Haswell-E doesn't have much over Ivy

[Tutor]

I think that there are two independent questions here that are being treated as one.

#1

Will the Haswell-EP systems offer greatly improved performance, so that most people with an MP6,1 will want to upgrade?

No. The improvement will be minor for most apps. If you have an app that benefits from AVX2, it might be worth considering. Otherwise, stick with your MP6,1 until a big improvement *on your workflow* happens.​

#2

Should Apple update the MP6,1 with Haswell-EP?

Absolutely. If Apple doesn't, "death of Mac Pro" stories will be everywhere. Apple needs to show a commitment to the platform, and ignoring a CPU refresh is showing a lack of commitment.​

I agree fully. While the benefits of Haswell-EP may be minor to current MacPro6,1 owners, there are lots more current owners of only older Mac Pros. For them that minor increase over Ivy Bridge might be just what it takes to get them to purchase a Haswell MP (and there're likely some Windows system users who are similarly situated who might too jump to a Haswell MP, particularly if the GPUs are also upgraded and there are no driver and 4k issues).

 

[deconstruct60]

I think that there are two independent questions here that are being treated as one.

#1

Will the Haswell-EP systems offer greatly improved performance, so that most people with an MP6,1 will want to upgrade?
​

That is a deeply dubious question. The implicit implication of "so that most people with a MP6,1 will ..." is the highly suspect part. The overwhelming majority of folks who just bought a Mac Pro don't need a replacement since they just did a replacement.

I do think there is bound to be a rash of threads rationalizing ignoring an upgrade. However, the MP6,1 was good enough to upgrade to previously and frankly that doesn't particularly change when the next iteration comes along regardless of new Intel codenames or increments to "v" after the name.

The primary customers for a E5 v3 Mac Pro are customers still on Xeon 3xxx/5xxxx stuff (or in even deeper denial with PPC stuff). It isn't just the 3% percent between v2 and v3 they are missing. It is compounded, multi generational gap that are missing. 1.03 * 1.03 * 1.03 is a 9.2% jump. It isn't precisely linear.

#2

Should Apple update the MP6,1 with Haswell-EP?

Absolutely. If Apple doesn't, "death of Mac Pro" stories will be everywhere.
​

Those aren't the biggest threat. A much bigger threat are going to be competitive workstation sale folks who will trot out the notion that Apple is an unreliable systems vendor. Apple left EU customers hanging in the wind for almost a year because they went "dark" on updates. If customers upgrade cycle doesn't match Apples multi-year drought between updates cycle do they take lumps or sync up with Apple ?

It is also not particularly competitive at the higher end of the product line up. The 1680 v3 is likely substantially cheaper (i.e., priced as a 1600 series GPU) as opposed to the customize 2600 10 core the 1680 v2 was. The 1680 v2 was pushed out about a generation too soon to be economical.

Similarly they will be down 2 cores at the top end versus competitors. For folks that have 10-14 workload threads it will make a difference. Apple already lacks dual CPU package systems. Apple can bet on more cores per socket but then can't also walk away from more cores per socket at the same time.

Frankly waiting for E5 v4 or v5 isn't going to make any revolutionary difference on the single core, single thread, drag racing benchmarks which are largely coupled to clock speed. At the same price points, there probably isn't going to be a radical change in base clock. Running pentium 4 optimized code faster is not going to see major improvements over time. Waiting until the next update buys neither Apple or customers much on that front along this product line.

For Xeon E5 Intel is largely going to take process improvements to add more cores rather than primarily pure base clock speed.​

 

[Tutor]

That is a deeply dubious question. The implicit implication of "so that most people with a MP6,1 will ..." is the highly suspect part. The overwhelming majority of folks who just bought a Mac Pro don't need a replacement since they just did a replacement.

I do think there is bound to be a rash of threads rationalizing ignoring an upgrade. However, the MP6,1 was good enough to upgrade to previously and frankly that doesn't particularly change when the next iteration comes along regardless of new Intel codenames or increments to "v" after the name.

The primary customers for a E5 v3 Mac Pro are customers still on Xeon 3xxx/5xxxx stuff (or in even deeper denial with PPC stuff). It isn't just the 3% percent between v2 and v3 they are missing. It is compounded, multi generational gap that are missing. 1.03 * 1.03 * 1.03 is a 9.2% jump. It isn't precisely linear.





