New Mac Pro

[L. Langer]

LGA 2011 is at the end of its run. Unlikely to be any further chips released for the Mac Pro than what is out now.

Broadwell Xeons will be next and they use LGA 2011.

 

[DJ Meyer]

Broadwell Xeons will be next and they use LGA 2011.

Haswell-EP is next and uses LGA 2011-3. Requires C610 chipset and DDR4 RAM.

It is similar but no, not socket compatible with Sandy and Ivy.

Broadwl is the successor to Haswell and will work in 2011-3.

 

[Jeff Kilgroe]

Actually we don't know what will be next. Haswell and Broadwell Xeons are both on the table. Haswell has not made the full push into the top end desktop CPUs as it offers nothing on that end of things over the current Ivy Bridge CPUs like the i7-4960X. IMO, if they release Haswell Xeons, we would gain nothing and it would just be about marketing and getting people to buy new workstations. Broadwell would bring the full transition to DDR4 and PCIe v4.

I think it will come down to timing in the market and how sales go. If they push to release Haswell Xeons by the end of this year, then we probably won't see any updated E5-x600V2 Xeon chips. If they do the smart thing and hold off the next Xeon update to release Broadwell based chips sometime in late 2015, we will probably see a few new chips trickle out later this year. There has been an E5-2699V2 3.1GHz 12-core that has appeared in some parts lists only to disappear. I think it's a higher TDP / 150W chip though. There is supposed to be a 10-core E5-1690V2 at 2.8GHz that has yet to materialize.

The current Ivy Bridge Xeons will be the last to use the Socket LGA2011 as it exists now.

 

[Wayne Morellini]

Larrabee, opps, did I say that. What was the other name of their massive core project. We really need to get away from the monolithic processor idea if throwing everything including 4 kitchen sinks into a processor to get modest performance gains if it were on the same process of the last chip. A stripped firm redesign is needed that goes faster with more processors in the same space. Intel is working on one, but we have Arm already as a partial solution.

We need to rearrange the data workflow too, having large on cycle random access local on chip/die memory surrounded by as fast as possible surrounding chip memory going to high speed memory like storage array. When something like QDCA comes through and reduces heat production we can have cube like chips reducing the cases of strict need for external memory.

 

[Wayne Morellini]

Thinking that of this again, a Xeon with their massive core array. The marketing is relatively simple, have the tools ready, Open CL is a tool, Release a Xeon where one processor is replaced by an processing array, let the application programmes build in support, then over time move to replacing most of the normal Xeon processors.

Notice I say "processing array", as a stripped down PC processor may not prove to be the best solution. So, another solution become evident, a virtual solution on software that allows the array to change over time, a virtual engine language solution.

 

[Jeff Kilgroe]

Thinking that of this again, a Xeon with their massive core array. The marketing is relatively simple, have the tools ready, Open CL is a tool, Release a Xeon where one processor is replaced by an processing array, let the application programmes build in support, then over time move to replacing most of the normal Xeon processors.

This is actually where they're headed with their MIC (Xeon Phi) tech. Eventually they will merge MIC (Many Integrated Cores) GPU units directly into CPU chips. We're still a few years out and it will probably happen with the die shrink to 15nm (speculative). But We'll probably see single 4GHz Xeon CPUs with 12 to 16 regular cores plus one or two MIC cores providing a whole bunch of stream processors or x86 micro-cores each.

Unfortunately, Xeon Phi hasn't been catching on in the market. I think Intel needs to find a better way to market these cards. They seem promising, yet no one is really interested in working with them.

 

[Bruce Allen]

This is actually where they're headed with their MIC (Xeon Phi) tech. Eventually they will merge MIC (Many Integrated Cores) GPU units directly into CPU chips. We're still a few years out and it will probably happen with the die shrink to 15nm (speculative). But We'll probably see single 4GHz Xeon CPUs with 12 to 16 regular cores plus one or two MIC cores providing a whole bunch of stream processors or x86 micro-cores each.

