as the LGA1366 SP boards typically containing 6x DIMM slots on an ATX format. Same with DP versions, but it's possible to exceed that (up to 9x per CPU per Intel spec = 3x DIMM's max via interleaving per memory channel; thus 3x channels * 3x DIMM's = 9 DIMM's per CPU max) on a larger PCB format, such as E-ATX/SSI-EEB. And there are DP server boards with 18x DIMM slots on a single board (not just a paper specification that was never implemented).

 

[wallysb01]

Other board makers have been doing this all along, such as the LGA1366 SP boards typically containing 6x DIMM slots on an ATX format. Same with DP versions, but it's possible to exceed that (up to 9x per CPU per Intel spec = 3x DIMM's max via interleaving per memory channel; thus 3x channels * 3x DIMM's = 9 DIMM's per CPU max) on a larger PCB format, such as E-ATX/SSI-EEB. And there are DP server boards with 18x DIMM slots on a single board (not just a paper specification that was never implemented).

I see. Thanks for the clarification. I would be rather underwhelmed I guess if the Mac Pro keeps the same number of slots. 16GB sticks are still pretty expensive, and it would be nice to be able to get up to 96-128GB with 8GB sticks. Leaving the possibility to get up to 192-256 later once the 16GB sticks are cheaper.

Oh well, I'm not sure we'll be buying a Mac Pro if the next generation is mostly just a CPU upgrade. Might just have to build some Ubuntu machine.

 

[Inconsequential]

I see. Thanks for the clarification. I would be rather underwhelmed I guess if the Mac Pro keeps the same number of slots. 16GB sticks are still pretty expensive, and it would be nice to be able to get up to 96-128GB with 8GB sticks. Leaving the possibility to get up to 192-256 later once the 16GB sticks are cheaper.

Oh well, I'm not sure we'll be buying a Mac Pro if the next generation is mostly just a CPU upgrade. Might just have to build some Ubuntu machine.

It will be CPU + GPU + TB + Some other minor changes.

The CPU performance is going to be fairly hefty!

 

[Rustus Maximus]

It will be CPU + GPU + TB + Some other minor changes.

The CPU performance is going to be fairly hefty!

Looking forward to the CPU, GPU boost...TB...still don't really see what it brings to the Mac Pro platform, if anything at all.

 

[nanofrog]

The CPU performance is going to be fairly hefty!

It will be nice, but it may not be as "hefty" as you think. Figure 15 - 20% on avg. (depends on the software), clock per clock going by consumer versions of Sandy Bridge vs. consumer Nehalem/Westmere performance comparisons.

To me at least, the more interesting aspects are the improvements with I/O between the CPU and rest of the system that began with Nehalem (increase in memory bandwidth, ... sorts of changes over what was begun with Nehalem). Part of this is integrating areas that were previously located on additional semiconductors to the CPU's die (i.e. Nehalem moved the memory controller to the CPU die, now they've also included the PCIe controller on the CPU die). By doing so, they can speed things up as well as reduce the number of components needed on a board, which reduces manufacturing costs.

Looking forward to the CPU, GPU boost...TB...still don't really see what it brings to the Mac Pro platform, if anything at all.

TB doesn't offer anything in terms of performance improvements for desktop use only (TB is built on a PCIe 4x lane interconnect, and then reduces the throughput due to the additional latency and protocol conversions).

It can offer convenience and the ability to share peripherals with other systems however.

Unfortunately, Apple will probably use it for the graphics interconnect, going by the articles on the front page about the upcoming monitors. And sell new adapters/TB hubs to get additional cash out of users to make it work with existing monitors.

 

[Hellhammer]

It will be nice, but it may not be as "hefty" as you think. Figure 15 - 20% on avg. (depends on the software), clock per clock going by consumer versions of Sandy Bridge vs. consumer Nehalem/Westmere performance comparisons.

If the 8-core parts are real and come with decent clock speed, I think the CPU boost will be pretty good. You get 33% more cores and ~20% clock for clock increase (assuming the clock speed is ~3GHz, otherwise it will lower the increase).

 

[nanofrog]

If the 8-core parts are real and come with decent clock speed, I think the CPU boost will be pretty good. You get 33% more cores and ~20% clock for clock increase (assuming the clock speed is ~3GHz, otherwise it will lower the increase).

Some applications will benefit more than others (those capable of true n core multi-threading), but it will depend on the specific usage as to how applicable this will be. And this also assumes that any other bottlenecks such as storage are solved.

For example, if a user will only spend 25% of their time on something like After Effects, and 75% on Photoshop, most of those cores will spend most of their time idle.

As per the 8 core parts, those appear to only be the DP and later MP parts, and clocked lower to reduce power consumption (not seen a consumer SP CPUID version with 8 cores yet, so I'm making an educated guess based on what information I've seen, such as them actually exceeding their stated TDP - based on PSU current specifications).