Those aren't the biggest threat. A much bigger threat are going to be competitive workstation sale folks who will trot out the notion that Apple is an unreliable systems vendor. Apple left EU customers hanging in the wind for almost a year because they went "dark" on updates. If customers upgrade cycle doesn't match Apples multi-year drought between updates cycle do they take lumps or sync up with Apple ?

It is also not particularly competitive at the higher end of the product line up. The 1680 v3 is likely substantially cheaper (i.e., priced as a 1600 series GPU) as opposed to the customize 2600 10 core the 1680 v2 was. The 1680 v2 was pushed out about a generation too soon to be economical.

Similarly they will be down 2 cores at the top end versus competitors. For folks that have 10-14 workload threads it will make a difference. Apple already lacks dual CPU package systems. Apple can bet on more cores per socket but then can't also walk away from more cores per socket at the same time.

Frankly waiting for E5 v4 or v5 isn't going to make any revolutionary difference on the single core, single thread, drag racing benchmarks which are largely coupled to clock speed. At the same price points, there probably isn't going to be a radical change in base clock. Running pentium 4 optimized code faster is not going to see major improvements over time. Waiting until the next update buys neither Apple or customers much on that front along this product line.

For Xeon E5 Intel is largely going to take process improvements to add more cores rather than primarily pure base clock speed.

deconstruct60,

Thanks for additional, specific reasons why Apple shouldn't walk away from Haswell.

 

[VirtualRain]

I think that there are two independent questions here that are being treated as one.

#1

Will the Haswell-EP systems offer greatly improved performance, so that most people with an MP6,1 will want to upgrade?

No. The improvement will be minor for most apps. If you have an app that benefits from AVX2, it might be worth considering. Otherwise, stick with your MP6,1 until a big improvement *on your workflow* happens.​

#2

Should Apple update the MP6,1 with Haswell-EP?

Absolutely. If Apple doesn't, "death of Mac Pro" stories will be everywhere. Apple needs to show a commitment to the platform, and ignoring a CPU refresh is showing a lack of commitment.​

I would add two additional questions to this we're trying to answer here...

#3

Should I buy a nMP now or wait for Haswell?

If you need a new workstation with 4-8 cores, there's no point in waiting... buy a 6,1 now. The only reason to wait is if a 12+ core system is a requirement where Haswell will likely offer some savings or more cores/$.
​

#4

Will Apple update the MP6,1 with Haswell-EP?

Of course they "should" update the MP6,1 with Haswell as many have discussed, however, there is no guarantee that they will. I think it's unwise to assume the update cycle revolves strictly around Intel on this product given the importance of the GPUs.

Apple effectively has three options:
1. Update the nMP when Haswell-EP becomes available regardless of available GPUs - Many people have stated why they should do this even though there are only incremental gains to be had
2. Update the nMP with Haswell-EP but only when next-gen GPUs become available from AMD - which would mean a delayed refresh but one that carries more impact
3. Skip Haswell-EP and wait for Broadwell-EP - I'm not sure this makes sense but they have clearly done this with the Mac Mini so it's possible, although perhaps not very likely.
​

 

[goMac]

Should I buy a nMP now or wait for Haswell?

If you need a new workstation with 4-8 cores, there's no point in waiting... buy a 6,1 now. The only reason to wait is if a 12+ core system is a requirement where Haswell will likely offer some savings or more cores/$.
​

I'd argue if you're concerned about GPU performance you should also wait.

 

[deconstruct60]

I would add two additional questions to this we're trying to answer here...

#3

Should I buy a nMP now or wait for Haswell?

If you need a new workstation with 4-8 cores, there's no point in waiting... buy a 6,1 now.​

this has little to nothing to do with core count. If actually need a Mac Pro ( missing out on business demands because don't have one) then should buy one.

Is an update immediately eminent because the Core i7 59xxk models have dropped? No. Apple isn't going to use Core i7's and the Xeon E5 v3 isn't necessarily immediately coming at the same time ( For Xeon E5 v1 they were displaced almost a whole quarter from the Core i7 release. )

if the current Mac Pro was not "fast enough" to warrant a purchase then it still isn't.

#4

Will Apple update the MP6,1 with Haswell-EP?