Unfortunately, Xeon Phi hasn't been catching on in the market. I think Intel needs to find a better way to market these cards. They seem promising, yet no one is really interested in working with them.

Intel don't need marketing. They need to just make those Xeon Phi cards do something useful like Resolve or V-Ray faster than the nVidia cards already do for less money.

There's a big difference between "promising" and "better than what's currently on the market".

I remember being at an Autodesk event called 3December back in 1999 or so in South Africa. They were trying to convince us that Itanium was going to be the future. But the demos just weren't that good. And AMD's less-elegant 64-bit instruction set turned out to be the way to go.

I guess that's why I just like nVidia's efforts as opposed to AMD and Intel's Xeon Phi. The nVidia stuff works and they put in the resources to work with the companies needed to get actual results.

If Intel and AMD actually feel like actually bothering to compete on a serious level, great. But I'll judge that in "can I use this hardware to make apps I want to use run faster for cheaper" rather than "ooh cool their architecture is so theoretically promising and has nice big FLOPS numbers" or whatever.

BTW Jeff, don't you think there will at least be a 1-series Xeon version of Haswell next year? It'd be awfully odd if there is an 8-core Haswell-E i7 Extreme processor but no Xeon model.

Bruce Allen
www.boacinema.com

 

[Brian Timmons]

A very thorough new Mac Pro teardown from ifixit.com

http://www.ifixit.com/Teardown/Mac+Pro+Late+2013+Teardown/20778#s56812

Interesting that the GPU cards are slightly different from each other and that the second
GPU is where the SSD is installed.

Brian Timmons
BRITIM/MEDIA

 

[Antony Newman]

I believe that 15c xeon were due at the end of 2014 Q1 - but Mac OS may have issues running with more than 12c's.

'12c 150w Xeon E5-2699 v2 / 3.1GHz' - I'll pass on this one as it's more than the 130w TDP.

This leaves two (current) CPUs that can be upgraded to that Apple don't offer:

10c 130w Xeon E5-2690 v2
http://ark.intel.com/products/75279
Turbo : 1c=3.6GHz/3.5/3.4/3.3/3.3/3.3/3.3/3.3/3.3/3.3GHz=10c

8c 130w Xeon E5-2667 v2
http://ark.intel.com/products/75273
Turbo : 1c=4Ghz/3.9/3.8/3.7/3.6/3.6/3.6/3.6GHz=8c

I'm currently thinking the 8c is the best compromise.

AJ

 

[Brian Merlen]

just read the tear down and i have to say I like what I am hearing about the possibility of upgrades. I could theoretically start bare bones with a stock 4 core...maybe up the GPUs since as of now I don't think there is third party options or retailers for GPU yet.... might not ever be, or may take a while... so for safety i get stock besides the GPUs, which i max out to D700s...

So that leaves me with a fairly decent processor speed on 4 cores, kind of low on ram, and SSD size fine for my needs since it runs OS and apps only...

and later I can up the cores AND ram, and perhaps one day even GPU...

i really dont like the fact they took away what little GPU third party control we have and its almost a deal breaker on me from a wtf stand point, but one can hope the industry of GPU makers will lend us future options (in a post PC world one can hope)... but i have to admit servicing a hackentosh in comparison is less then ideal is my current belief, can be done but i'd really not like to have to troubleshoot a hackentosh on set with clients on hand is my primary concern over that route now, cuz even if its not the hacks fault i am sure it will be blamed by the clients and potentially reflect upon me. I have met people who just dont trust hacks no matter how much more power they can provide. so part of me wants to pay the apple brand tax simply to say "aha it cannot be a hacks fault cuz my hardware is not third party" when issues arise..

it will limit to me to who i can blame, coupled with applecare which has helped me out in the past with my currently in use but somewhat dated machine. i have been thinking and debating an update for a long time, have presented all kinds of different ideas on the matter...was initially pretty disappointed in the PCI slot issue, but i think after reading the above tear down at least 2 out of 3 of the potential things I would ever likely want to expand can currently be done, and perhaps the GPU thing will work itself out and I can get away with a fully stock model, but I think hedging myself there is a wise one for now...

