Ivy Bridge E5 Xeon Part Numbers & Specs Leaked

[VirtualRain]

Found this... http://www.techpowerup.com/mobile/1...xeon-e5-processors-specifications-leaked.html

Can anyone decipher those part numbers into core counts? I assume the cache size can tell us if the model number can't. If that's true, is there 2.5MB Cache per core for 8, 10, and 12 cores here?

Any conjecture on what max TDP Apples new cooling solution can support?

 

[Peace]

Well the guy says they come in 8,10 and 12-cores. So by looking at the cache size you should be able to tell.

8-core has 20MB cache
10-core has 25MB cache
12-core has 30MB cache.

 

[VirtualRain]

Well the guy says they come in 8,10 and 12-cores. So by looking at the cache size you should be able to tell.

8-core has 20MB cache
10-core has 25MB cache
12-core has 30MB cache.

Ha, yeah, I guess you wrote this while I was updating my post.

Some of it doesn't make sense though... A 3.5GHz 10 Core and a 3GHz 10 core both with a TDP of 130W?! The 2643 must be an 8 core chip.

Edit, my guess is the 264x and 265x are 8 core, 266x, 267x, 268x are 10 core, and 269x are 12.

If all this is accurate, I'll take the 2643 3.5GHz Octo in my new Mac Pro please

My last comment: these are not going to be cheap!

 

[VirtualRain]

Just organizing this better and appending core-counts based on cache size...

Xeon E5-2640 V2 : 2.0 GHz, 20 MB L3, TDP 95 Watts (8-Core)
Xeon E5-2643 V2 : 3.5 GHz, 25 MB L3, TDP 130 Watts (8-Core or 10-Core?)
Xeon E5-2650 V2 : 2.6 GHz, 20 MB L3, TDP 95 Watts (8-Core)
Xeon E5-2650L V2 : 1.7 GHz, 25 MB L3, TDP 70 Watts (8-Core or 10-Core?)
Xeon E5-2660 V2 : 2.2 GHz, 25 MB L3, TDP 95 Watts (10-Core)
Xeon E5-2670 V2 : 2.5 GHz, 25 MB L3, TDP 115 Watts (10-Core)
Xeon E5-2680 V2 : 2.8 GHz, 25 MB L3, TDP 115 Watts (10-Core)
Xeon E5-2687W V2 : 3.4 GHz, 25 MB L3, TDP 150 Watts (10-Core)
Xeon E5-2690 V2 : 3.0 GHz, 25 MB L3, TDP 130 Watts (10-Core)
Xeon E5-2695 V2 : 2.4 GHz, 30 MB L3, TDP 115 Watts (12-Core)
Xeon E5-2697 V2 : 2.7 GHz, 30 MB L3, TDP 130 Watts (12-Core)

The two that don't make sense to me:
- The 2643 with 25MB L3 but a clock speed and TDP that don't make sense compared to other 10-core parts
- The 2650/50L... these are both likely 8-Cores

 

[Umbongo]

The Xeon E5-2643 V2 either does not have the cache they have listed, or Intel are not disabling it all somehow. It's either a 4 or 6 core CPU.

The Xeon line is also larger than what they have listed there as these are all $900+ models.

 

[thekev]

The Xeon line is also larger than what they have listed there as these are all $900+ models.

Yeah I was just thinking it doesn't list any Ivy 1600s cpus.

 

[deconstruct60]

Any conjecture on what max TDP Apples new cooling solution can support?

In most configurations, probably 130W because that what is the E5 1600 parts not listed here come in at. Makes even more sense if going to be swapping those E5 1600 parts out for 130W 12 cores parts at higher BTO configurations.

Whether can get also get two W9000 derivates in there also is open question.
Depends on how effective the single fan is. Even if can it is doubtful though that Mac Pro will run 'quiet' though when turn on everything at 75-100%.

 

[Tutor]

Just organizing this better and appending core-counts based on cache size...