Of course they "should" update the MP6,1 with Haswell as many have discussed, however, there is no guarantee that they will. I think it's unwise to assume the update cycle revolves strictly around Intel on this product given the importance of the GPUs.​

Problem is that the GPUs have updated also. The new AMD GCN 1.1 products are OpenCL 2.0 capable and the current GPU cards are not. Several of the new ones may have DSPs ( depends up if dressing up mainstream or really getting FirePro variants ) that Apple might leverage so audio customers don't have underwear in a twist over two GPU cards.


The PCH has moved. the CPU has moved, the GPUs have moved, the memory tech has moved. Almost everything have moved on. Apple could snore on all that. But it is gross mischaracterization that only the x86 cores have progressed here. The Mac Pro 2013 was late..... waaaaaaaaaaaay late.
About the only the tech there that was bleeding edge was the TB v2 and then only because it is was in the ramp to volume production phase of manufacturing.

Waiting for E5 v4 is something Apple could do, but it is a pretty bonehead move.

Apple effectively has three options:
1. Update the nMP when Haswell-EP becomes available regardless of available GPUs - Many people have stated why they should do this even though there are only incremental gains to be had

Yet another bonehead option. If the GPUs were coming 8-9 months in the future then perhaps. But the new GPUs are already announced or released . The FirePro 8100/9100 are out:


http://www.amd.com/en-us/products/graphics/workstation

The 7100 is already announced and should ship relatively soon:

http://www.anandtech.com/show/8371/amd-firepro-w7100-w5100-w4100-w2100

News about Intel Xeon E5 v3 is leaking like a sieve right not but Intel actually hasn't officially announced them yet. In contrast, the new GPUs have already been announced. ( even if Apple switched over to Nvidia they announced the same time AMD did. )

It is likely only the delayed OS X drivers that are the only hold up. The hardware exists. It would be extremely ass-backwards to skip a system when hardware is available. Apple doesn't need to release anything right now anyway so even if need until January to finish up the drivers it isn't a problem.
[ Turning an update on the Mac Pro out in less than 12 months is baseless in need (they have only had a couple months of steady state supply-demand ) and actually would be setting unrealistic expectations for the future. ]

if the GPUs and CPUs drift largely out of synch then doing one without the other makes sense on the Mac Pro. But when upgrades launch inside of 3-4 months of each other, it doesn't much any sense to separate them. Especially on a product that would be about an approximately 12 month upgrade cycle. Going from 12 to 15 isn't the end of the world. Frankly with the rapid firmware fixes that followed the last Mac Pro release the bake time isn't excessively long here either anyway.

2. Update the nMP with Haswell-EP but only when next-gen GPUs become available from AMD - which would mean a delayed refresh but one that carries more impact

The basic hardware is out. What waiting on is only Apple's custom board hardware tweaks and drivers (the latter at low level is still AMD ).

3. Skip Haswell-EP and wait for Broadwell-EP - I'm not sure this makes sense but they have clearly done this with the Mac Mini so it's possible, although perhaps not very likely.

There is no 100% proof right now that Apple has skipped Haswell for the Mini. The Mini hasn't shipped but unless shifting to Core M CPUs there is marginal benefits that Broadwell is going to buy the Mini over a straightforward update of the Mini with Haswell.

Xeon E5 v4 (Broadwell-EP) isn't bringing magic pixie dust any more than delaying until Xeon E5 v2 brought magic pixie dust for the Mac Pro 2013. The delta from v3 to v4 isn't going to be huge. It will probably be almost the same delta this thread started off "complaining" about not being worthy of an upgrade. Apple stalling until late 2013 to synch up with Thunderbolt v2 didn't "buy" a whole lot one the x86 performance front over most of the line up. They got core count parity at the top end 12 cores ( same max core count as 2012 Mac Pro ) but not a huge leap over what Xeon E5 v1 would have done paired with TB v1.

 

[deconstruct60]

I'd argue if you're concerned about GPU performance you should also wait.

Very similarly to the CPU, the performance changes are going to be incremental. There is no general revolution coming. [ Unless there is a major change in breadth of Vendors offered , no revolution there either. ] There are some bigger future ( OpenCL 2.0 ) upticks than there is major performance bumps.

 

[Tutor]

When I was a child surprises were good, but few children need + can afford a MacPro.

I would add two additional questions to this we're trying to answer here...

#3
Should I buy a nMP now or wait for Haswell?