gotta wonder how it will even run with so much Vram vs so few cores and ram lol!!! i really just dont wanna embrace a windows 7 universe is another issue, unless i get a Miro or something and really need to i wanna avoid that route as long as possible


PS does anyone with both a Rocket X and Dragon also have a Mac Pro and can do tests comparing it with Rocket vs GPU only for playback and a variety of encoding? (Even better if they buy the 4 core mac pro!) Cuz before I blow this cash I should probably figure out how well it will run without a rocket too since it has no slots and will always be limited by the TB2 should I ever spring for it

 

[Eric Santiago]

Has anyone looked into matching old tech with any TB/TB2 enclosures e.g. ROBEN-3.

Trying to merge an Avid HBA card (Nitris DX) and MAGMA HBA (Digidesign Magma Expansion Chassis) along with ROCKET/DeckLink (dream set-up).

I know off the ATTOs working with TB/TB2 so Im assuming itll be fine with the ROBEN/SONNET options.

 

[Maz Mawlawi]

It is amazing to me that in this day of age people still cling so personally to a specific OS platform...you really should open up your horizons and not be afraid to be cross platform. Today windows and MAC systems performs virtually identical...and most software is now cross platform. Use the best tools for the job. We found that windows workstations perform much better and are faster (due to better hardware options) than macs...on the other hand MAC laptops are way better than windows one....hence we live in a hybrid world. I find the trash can is to limiting....everything has to be external...it's gonna end up looking like an octopus not to mention the high cost of every TB2 enclosure and the possible limitation with rocket x etc.... Not sure why you some are so afraid of PC/windows...been running workstations for 3 years now non stop and not a single one failed or crashed or gave us any trouble. On the other hand PC laptops really suck....so I'd stay MAC all the way on that.

 

[Eric Bowen]

http://www.extremetech.com/computin...-offer-ddr4-support-up-to-eight-cores-in-2014

Haswell E brings DDR4 to the market

Dram wafer fabs are already starting to retool some lines for DDR4.

 

[Jeff Kilgroe]

SHIPPED!

Looks like I get the Mac Pro wednesday.

CTO 8-Core, 64GB, 1TB SSD, D700 GPUs.


I ordered a 24" 4K monitor from Dell to play with. Don't know when it will show up...

 

[FrankMcPartland]

Congrats Jeff! I look forward to any 4k monitor advice you have!

 

[M Most]

It is amazing to me that in this day of age people still cling so personally to a specific OS platform...you really should open up your horizons and not be afraid to be cross platform. Today windows and MAC systems performs virtually identical...and most software is now cross platform.

I agree with your general sentiment, but saying that "most software" is now cross platform is a bit misleading, since it is ignoring some programs that have a very significant presence in the media industry, specifically Final Cut Pro (multiple versions), Autodesk Smoke, Logic Pro - all of which are Mac only - and Flame Premium (which includes Lustre), Mistika, Baselight, and a few others that are Linux only. Granted, there are an awful lot of programs that are cross platform, but we're not talking about the general computing world, we're talking about the media creation industry.

 

[Eric Bowen]

The Media Creation world also has equivalents on Windows such as Avid, Adobe, Fusion, Davinci, Nuendo, Cubase, and Pro Tools. So platforms are really just dependent on what applications the editor wants to use.

 

[Isaac Marchionna]

SHIPPED!

Looks like I get the Mac Pro wednesday.

CTO 8-Core, 64GB, 1TB SSD, D700 GPUs.


I ordered a 24" 4K monitor from Dell to play with. Don't know when it will show up...

Congrats, I wonder if mine will be soon, or if like my usual luck I'll get mega-boned.

 

[Jeff Kilgroe]

Intel don't need marketing. They need to just make those Xeon Phi cards do something useful like Resolve or V-Ray faster than the nVidia cards already do for less money.