Xeon E5-2640 V2 : 2.0 GHz, 20 MB L3, TDP 95 Watts (8-Core)
Xeon E5-2643 V2 : 3.5 GHz, 25 MB L3, TDP 130 Watts (8-Core or 10-Core?)
Xeon E5-2650 V2 : 2.6 GHz, 20 MB L3, TDP 95 Watts (8-Core)
Xeon E5-2650L V2 : 1.7 GHz, 25 MB L3, TDP 70 Watts (8-Core or 10-Core?)
Xeon E5-2660 V2 : 2.2 GHz, 25 MB L3, TDP 95 Watts (10-Core)
Xeon E5-2670 V2 : 2.5 GHz, 25 MB L3, TDP 115 Watts (10-Core)
Xeon E5-2680 V2 : 2.8 GHz, 25 MB L3, TDP 115 Watts (10-Core)
Xeon E5-2687W V2 : 3.4 GHz, 25 MB L3, TDP 150 Watts (10-Core)
Xeon E5-2690 V2 : 3.0 GHz, 25 MB L3, TDP 130 Watts (10-Core)
Xeon E5-2695 V2 : 2.4 GHz, 30 MB L3, TDP 115 Watts (12-Core)
Xeon E5-2697 V2 : 2.7 GHz, 30 MB L3, TDP 130 Watts (12-Core)

The two that don't make sense to me:
- The 2643 with 25MB L3 but a clock speed and TDP that don't make sense compared to other 10-core parts
- The 2650/50L... these are both likely 8-Cores

The "2" in, e.g., 2650, is for dual CPU systems. Here we're talking about a single CPU system - the 2013/2014 Mac Pro. That "2" adds to the cost. Probably should be looking for Xeon CPUs that begin with a "1", although a CPU model that begins with a "2" could be used, it's just be more expensive because of the QPI cost factor. I believe that Apple would lean toward lower cost , if at all possible, though not necessarily reflecting a lower price to us. If Apple does use the 26xx, here's my guess for the 26xx lineup:

Xeon E5-2640 V2 : 2.0 GHz, 20 MB L3, TDP 95 Watts (8-Core) (8 x 2 = 16 GHz)
Xeon E5-2660 V2 : 2.2 GHz, 25 MB L3, TDP 95 Watts (10-Core) (10 x 2.2 = 22 GHz)
Xeon E5-2695 V2 : 2.4 GHz, 30 MB L3, TDP 115 Watts (12-Core) (12 x 2.4 = 28.8 GHz)

because of TDP (only one system fan) and steady step up in speed.

 

[Umbongo]

The "2" in, e.g., 2650, is for dual CPU systems. Here we're talking about a single CPU system - the 2013/2014 Mac Pro. That "2" adds to the cost. Probably should be looking for Xeon CPUs that begin with a "1", although a CPU model that begins with a "2" could be used, it's just be more expensive because of the QPI cost factor. I believe that Apple would lean toward lower cost , if at all possible, though not necessarily reflecting a lower price to us.

Apple are using either an E5-2695 V2 or E5-2697 V2 to get 12 cores. Hopefully (surely) the E5-1620 V2 (3.7GHz 4-core) and E5-1650 V2 (3.4GHz 6-core) are available too. Xeon E5-2690 V2 could fit in nicely too with how Apple do things.

 

[rezwits]

In most configurations, probably 130W because that what is the E5 1600 parts not listed here come in at. Makes even more sense if going to be swapping those E5 1600 parts out for 130W 12 cores parts at higher BTO configurations.

Whether can get also get two W9000 derivates in there also is open question.
Depends on how effective the single fan is. Even if can it is doubtful though that Mac Pro will run 'quiet' though when turn on everything at 75-100%.

I think the Video card from what I have heard is the:

http://www.amd.com/us/products/desktop/graphics/7000/7970/Pages/radeon-7970.aspx

It's got the stream processors and is comparable.