If you need a new workstation with 4-8 cores, there's no point in waiting... buy a 6,1 now. The only reason to wait is if a 12+ core system is a requirement where Haswell will likely offer some savings or more cores/$.

I agree whole-heartedly generally, but "need" is the key for both sentences and "need" needs to be tied down temporally.

Apple's potential customers for the MacPro6,1 and a Haswell MacPro are those with a need to have or own what the MacPro provides (including just the prestige of owning one) and have the financial resources to purchase one. They consist of those who've never owned a MacPro and those who have owned previous MacPro versions. As Aiden points out, those with a substantial understood need for AVX2 and now own a MacPro6,1 would be good candidates for Haswell MacPros, even if they perceive all the other advantages to be minimal. The same would apply to all other MacPro owners and others who do not own a MacPro, but to this group what a Haswell MacPro offers might not be perceived by some of them to be minimal.

Thus, if someone recognizes that he/she has a current need for a 4-8 core 2013 MacPro (i.e., all/much of that the 2013 MacPro has to offer), then they should purchase a 4-8 core 2013 MacPro now. As you state below, there is no guarantee that Apple will release a Haswell MacPro.

#4

Will Apple update the MP6,1 with Haswell-EP?

Of course they "should" update the MP6,1 with Haswell as many have discussed, however, there is no guarantee that they will. I think it's unwise to assume the update cycle revolves strictly around Intel on this product given the importance of the GPUs.​

True, but if the GPU cycle, over which AMD has lots more control than Apple, is widely out of line with new CPU releases, then, as deconstruct60 put it: "A much bigger threat are going to be competitive workstation sale folks who will trot out the notion that Apple is an unreliable systems vendor. Apple left EU customers hanging in the wind for almost a year because they went "dark" on updates. If customers upgrade cycle doesn't match Apples multi-year drought between updates cycle do they take lumps or sync up with Apple ?" Apple is too secretive about it's plans, such that outsiders don't know whether Apple is waiting on a component update, waiting to quit MacPro production, or whatever else. So Apple needs to timely speak up and/or timely deliver on a Maxwell MacPro. When I was a child surprises were good, but I, like most MacPro purchasers, am not now a child. Surprise releases or surprise lack of updates, rub me wrong. As an adult I plan significant purchases and Apple's competitor, including those who market computer parts, don't treat their potential customer like children. They don't play hide the ball.

Apple effectively has three options:
1. Update the nMP when Haswell-EP becomes available regardless of available GPUs - Many people have stated why they should do this even though there are only incremental gains to be had;

Those gains might appear to be incremental, at least, to those without a perceived need for AVX2 and (1) with 6,1s currently, (2) tricked out self-builds and (3) tricked out MacPro5,1 with X5680s and especially X5690s and recent GPUs (for whom the 6,1 may be perceived as simply incremental). I respectfully submit all these three groups together are smaller than those others with only (factory state) cMPs or pre-2009 MacPros (that may or may not have been modified) + those who are first time MacPro purchasers + those who have never owned a computer + those who have owned an iMac, Mini, etc. (but never owned a Mac Pro) + those who have been Windows or other OS users. Thus, Apple should release a Haswell MP for these individuals and businesses, regardless of whether AMD has made next-gen GPUs available (unless we're talking about a delay of only a few months). Also, Apple should promptly publicly announce its intentions and the reasons for its delay. This is particularly so because of Apple's history of wavering in this regard.

2. Update the nMP with Haswell-EP but only when next-gen GPUs become available from AMD - which would mean a delayed refresh but one that carries more impact;

No, but if Apple delays see prior response.

3. Skip Haswell-EP and wait for Broadwell-EP - I'm not sure this makes sense but they have clearly done this with the Mac Mini so it's possible, although perhaps not very likely.

No, but if Apple delays see prior responses.​

 

[VirtualRain]

Problem is that the GPUs have updated also. The new AMD GCN 1.1 products are OpenCL 2.0 capable and the current GPU cards are not. Several of the new ones may have DSPs ( depends up if dressing up mainstream or really getting FirePro variants ) that Apple might leverage so audio customers don't have underwear in a twist over two GPU cards.

I don't consider GCN 1.1 "next-gen"... Hawaii is just a small incremental improvement on Tahiti. We won't see true "next-gen" performance improvements until AMD moves to 20nm with a new architecture. While Apple might offer variants of the Wx1xx series in the next Mac Pro refresh, those are about as unexciting as Haswell and the word is that they are running hot which means binning for the nMP cooling system is likely to yield poor volumes = expensive or not worth it.

