What are the chances - Mac Pro / xMac with SB-E

[theSeb]

The Sandy Bridge E 3960 makes itself into a compelling choice for those looking for a single CPU Sandy Bridge workstation. What are the chances that we may see a single CPU Mac Pro released based on the SB E CPU? I assume the general consensus is none, which is a pity because it will suit my workloads perfectly and I don't want to leave Mac OSX.

http://www.anandtech.com/show/5091/...y-bridge-e-review-keeping-the-high-end-alive/

 

[khollister]

Calm down - This part will probably have about a 10-15% performance advantage at most over the Westmere 3.33 6 core Xeon available in the current MP. Other than bragging rights, I fail to see how that is worth ditching OS X and what I presume is an investment in professional software that takes advantage of that.

The bigger new is that the Xeon part will have 8 cores - that should put a 3.33 Ghz SP MP at well over a 20,000 Geekbench score. That is knocking on the door of the current top end 12-core DP machine. Now that's something to get excited over. Of course, Intel hasn't released those parts yet to anyone.

 

[thekev]

Daaaammmnn you Khollister...... now I'm thinking about how fast that thing would render.

Also given that we're talking about a workstation here, the increased per core performance is very welcome. We are talking about a workstation rather than a server so cramming the same amount of raw power into fewer cores is generally better as not every function scales well.

 

[theSeb]

I am pretty calm, but I am just getting bored of waiting.

 

[khollister]

I am pretty calm, but I am just getting bored of waiting.

Understand, but jumping to a PC/Windows won't help much - everyone is waiting on the same Intel Xeons. A move at this point would get you SATA III, but no Thunderbolt and a very minimal CPU improvement.

Unless Apple pulls the plug (too early IMHO), you will get a 8 core SP Xeon with SATA III, TB and maybe PCIe 3.0 if the standards committee gets their ass in gear sometime Q1 next year. I doubt PCI 3 because I suspect Apple (and everyone else) has the boards already designed and is just waiting on Intel for last minutes tweaks.

----------

Daaaammmnn you Khollister...... now I'm thinking about how fast that thing would render.

Also given that we're talking about a workstation here, the increased per core performance is very welcome. We are talking about a workstation rather than a server so cramming the same amount of raw power into fewer cores is generally better as not every function scales well.

Just imagine the DP version - we are probably talking over 30,000 Geekbench scores - 16 cores.

 

[nanofrog]

Unless Apple pulls the plug (too early IMHO), you will get a 8 core SP Xeon with SATA III, TB and maybe PCIe 3.0 if the standards committee gets their ass in gear sometime Q1 next year. I doubt PCI 3 because I suspect Apple (and everyone else) has the boards already designed and is just waiting on Intel for last minutes tweaks.

Hopefully, the CPU's and boards will release with PCIe 3.0 locked in rather than gain certification after the fact (could get ugly if this happens).

But there should be a Xeon based on the X3690, and at 3.9GHz on 6 cores, should be able to produce an nice balance of single and multi-threaded performance. And at a decent cost as well (SP variants don't have the premium prices associated with DP CPUID's <2nd QPI channel active>).

I also hope that Apple doesn't jump the gun and cut the MP before the SB/IB LGA2011 Xeons are released (too soon IMO from a workstation performance standpoint).

 

[khollister]

Hopefully, the CPU's and boards will release with PCIe 3.0 locked in rather than gain certification after the fact (could get ugly if this happens).

But there should be a Xeon based on the X3690, and at 3.9GHz on 6 cores, should be able to produce an nice balance of single and multi-threaded performance. And at a decent cost as well (SP variants don't have the premium prices associated with DP CPUID's <2nd QPI channel active>).

I also hope that Apple doesn't jump the gun and cut the MP before the SB/IB LGA2011 Xeons are released (too soon IMO from a workstation performance standpoint).

Right - I think the transition point of eliminating the MP as we know it today is around Haswell. I think the MP has a couple more updates (both on LGA2011). They need 8 core consumer chips, a more mature TB eco-system and some more maturity to the GrandCentral architecture to enable more widespread slave processing.