You know this one:

https://www.macrumors.com/2013/03/1...d-7950-mac-edition-graphics-card-for-mac-pro/

but 70 like this

https://forums.macrumors.com/threads/1295589/

 

[rGiskard]

The odd one to me is that 3.4 GHz 150W 10-core chip that is slower than the 3.5GHz 130W 10-core chip. Not sure what the "W" at the end of the part number means.

 

[VirtualRain]

The odd one to me is that 3.4 GHz 150W 10-core chip that is slower than the 3.5GHz 130W 10-core chip. Not sure what the "W" at the end of the part number means.

The W is reserved for Intel's hottest CPUs.

As I said above, there's something not right about that 3.5GHz chip... something is wrong... either it's an 8 core or the clock speed is misquoted, or something. It just doesn't make sense for a 3.5GHz 10 core to have a TDP of 130W.

Yeah I was just thinking it doesn't list any Ivy 1600s cpus.

Is there a list somewhere, or are we still waiting for those part numbers to leak?

 

[rGiskard]

Why not? It's the fastest 10 core they offer, so it seems reasonable that it's TDP is higher. That's not unusual when trying to eek out every MHz possible from a design.

 

[deconstruct60]

Just organizing this better and appending core-counts based on cache size...

The way the design goes cores and L3 come together in layers. One layer is

[x86 core] [ ring bus ] [ 2.5 L3 cache ] [ ring bus ] [ x86 core]

You just keep "stacking" those up ( horizontally across the die surface) to go from 2 , 4 , 6 , 8 , 10 , 12 etc.
You could pehaps disable some of the cache/core of a layer but that's the normal progression.

Xeon E5-2643 V2 : 3.5 GHz, 25 MB L3, TDP 130 Watts (8-Core or 10-Core?)

I suspect that is far more likely some kind of typo. The current E6 2643 is a 3.3GHz 4 core 10MB. If drop a 0.2GHz speed bump ( like Intel is doing on the 1600's v2 increases ) then that would be 3.5GHz 4 core 10MB L3.
The 3.5GHz hits the speed right on the mark.

Three layers where the two cores were flipped off (or blank) purely to goose the L3 leves higher seems a bit extreme. Those 3 layers are only going to be accessible by ring bus only.

Xeon E5-2650L V2 : 1.7 GHz, 25 MB L3, TDP 70 Watts (8-Core or 10-Core?)

As above yet another L3 cache typo. The current 2650 and 2650L both having L3 20MB would make much more sense. It is just simpler and cheaper to use the exact same 8 core variant ide and dial down the clock and the turbo range to much lower levels.

- The 2650/50L... these are both likely 8-Cores

[/quote]

The current E5 2650 is an 8 core. That is a huge jump on GHz for core count being constant. but the price point is probably much better for that speed ( effectively old 2680 numbers )

 

[Tutor]

... . Xeon E5-2690 V2 could fit in nicely too with how Apple do things.

I believe the TDP max will be 115W/quiet because of the one fan issue.

 

[deconstruct60]

Is there a list somewhere, or are we still waiting for those part numbers to leak?

Can wait for the leak but probably will not be much different than these numbers which have leaked.

http://www.cpu-world.com/news_2013/...ge-E_extreme_CPUs_to_launch_in_September.html

Just substitute E5 1620 v2 , 1650 v2 , 1660 v2 in the first column in that table.

 

[deconstruct60]

The "2" in, e.g., 2650, is for dual CPU systems. Here we're talking about a single CPU system - the 2013/2014 Mac Pro. That "2" adds to the cost. Probably should be looking for Xeon CPUs that begin with a "1", although a CPU model that begins with a "2" could be used, it's just be more expensive because of the QPI cost factor. I believe that Apple would lean toward lower cost ,

Apple will lean to lower costs so the core of the line up will probably use 1600 parts. But they are also going to be looking to show they are not backsliding. The entry level current dual package Mac Pro has 12 cores.
For the new and improved Mac Pro to top out at just 6 cores would be kind of embarrassing to those that just count x86 cores or are bounded by x86 cores restricted software.