----------

Those gains might appear to be incremental, at least, to those without a perceived need for AVX2 and

I see a lot of mumbling about AVX2, but AVX2 is just a poor man's version of GPU acceleration which the nMP solves in a completely different way from typical mainstream Haswell solutions. AVX2 offers Gflops of FP performance compared to a GPUs Tflops. Anyone optimizing their code for AVX2 and not GPGPU is not developing apps for workstation users. I would expect you to be schooling me in this...

 

[VirtualRain]

Haswell-E doesn't have much over Ivy

I think one thing all the Haswell backers may be overlooking, is that Intel is on a tick-tock cycle with a Tick representing a die-shrink and a Tock representing a new architecture. Ivy Bridge was a Tick (die-shrink) that offered amazing new core counts on a single die. Hawsell is a tock that offers some new features and efficiency (and a bit of a price break) that effectively amounts to a pile of magic beans.

Let's face it, Intel's most recent tock microarchitecture home run came with Nahelem (way back in 2009) with the integration of the NB and elimination of the FSB. Their major focus on improvements in microarchitecture these days are on the IGP which is of little value to workstation users running one or more discrete GPUs. In the coming generations, the major gains are going to be on Ticks where a die-shrink can yield a lot more silicon to add cores within a given TDP budget. I think the smartest time to buy an Intel CPU is now on the Tick, not the Tock. Ivy Bridge is a Tick. Haswell is a Tock. Broadwell is a Tick. Expect real improvements with Broadwell. Hasswell's writing is on the wall.

 

[goMac]

I don't consider GCN 1.1 "next-gen"... Hawaii is just a small incremental improvement on Tahiti. We won't see true "next-gen" performance improvements until AMD moves to 20nm with a new architecture. While Apple might offer variants of the Wx1xx series in the next Mac Pro refresh, those are about as unexciting as Haswell and the word is that they are running hot which means binning for the nMP cooling system is likely to yield poor volumes = expensive or not worth it.

I don't think there is anything minor about the gains at all. They look like they're between 15%-20%, which is pretty decent:
http://www.tomshardware.com/reviews/firepro-w9100-performance,3810-12.html
http://www.tomshardware.com/reviews/firepro-w9100-performance,3810-13.html

A few of the benchmarks are more even, but more often than not the W9100 is showing very impressive gains.

As the link notes:
"For the first time since 2007, AMD has a FirePro-branded card based on a really big GPU. At 6.2 billion transistors, the Hawaii processor boasts 44-percent more logic than the FirePro W9000’s Tahiti chip."

The much higher transistor count is very hard to argue with.

Given how tight the performance is on the current cards, and how more 4k displays are on the horizon, a 20% gain in GPU power could be a make or break for some users.

The power draw and heat are in line with the W9000, and Apple will probably under clock a smidge like they did with the W9000.

Also:
"The FirePro W9100 is the first (and currently only) card that can drive up to six 4K monitors at full resolution, even if that means stepping down to 30 Hz when more than three are connected. A massive 16 GB of fast GDDR5 memory is more than enough for anything that you can throw at it."

Would be a nice alignment with the Mac Pro's six Thunderbolt ports.

 

[Tutor]

Hickory Tickery Tock - Applications and true "Need" regulate my clock

... .
I see a lot of mumbling about AVX2, but AVX2 is just a poor man's version of GPU acceleration which the nMP solves in a completely different way from typical mainstream Haswell solutions. AVX2 offers Gflops of FP performance compared to a GPUs Tflops. Anyone optimizing their code for AVX2 and not GPGPU is not developing apps for workstation users. I would expect you to be schooling me in this...

Some may be poor men/women because AVX and/or AVX2 is what the developer of their important application(s) chose to speed it up. I never said I preferred AVX(2) gains over GPGPU gains because those technologies, in the end, are relevant only in the context of one's important applications and though one might prefer one technology over the other in the abstract, one's limited by what the developer of the application does and the nature of the application.

Personally, I do prefer CUDA GPUs over CPUs because my important applications are 3d and video applications that take advantage of them; but I also recognize that GPUs (CUDA and OpenCL) may not perform important compute function for someone else's important application(s). Likewise, however, I also recognize that increasing numbers of CPU cores (with attendant decreasing core speeds) may not be important [and could, in fact, be a hinderance] to some else's important application(s). Some applications benefit more from fewer, but higher speed cores and make not take advantage of CUDA or OpenCL.