Oh, they do need some marketing help. They need to do what nVidia does. They need to give these cards to software developers and hold their hands along the way as they write software for this hardware. They need to send representatives around to the major software companies who know the hardware and developer tools inside and out, who can offer assistance.

There's a big difference between "promising" and "better than what's currently on the market".

Yep.

I remember being at an Autodesk event called 3December back in 1999 or so in South Africa. They were trying to convince us that Itanium was going to be the future. But the demos just weren't that good. And AMD's less-elegant 64-bit instruction set turned out to be the way to go.

Itanium was a gross mis judgement by Intel as to where the industry was headed. Perhaps it could have been great, we will never know. But it was taking the concept of if already bloated instruction sets and pipelines and compounding them and combining them into even larger and more parallelized sets. That's where the whole Intel EPIC (Explicitly Parallel Instruction Computing) architecture came in. It was more akin to the MIPS microprocessors in the vein of the R10K and R14K, but with a rather diverse instruction set implementation and even some downward compatibility with x86. Who knows what would have happened if it had caught on. I actually thought the IA64 instruction set showed a lot of promise, better than what AMD had. Unfortunately, IA64 was perched atop a bloated pile of poo. Anyway, file that under the what-ifs. Computing would be in a vastly different place if not for AMD's surge of innovation and battles with Intel back then.

I guess that's why I just like nVidia's efforts as opposed to AMD and Intel's Xeon Phi. The nVidia stuff works and they put in the resources to work with the companies needed to get actual results.

It's all about the resources and how they, nVidia, invest themselves in getting developers to support their products. Doesn't matter how large or small you or your project is, nVidia takes care of those developing for their products.

BTW Jeff, don't you think there will at least be a 1-series Xeon version of Haswell next year? It'd be awfully odd if there is an 8-core Haswell-E i7 Extreme processor but no Xeon model.

I think it all comes down to sales figures and ROI of the current Xeon lines. The E5 Sandy Bridge based Xeons didn't sell as well as anticipated, even with the long drought leading up to their launch. They were priced high and many integrators found the diverse offerings to be somewhat confusing. Pricing was a lot higher than previous Xeon CPUs and for many, the Westmere Xeons still held far better value. Intel knows this, that's why they still make them even though they were supposed to discontinue them over 18 months ago.

The new E5V2 Xeons just compound this mess. They offer a few more cores with some models, but at the expense of higher prices and slower CPU clocks. We're seeing a much more defined catalog of offerings where we get to pick either cores or speed, not both. Some chip models like the E5-2699V2 have appeared, disappeared, reappeared and disappeared again from product sheets. I do think we'll see a few updated CPUs in the next year. I don't know if it will be Haswell based or a continuation of the Ivy based ones. The market doesn't want a Haswell based Xeon, I can tell you that. The market wants Broadwell with PCIe v4. If Intel drops a Haswell Xeon turd on us in late 2014 or early 2015 when Broadwell systems are about to start finding their way onto desktops, then consumers are going to be generally disinterested in those Xeons.

Initial reports on overall E5V2 sales are kinda what I expected and aren't that great. Very few are upgrading CPUs as the newer chips don't offer enough performance gain to justify the high cost of an upgrade to already expensive CPUs that only had 18 months or less to reach their ROI. OTOH, There's a surplus of E5-2600 V1 chips out there because as soon as the V2's launched, everyone quit buying the V1's. Westmere Xeons continue to sell briskly and have sold many times more than E5 Xeons have and probably ever will. Even with the new Mac Pro using the E5V2's and Dell and others sticking the low-clock models in workstation class notebooks.

Intel doesn't make their roadmaps public anymore. We can only go off of hints they drop at venues like IDC and some of the leaks they allow to come out.

 

[Mark Phelan]

SHIPPED!

Looks like I get the Mac Pro wednesday.

CTO 8-Core, 64GB, 1TB SSD, D700 GPUs.


I ordered a 24" 4K monitor from Dell to play with. Don't know when it will show up...

Will be very interested in your thoughts once you've had a chance to play with it. Thanks for your other insights too.