The music guys are already way down that road. I see more and more folks talking about i7 quad MBP or Mini hosted DAW's with multiple PC or Mac Mini slaves using Vienna Ensemble Pro. VSL has revolutionized the industry with VEPro. The days of requiring a DP MP are over for that particular market segment. I'm going there myself next year with PC slave to host EWQL Hollywood Strings and Hollywood Brass libraries back to my MP Logic/Spectrasonics/Kontakt environment. EWQL Play runtime is more efficient on Windows, so a cheap i7-2600K with 32GB RAM and a couple SSD's is the ticket. If I needed NI Kontakt instances, I would probably use an i7 Mini with OS X. The beauty of VEPro is that once the slave is set up, the host OS is completely invisible to the user. You see a virtual instrument in a VEPro frame in your DAW just like it was running locally. You never touch the slaves except for admin stuff. Everything runs via GigE.

 

[deconstruct60]

Right - I think the transition point of eliminating the MP as we know it today is around Haswell. I think the MP has a couple more updates (both on LGA2011). They need 8 core consumer chips, a more mature TB eco-system and some more maturity to the GrandCentral architecture to enable more widespread slave processing.

Desktop Haswell doesn't have 8.

http://www.anandtech.com/show/5078/...abooks-gt3-gpu-for-mobile-lga1150-for-desktop

Intel is going to spend more of the additional transistor budget adding GPU infrastructure: cores, more GPGPU support, and likely memory I/O . The other major goal is power reduction. [ may see increase in PCI-e lane support: 20 is rather limited]. The x86 core count (4) stays the same.
The desktop market is moving to "System on a Chip" (SoC) where the old GPU, Northbridge, and Southbridge is largely being moved into one single package. The ever shrinking average selling price of a PC put a higher priority on lower costs.

There also won't be an 8 core SP Xeon E5. The 1600 series will be capped at 6 cores just like the i7 39xx . Only the dual package Mac Pros will have 16 core configs and they will not come in 3.3GHz config. You will be able to pick highest GHz or highest core count, but not both.

Choosing the Core i7 has no real advantage for Apple. The Xeon E5 1600 series is priced nearly the same (so system cost not really moving) and is more compatible with the components used in E5 2600 systems. The only reason to drop the 1600 is if they were going to drop the 2600. That seems unlikely because for the "high thread count, high I/O bandwidth" market, more is better (and can charge more).

 

[khollister]

Desktop Haswell doesn't have 8.

http://www.anandtech.com/show/5078/...abooks-gt3-gpu-for-mobile-lga1150-for-desktop

Intel is going to spend more of the additional transistor budget adding GPU infrastructure: cores, more GPGPU support, and likely memory I/O . The other major goal is power reduction. [ may see increase in PCI-e lane support: 20 is rather limited]. The x86 core count (4) stays the same.


There also won't be an 8 core SP Xeon E5. The 1600 series will be capped at 6 cores just like the i7 39xx . Only the dual package Mac Pros will have 16 core configs and they will not come in 3.3GHz config. You will be able to pick highest GHz or highest core count, but not both.

Choosing the Core i7 has no real advantage for Apple. The Xeon E5 1600 series is priced nearly the same (so system cost not really moving) and is more compatible with the components used in E5 2600 systems. The only reason to drop the 1600 is if they were going to drop the 2600. That seems unlikely because for the "high thread count, high I/O bandwidth" market, more is better (and can charge more).

The article the OP linked to indicates there will be 8 core SB-EP (aka Xeon) parts and states the other 2 cores are on-die but fused off for the i7-Extreme part. There is a photo of the die that appears to show two more cores on-chip as well.

I'll have to find the reference, but I know I read about Haswell 8 core mainstream devices. I'll see if I can find it again

UPDATE: At first blush it appears you are correct, 8 core for server parts, but only 4 for consumer. Looks like TDP and GPU are the big news along with 1600 memory.

Well, great - so all we are really getting in an update is SATA III and TB as well as an insanely expensive 16 core. This complicates the reasoning quite a bit.

 

[DisMyMac]

It's almost 10:30 EST. What time does Tim get to work out there?

 

[theSeb]

It's almost 10:30 EST. What time does Tim get to work out there?

Tim generally starts work on a Sunday evening.

 

[deconstruct60]

I'll have to find the reference, but I know I read about Haswell 8 core mainstream devices. I'll see if I can find it again

UPDATE: At first blush it appears you are correct, 8 core for server parts, but only 4 for consumer. Looks like TDP and GPU are the big news along with 1600 memory.

I have no doubt that some folks predicted 8 but that likely was not re-quoting an Intel source with knowledge.