I think one thing all the Haswell backers may be overlooking, is that Intel is on a tick-tock cycle with a Tick representing a die-shrink and a Tock representing a new architecture. Ivy Bridge was a Tick (die-shrink) that offered amazing new core counts on a single die. Hawsell is a tock that offers some new features and efficiency (and a bit of a price break) that effectively amounts to a pile of magic beans.

Let's face it, Intel's most recent tock microarchitecture home run came with Nahelem (way back in 2009) with the integration of the NB and elimination of the FSB. Their major focus on improvements in microarchitecture these days are on the IGP which is of little value to workstation users running one or more discrete GPUs. In the coming generations, the major gains are going to be on Ticks where a die-shrink can yield a lot more silicon to add cores within a given TDP budget. I think the smartest time to but an Intel CPU is now on the Tick, not the Tock. Ivy Bridge is a Tick. Haswell is a Tock. Broadwell is a Tick. Expect real improvements with Broadwell. Hasswell's writing is on the wall.

Not only as to Haswell, but every CPU's writing is on the wall, at least until the GHz Ghost (constraining 4-5+ GHz speeds) constrain expansion of CPU core counts as well. What may be the smarted time for a purchaser to make a purchase because one wants the most cores possible may not coincide with the time one truly needs to make a purchase. One who can wait a year to make a purchase of something that hasn't even been announced nor had its specs revealed isn't likely truly in need of the item. And until the Ghost expands turf to include core counts, there'll always be more cores coming in the next, or the next, or the next ... release. So one could wind up in a continual period of waiting and in the end miss important opportunities for business expansion; and when they finally do purchase - only end up with a system with the highest priced, fastest CPU(s) - well, only until the next swing of the pendulum the following year or so.

Tickery Tock History

(1) The top Workstation Nehalems (1st Gen - Tock ) were 4-cores [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon W5590 - AT80602000753AA (BX80602W5590).html ] $1,600 = price at introduction;

(2) The top Workstation Nehalems (2nd Gen a/k/a Westmeres - Tick) were 6-core [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon X5690 - AT80614005913AB (BX80614X5690).html ] $1,663 = price at introduction;

(3) The top Workstation SandyBridge (1st Gen - Tock) were 8-cores [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon E5-2690.html ] $2061 (box) = price at introduction;

(4) The top Workstation SandyBridge (2nd Gen a/k/a Ivy Bridge - Tick) were 12-cores [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon E5-2697 v2.html ] $2618 (box) = price at introduction;

(5) The top Workstation Haswell (1st Gen - Tock) will be an 18-core [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon E5-2699 v3.html ] cannot yet find price for this one, but the pre-order price of Xeon E5-2697v3 (a 14 core with a base clock of 2.6 GHz) is $2763 vs. $2,673 for Xeon E5-2697v2 (at same store) for a current 3.4% price difference. [ http://www.cpu-world.com/news_2014/2014080502_Xeon_E5-2600_v3_CPUs_are_available_for_pre-order.html ].

 

[deconstruct60]

....
A few of the benchmarks are more even, but more often than not the W9100 is showing very impressive gains.
...
"The FirePro W9100 i.... A massive 16 GB of fast GDDR5 memory is more than enough for anything that you can throw at it."
...

But is Apple going to put 16GB of fast GDDR5 memory on their custom card? Their track record isn't hot in this area.

D300 2GB ( versus W7000 4GB )
D500 3GB ( versus not quite as equivalent W8000 4GB )
D700 6GB ( versus W9000 6GB )

One out of three isn't a spectacular track record.

To shave costs it wouldn't be surprising to see Apple dump half of that 16GB of VRAM. It isn't like they haven't done it before. At that point, some of benchmarks are going to pull back. Would be rather weird to have nominal configs with 32GB of VRAM and only 16GB of RAM. Those with extra money to throw at D710(?) probably have money for more regular RAM but the fact these comes in pairs.... that is a ton of VRAM.

 

[iBug2]

#2

Should Apple update the MP6,1 with Haswell-EP?