There will probably be some BTOs to get back into the double digits with processors that cost higher than the E5 1660 tops out at ( ~$1000). It is a bit of "wasted" money, but the E5 2600 options don't even have to be part of their "good , better ,best " line up at all. Since BTO zone that is always where Apple gooses the profit margin a bit more. That extra QPI tax that Intel puts on they can either offset ( by just not going quite as high) or just use to print even more money ( slap the normal 30% mark-up on top).

This is pretty close to the old dual processor box without having to pay the space cost for a dual processor box.

There is probably a C602 chipset in there too. It has 10 SATA lanes that are being flushed down the drain. Two dead QPI links is kind of par-for-the-course in this new box.

 

[VirtualRain]

The way the design goes cores and L3 come together in layers. One layer is

[x86 core] [ ring bus ] [ 2.5 L3 cache ] [ ring bus ] [ x86 core]

You just keep "stacking" those up ( horizontally across the die surface) to go from 2 , 4 , 6 , 8 , 10 , 12 etc.
You could pehaps disable some of the cache/core of a layer but that's the normal progression.




I suspect that is far more likely some kind of typo. The current E6 2643 is a 3.3GHz 4 core 10MB. If drop a 0.2GHz speed bump ( like Intel is doing on the 1600's v2 increases ) then that would be 3.5GHz 4 core 10MB L3.
The 3.5GHz hits the speed right on the mark.

Three layers where the two cores were flipped off (or blank) purely to goose the L3 leves higher seems a bit extreme. Those 3 layers are only going to be accessible by ring bus only.





As above yet another L3 cache typo. The current 2650 and 2650L both having L3 20MB would make much more sense. It is just simpler and cheaper to use the exact same 8 core variant ide and dial down the clock and the turbo range to much lower levels.

The current E5 2650 is an 8 core. That is a huge jump on GHz for core count being constant. but the price point is probably much better for that speed ( effectively old 2680 numbers )

Can wait for the leak but probably will not be much different than these numbers which have leaked.

http://www.cpu-world.com/news_2013/...ge-E_extreme_CPUs_to_launch_in_September.html

Just substitute E5 1620 v2 , 1650 v2 , 1660 v2 in the first column in that table.

Makes perfect sense. I can't wait to see the configurations that Apple offers although I suspect the price points are going to hurt.

 

[rGiskard]

I believe the TDP max will be 115W/quiet because of the one fan issue.

Nothing in the design suggests a 115W max TDP. Each of the GPUs will pull considerably more than that, so a 130W or even 150W CPU seems entirely in line.

Remember, that triangular heatsink is freakin' HUGE, and the fan can move a lot of air. The intake air is going to be room temperature as well.

 

[beaker7]

I'll take 2 of the E5-2687w V2.

 

[wallysb01]

Nothing in the design suggests a 115W max TDP. Each of the GPUs will pull considerably more than that, so a 130W or even 150W CPU seems entirely in line.

Remember, that triangular heatsink is freakin' HUGE, and the fan can move a lot of air. The intake air is going to be room temperature as well.

You'd think it could certainly handle the 130W chips, because that's what the 1600 series are at. Wether they put the 150W chip comes down to pricing strategy as well.

At 3.4 GHz, that 10-core 150W (34 GHz by simple math for basic guide) chip should be faster than the the 12-core 2.4 and 2.7 GHz chip (28.8 and 32.4 GHz respectively).

So, my guess, other than the 1600 V2s, they go with one of the mid range 10-core chips (ie 2670/80), the 2690 and the 2697.

 

[deconstruct60]

You'd think it could certainly handle the 130W chips, because that's what the 1600 series are at. Wether they put the 150W chip comes down to pricing strategy as well.