 

[Hellhammer]

Some applications will benefit more than others (those capable of true n core multi-threading), but it will depend on the specific usage as to how applicable this will be. And this also assumes that any other bottlenecks such as storage are solved.

For example, if a user will only spend 25% of their time on something like After Effects, and 75% on Photoshop, most of those cores will spend most of their time idle.

I was talking about synthetic performance because that is what is used for comparison. It gets ridiculous if we start thinking about real world apps (too many variables) as it might bring no performance increase at all (no, Safari won't be any faster, it will still crash as often ). As we all like big numbers, it's always the maximum gain we talk about.

As per the 8 core parts, those appear to only be the DP and later MP parts, and clocked lower to reduce power consumption (not seen a consumer SP CPUID version with 8 cores yet, so I'm making an educated guess based on what information I've seen, such as them actually exceeding their stated TDP - based on PSU current specifications).

There appears to be 3GHz part as well, with a whopping 150W TDP.

 

[nanofrog]

I was talking about synthetic performance because that is what is used for comparison. It gets ridiculous if we start thinking about real world apps (too many variables) as it might bring no performance increase at all (no, Safari won't be any faster, it will still crash as often ). As we all like big numbers, it's always the maximum gain we talk about.

As long as the metrics are the same, then it will show the performance differences (synthetic or real world testing, where only the system is changed).

There appears to be 3GHz part as well, with a whopping 150W TDP.

We've seen 150W TDP's before in Xeons (in general), so that's not unusual. But where it gets sticky, is the PSU specs on the SB LGA2011 parts. The PSU spec data indicates they actually consume more (details of the conditions haven't been stated, but working backwards off of the current values, it's higher than 150W; closer to 180W IIRC).

This is also going to be an expensive part as well as hot, so it may not make it to a MP, as it would be too expensive after everything is accounted for (particularly Apple's gross margin).

Ultimately, we'll have to at least wait for quantity pricing of the DP parts before getting a better idea of what parts will be used.

 

[Hellhammer]

We've seen 150W TDP's before in Xeons (in general), so that's not unusual. But where it gets sticky, is the PSU specs on the SB LGA2011 parts. The PSU spec data indicates they actually consume more (details of the conditions haven't been stated, but working backwards off of the current values, it's higher than 150W; closer to 180W IIRC).

Maximum heat dissipation is usually 20-30% more than the actual TDP.

This is also going to be an expensive part as well as hot, so it may not make it to a MP, as it would be too expensive after everything is accounted for (particularly Apple's gross margin).

Definitely not for MP, I was just speaking of SB-E in general. I would be surprised to see any 8-core in MP before IB.

 

[nanofrog]

Maximum heat dissipation is usually 20-30% more than the actual TDP.

What I'm talking about is different (current values are noticeably higher than what would be expected on these parts).

It's more like 180W in reality, when the stated TDP is 130W due to 23A specified on the 8 pin VEPS connector (sorry about mixing the TDP values - I should have caught that). The point should be clearer now though, as a 50% increase from the stated value isn't minor (means beefier coolers, that tend to need more space).

 

[Inconsequential]

Maximum heat dissipation is usually 20-30% more than the actual TDP.

From what I recall of my overclocking days, we have to be careful with these figures for max power draw/heat dissipation.

AMD/Intel use different measurements. One sets the TDP as if 100% of it was being used (I.e. this never happens in reality) the other sets it at typical max (I.e. what the CPU would draw in normal use as a utter maximum).

I *think* AMD quotes it's TDP as 100% CPU use (I.e. every transistor is working - something that does NOT happen in real life).

Intel on the other hand quote it for something like video encoding, i.e. the closest the CPU ever gets to hardware, not software, 100% load.

This is also the reason why some CPU stability testers cause CPU temps higher than any typical desktop software, because they are using parts of the CPU that other software doesn't use concurrently.

For example my W3520 under video encoding load hits ~110W draw according to internal sensors (an external monitor shows within a few % the change from idle to load correctly so I'm happy it's reasonably correct).

Given it's rated to be a 130W chip, that leaves 20W to be accounted for!

From what I recall, only the SP systems have had 130W CPUs in them, the DP systems only have 95W chips in them?

 

[Hellhammer]

From what I recall, only the SP systems have had 130W CPUs in them, the DP systems only have 95W chips in them?

Back in the Harpertown days (i.e. 2008 MP), the DP had up to 150W CPUs. Since Nehalem, the DPs have been limited to 95W though, although higher TDP chips have existed.

 

[SamuraiArtGuy]

TB doesn't offer anything in terms of performance improvements for desktop use only (TB is built on a PCIe 4x lane interconnect, and then reduces the throughput due to the additional latency and protocol conversions).

It can offer convenience and the ability to share peripherals with other systems however.

Unfortunately, Apple will probably use it for the graphics interconnect, going by the articles on the front page about the upcoming monitors. And sell new adapters/TB hubs to get additional cash out of users to make it work with existing monitors.