Absolutely. If Apple doesn't, "death of Mac Pro" stories will be everywhere. Apple needs to show a commitment to the platform, and ignoring a CPU refresh is showing a lack of commitment.​

I highly doubt anyone will be actually thinking that Mac Pro will be killed if Apple does not do a Haswell refresh. They just spent millions of dollars for a major redesign and a new manufacturing plant for this thing.

They should update to Haswell and new FirePro's in any case but this is not a reason why.

 

[VirtualRain]

Tickery Tock History

(1) The top Workstation Nehalems (1st Gen - Tock ) were 4-cores [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon W5590 - AT80602000753AA (BX80602W5590).html ] $1,600 = price at introduction;

(2) The top Workstation Nehalems (2nd Gen a/k/a Westmeres - Tick) were 6-core [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon X5690 - AT80614005913AB (BX80614X5690).html ] $1,663 = price at introduction;

(3) The top Workstation SandyBridge (1st Gen - Tock) were 8-cores [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon E5-2690.html ] $2061 (box) = price at introduction;

(4) The top Workstation SandyBridge (2nd Gen a/k/a Ivy Bridge - Tick) were 12-cores [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon E5-2697 v2.html ] $2618 (box) = price at introduction;

(5) The top Workstation Maxwell (1st Gen - Tock) will be an 18-core [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon E5-2699 v3.html ] cannot yet find price for this one, but the pre-order price of Xeon E5-2697v3 (a 14 core with a base clock of 2.6 GHz) is $2763 vs. $2,673 for Xeon E5-2697v2 (at same store) for a current 3.4% price difference. [ http://www.cpu-world.com/news_2014/2014080502_Xeon_E5-2600_v3_CPUs_are_available_for_pre-order.html ].

Thank you... That's interesting... I hadn't looked at it from a top core count perspective, but it does illustrate that you generally gain 50% more cores on a Tick (die-shrink) and only 33% more cores on an Tock (architecture upgrade) but the latest gen may change that.

The net of it all though, is that Intel is smartly pursuing the massive shift to mobile and the shrinking form factors and power efficiency that's required for that segment, and either focusing less on top end raw performance (or running up against the laws of physics) with each new generation.

Anand noted in his 2013 nMP review that the CPU is effectively fading into the sunset when it comes to substantial increases in performance, and Apple was smart in moving to dual discrete GPUs in order to offer up some significant performance gains and a vibrant future for their workstation product. And Haswell is just further proof that clock-for-clock generational improvements have pretty much run out of gas.

I clearly seam to be going against the grain here in asserting that a Haswell refresh is not a particularly good buying event (if it's even offered), but I guess there's plenty of people that think it's great and the Mac Pro business is doomed without it, so as an Apple shareholder I hope they do a refresh . However, I'm not sure my advice to Mac Pro newcomers will change which is... Buy now, because you won't be missing much if and when a Haswell refresh is offered.

 

[deconstruct60]

I think one thing all the Haswell backers may be overlooking, is that Intel is on a tick-tock cycle with a Tick representing a die-shrink and a Tock representing a new architecture. Ivy Bridge was a Tick (die-shrink) that offered amazing new core counts on a single die.

what drugs are you smoking?


Xeon 5500 4 cores max (tock)
xeon 5600 6 cores max ( tick ) delta 2
Xeon E5 8 cores max ( tock) delta 2 ( same delta as above).
Xeon E5 v2 10-12 cores max (tick) delta 2-4 [ the 12 was going to supersized die pattern ]
Xeon E5 v3 12-14 cores max (tock) delta 2

The "tock" is second generation on the same process tech. The 2nd generation is going to pick up design optimizations for that particular process (versus just shrinking what already got). It also a more mature process with better yields and can more easily tolerates larger dies. Adding more cores to larger dies is straightforward. TIcks are likely more profitable ( shrinks and wafer increase for same price) for Intel but singular primary mechanism for core count increase .... not.

Xeon E5 v4 may get another 4 count delta in the "if have to ask you can't afford it" models, but at the bulk of the product line it likely going to be a 2 just like last couple iterations. There is no broad spectrum, "amazing" core count revolution on a single shrink. It isn't in the track record.

Hawsell is a tock that offers some new features and efficiency (and a bit of a price break) that effectively amounts to a pile of magic beans.

Their major focus on improvements in microarchitecture these days are on the IGP which is of little value to workstation users running one or more discrete GPUs.