Can't keep leveraging "it is just another 20W" to squeeze at 150W CPU in there. The GPUs in the W9000 configuration are much bigger TDP issue and extremely likely the CPU is just going to have to fit inside of some boundary. Going past 130W is just doubtful. Given two GPUs at 250+W + 130W is already in the 630+W zone. 3 Thunderbolt network power distribution at 10W each ( 660+W ). USB 3.0 4.5W per port ( 680+W ).
All the air is not going through the central pipe. ( It can't there is stuff to cool on the outside of the cards as well. )

The cylinder is only 6.6" ( 168mm ) in diameter. There is not a enormous difference from what a 100-120mm fan would do. It's more but it isn't double the amount of air.

The heat sink just repositions the heat for removal. You can't dump and store it there for long computational runs. For minutes perhaps, but if grinding at top speed for hours eventually it will get to full capacity.

Lots of indications here that because they don't have to deal with random 3rd party cards (and random CPUs) , Apple shrank the thermal management down to the limits of what they need and not much more. (e.g., where they point out that different sides of the triangle can pump out more than their share heat if the other side aren't. )

The "crazy high" priced 150W model is likely not going to make any pricing cut. But also probably bumps up against pushing requirements for a bigger box and/or additional cooling complexity.

At 3.4 GHz, that 10-core 150W (34 GHz by simple math for basic guide) chip should be faster than the the 12-core 2.4 and 2.7 GHz chip (28.8 and 32.4 GHz respectively).

If the price performance isn't linear then it isn't money well spent.

So, my guess, other than the 1600 V2s, they go with one of the mid range 10-core chips (ie 2670/80), the 2690 and the 2697.

The problem for the 2670 and 2780 is that too are outside the zone of processors that Apple has tended to use ( Jumping from E5 1660's (~1,000 ) point right up to $1500 range ( past Apple's typical $1400-1500 zone for the top). I can see them compelled to put the > $1700 12 core into the mix but $500 (pre 30% mark-up jump) is going to leave some folks behind. probably just one 10 core option and then the "if you have to ask can't afford it" 12 core option.

 

[tomvos]

Given two GPUs at 250+W + 130W is already in the 630+W zone. 3 Thunderbolt network power distribution at 10W each ( 660+W ). USB 3.0 4.5W per port ( 680+W ).
All the air is not going through the central pipe.

Well, looks like the central pipe acts as a turbine if the load exceeds ~ 500W. If the new Mac Pro pushes enough air through, I think it might even levitate.

 

[deconstruct60]

Well, looks like the central pipe acts as a turbine if the load exceeds ~ 500W.

The central pipe is completely stationary. In no way do it act like a turbine at all. The fan acts like a turbine. The pipe is stationary with straight lines ( from top to bottom) which allows extremely clean, laminar airflow. There is no compression mechanism evident either, not is it desirable. A turbine ( and chaotic flow) is exactly what you do NOT want in a heat exchanger.

If the GPU loads exceed 500W the pipe will try to pump more heat at the CPU. If the CPU happens to be running cooler than its TDP max point that the CPU's "share" of the heat exchanger can be use to handle the GPU 'overflow'. However, if the CPU is using its full 'share' there is no place for overflow to go.

With a saturated heat sink, at some point the thermal management systems of the GPUs and/or CPU will down clock the subsystems. The fan will have time to catch up to bleed some heat out of the heat exchanger.

That design is one of that probably leverages on a bit that most applications cannot keep all three major computation engines going at full blast for extended periods of time. For primarily human driven interaction programs that is probably a safe bet. A person will periodically stop and look at something. They don't work non-stop at high rates of speed (relative to computer internals ) for hours on end.

If the new Mac Pro pushes enough air through, I think it might even levitate.

Impossible since the forces are going in the opposite direction. The air is shot out up (and maybe a bit to the sides ). That means the reactive force on the Mac Pro is down. It is like a jet engine pointed directly at the ground. ( not one pointed straight up at the sky).

 

[beaker7]

Amazing how much you all know about a machine you've never used, tested, or likely even seen in person.