I am not kidding myself that with Apple's current consumer focus, we're not going to see anything special for the MPs next update - and I'd be surprised if they turn up before 1Q2012. We're looking at the Sandy Bridge Xeon E processors, probably GPU bumps to match industry median, Lion installed w no downgrade path and of course, Thunderbolt.

TB is where it's either interesting or a point of concern. Apple is pushing the TB port AS the only display connection on the Air, MacBook and the new iMacs. On the MP's where we are both using third party Graphics Cards with their own ports, and pro users routinely use non-apple (non-glossy!) displays - our needs lean towards standard display connectors.

Of course, Apple's newest display - the 27" Apple Thunderbolt Display now connects via thunderbolt, so doesn't even have a standard connector. The apple store is still showing the 27" Apple Cinema Display on the Mac Pro page, but shown nowhere else on the site or store. As the current Radeon cards sport two Mini DisplayPort outputs, and one dual-link DVI port - they still work with third party displays.

I can't imagine that apple would not provide appropriate display ports on the graphics cards, even with the Thunderbolt port/s. The inevitable TB port would be of more interest for high end peripherals and storage devices - which for the most part don't exist yet. But unless they release a non-glossy apple display (which ain't goona happen ) they have to give us our DVI ports.

 

[tamvly]

I think SamuraiArtGuy has it just about right. I, too, wonder how I'll be connecting to my 30" Apple display to a new MP. Hopefully, not through one of Apple's unfortunate TB to DVI adapters. (I use one to connect a MBP 11 to a NEC display - it does not inspire confidence.)

I would hope for a decent CPU bump and SATA III, as well as maintaining/improving the current four drive internal arrangement.

 

[SamuraiArtGuy]

Of course, Apple's newest display - the 27" Apple Thunderbolt Display now connects via thunderbolt, so doesn't even have a standard connector. The apple store is still showing the 27" Apple Cinema Display on the Mac Pro page, but shown nowhere else on the site or store.

UPDATE: I dug back in again, and actually both the 27" Apple Cinema Display and the 27" Apple Thunderbolt Display are in the still store in the "Displays & Graphics" section - along with a PILE of assorted adapters.

But I don't really see Apple doing the pros any special favors these days. The Cinema Display could very easily be EOL'd with the next MP update in favor of the Thunderbolt Display.

----------

I would hope for a decent CPU bump and SATA III, as well as maintaining/improving the current four drive internal arrangement.

We'll get the CPU bump, that has more to do with Intel than Apple, and they pretty much have to go there for parity with the workstation market. But I wouldn't hold my breath for eSATA or SATA III, since Apple has hitched their wagon to Thunderbolt as the high speed "everything" port. We'll see what we get. Maybe on the internals we'll see some upgrades. I don't think the overall case design will change much if at all, despite rumors of a new, smaller, form factor, and they'll keep the four drive bay configuration. But it is, yes, still cosmetically identical to my early model G5 tower.

I miss the days when Apple's towers were the heaviest iron for the money and just plain DOMINATED for the price point, and got teh shiny FIRST. I guess I'm gettin' old and jaded.

 

[MyFakeAcc]

Exclusive: Sandy Bridge-E to ship without cooler :/

The upcoming Core i7 processor from Intel, the 3820, 3930K and the 3960X will all ship without a cooler in the box
[...]
On top of this, we're hearing that although the rated TDP is 130W these beasts are consuming closer to 180W and that's without even overclocking them. In fact, according to PSU design guidance we've seen, Intel is telling power supply makers to make sure their Sandy Bridge-E PSUs can cope with a peak current of 23A on the 12V2 rail and based on an 80 percent or better efficiency rating of the PSU.

As such it might be a very good idea to invest in a water cooling kit if you're planning on going Sandy Bridge-E

None of these are possible MP 2011 targets, right?

 

[Umbongo]

Exclusive: Sandy Bridge-E to ship without cooler :/

None of these are possible MP 2011 targets, right?

Apple would likely use the Xeon versions of them which are the Xeon E5 1620, 1650 and 1660.

 

[Eidorian]

Exclusive: Sandy Bridge-E to ship without cooler :/

None of these are possible MP 2011 targets, right?

Very few overclockers use a stock cooler. I can imagine the Xeons still including them.

 

[Hellhammer]

I can imagine the Xeons still including them.

The current Xeons don't include a cooler.

http://www.newegg.com/Product/Product.aspx?Item=N82E16819117242

 

[Eidorian]

The current Xeons don't include a cooler.

http://www.newegg.com/Product/Product.aspx?Item=N82E16819117242

I recall a mention that some SKUs would not include coolers. Is it all of them now?

 

[Hellhammer]

I recall a mention that some SKUs would not include coolers. Is it all of them now?

Looks like some of the SP CPUs include coolers:

CPUs with coolers
CPUs without coolers