There will likely be an "Extreme" part that is based on the Xeon config (just like this i7 39xx series is. ). That will probably have 8 (with two cores flipped off .... again just like the current versions). Or , if the GPGPU abilities are good enough at that point, the Haswell Xeon will also temporarily plateau at 8 x86 cores and add a GT1 ( the lower performance and highly likely much lower transistor budget GPU variation) to the mix.

If you look at the Anantech (http://www.anandtech.com/show/5091/...dy-bridge-e-review-keeping-the-high-end-alive) and Tom's Hardware (http://www.tomshardware.com/reviews/core-i7-3960x-x79-performance,3026.html) reviews there is a bit of moaning and groaning over not having Quick Sync on these "Extreme" parts. Adding a relatively small GPU to the Xeon with just enough budget to do Quick Sync would eliminate that groaning. If the OpenCL drivers are good enough by that point even a server could get some computational "pop" from some GPGPU cores. Having a GPU present also makes the Thunderbolt system implementation much easier.

Intel could save the greater than 8 core range for the E7s. That would make some sense. With 8 cores you need a quite large L3 cache and more memory channels to be effective. Large L3 means larger die (memory consume lots of space) which means more expensive part.
By 2013 even the workstation class boxes will be under more pressure to drop system build costs. Many server boxes have nominal graphics connectors to connect a console monitor (although usually VGA); so it isn't like they are completely devoid of GPUs now.

I suspect the Haswell Xeon E5 will get a GPU instead of more x86 cores. the 8 x86 core present will get faster and more efficient so their would be improvements. Just not brute force improvements of bumping core count.

 

[nanofrog]

Right - I think the transition point of eliminating the MP as we know it today is around Haswell.

This is what I'm seeing as well (not the mainstream consumer socket, but the enthusiast parts built on the same socket as SP Xeons, such as those on LGA1366 and LGA2011).

Desktop Haswell doesn't have 8.

That article is referring to the mainstream LGA1150 socket though, not the enthusiast socket.

As per the direction you've mentioned for the mainstream parts, it's definitely intended to create SoC solutions. Intel has been integrating sections of previous sub-components onto the CPU die (i.e. PCIe and memory controllers).

I have no doubt that some folks predicted 8 but that likely was not re-quoting an Intel source with knowledge.

I've not seen an article yet, but it would fit with their more recent history, assuming they don't change their current stance with enthusiast parts/SP Xeons and ditch adding a pair of cores in favor of adding an IGP (where my current logic is coming from).

My main reasoning for this, is most users are willing to add in their own GPU card anyway to meet their specific needs (users can't change the CPU's die, so more cores on a single die, and perhaps a few more improvements, even if just perception, would continue to generate sales). Particularly for those that could switch from DP systems to a less expensive SP system that would fulfill their needs in terms of core and I/O performance (i.e. much higher clocked SP vs. slower clocked DP with only a pair of additional cores; say 500MHz+ per core in non-Turbo mode).

Combine this with the fact a decent graphics card is likely cheaper than the cost difference between an SP and DP system, it would be rather attractive for workstation users (consumer enthusiast parts for those that don't need ECC, and Xeon for those that do).

 

[Adamantoise]

Understand, but jumping to a PC/Windows won't help much - everyone is waiting on the same Intel Xeons. A move at this point would get you SATA III, but no Thunderbolt and a very minimal CPU improvement.

Unless Apple pulls the plug (too early IMHO), you will get a 8 core SP Xeon with SATA III, TB and maybe PCIe 3.0 if the standards committee gets their ass in gear sometime Q1 next year. I doubt PCI 3 because I suspect Apple (and everyone else) has the boards already designed and is just waiting on Intel for last minutes tweaks.

----------



Just imagine the DP version - we are probably talking over 30,000 Geekbench scores - 16 cores.

Thunderbolt is not worth waiting for. It's great on paper, but has close to 0 practical use ... For now anyway.

 

[DisMyMac]

Thunderbolt is not worth waiting for. It's great on paper, but has close to 0 practical use ... For now anyway.

They finally have expresscard adapters for TB: http://www.sonnettech.com/product/echoexpresscard34thunderbolt.html

So it's actually an adapter for an adapter, which allows connections like eSATA. (I'm almost positive they advertized USB3 support, but that seems to have mysteriously vanished from their ad.)