Right, because the precision gating and power control hasn't changed much in last two iterations. *cough* not. The IGP isn't part of the x86 microarchitecture. There is multiple stuff on the die at this point besides just the x86 cores.

Xeon E5 got PCIe v3 before the mainsream CPU did. It has a much higher lane count. Much larger caches. There is a significant amount of non-core logic that is consuming a rather large transistor budget on the Xeon E5's. No IGP but it isn't like there other non core logic is exactly the same as the mainstream CPU designs.

In the coming generations, the major gains are going to be on Ticks where a die-shrink can yield a lot more silicon to add cores within a given TDP budget.

Arm flapping and hand waving. Intel has added core counts on each tick/tock iteration. Intel has control over pricing. For $1,500-3,000 CPU packages can do alot without tip-toeing around process shrinks.

I think the smartest time to buy an Intel CPU is now on the Tick, not the Tock. Ivy Bridge is a Tick. Haswell is a Tock. Broadwell is a Tick. Expect real improvements with Broadwell. Hasswell's writing is on the wall.

Tick comes at the end of the chipset renewal cycle. Which means if buy top end Tick model then there is nothing left to upgrade to later. It is a dead end.

As the process shrinks become more difficult the Ticks are also going to be more highly variable in time. (i.e, Broadwell is how far behind schedule at this point? The next shrink isn't going to be any easier. )

Someone can pick "Tick" for similar reasons can choose 'tails" (versus "heads") on a coin flip. Something feel like doing. But some technological "ticks" are better than "tocks" is up there with some technological reason "tail" is better than "heads". Largely unmotivated.

 

[Tutor]

The 18-core Xeon E5-2699 v3 Is A Multicore Rendering Monster

... .

Aside 1 - Here's an example of how two Xeon E5-2699 v3 Haswells will perform in a 3d rendering application such as Cinema 4d (an application that craves all of the cores that you can throw at it): http://www.chiploco.com/xeon-e5-2699-v3-dual-cpu-benchmarks-35897/ . In Cinebench R15, all 36 cores/72 threads score 4,542 points. Thus, in single socket CPU system it should easily score more than 2,250+ points. For those whose businesses rely on core-hungry applications like Cinema 4d, that would be extremely enticing. One Xeon E5-2699 costs about $4,058.

Aside 2 - Probably not at all important to MacPro users (unless Apple underclocks them properly) is that like the Nehalems/Westmeres (and unlike the Sandy and Ivy Bridges), these CPUs have unlocked bins and can be significantly clock tweaked. But as this article [ http://www.chiploco.com/haswell-ep-xeon-e5-2699-v3-benchmarks-35799/ ] shows, it must be done properly and with appropriate coolers.

Update - A pair of them achieves Geekbench 3 scores of 3,354 (Single-Core) and 80,023 (Multi-Core) [ See http://browser.primatelabs.com/geekbench3/662846 ].

 

[goMac]

But is Apple going to put 16GB of fast GDDR5 memory on their custom card? Their track record isn't hot in this area.

D300 2GB ( versus W7000 4GB )
D500 3GB ( versus not quite as equivalent W8000 4GB )
D700 6GB ( versus W9000 6GB )

One out of three isn't a spectacular track record.

To shave costs it wouldn't be surprising to see Apple dump half of that 16GB of VRAM. It isn't like they haven't done it before. At that point, some of benchmarks are going to pull back. Would be rather weird to have nominal configs with 32GB of VRAM and only 16GB of RAM. Those with extra money to throw at D710(?) probably have money for more regular RAM but the fact these comes in pairs.... that is a ton of VRAM.

Yes, they could under clock and cut the VRAM amounts (I doubt they would cut VRAM or drastically cut VRAM speed by an entire memory class but they will likely under clock.) Given that today's cards are under clocked as well, the performance delta stays about the same.

 

[riggles]

I'm confused..

Maybe someone could help me connect the dots here.

What do the new Haswell-E CPUs have to do with the Mac Pro? Mac Pros have always used Xeon chips, so don't we need to wait for Haswell-EP specifics? Or is there some performance connection between E and EP?

 

[VirtualRain]

what drugs are you smoking?
...
*cough* not.
...
Arm flapping and hand waving.
...
Someone can pick "Tick" for similar reasons can choose 'tails" (versus "heads") on a coin flip.

Well, we may not always agree, although I think despite your deconstructive argumentative writing style, we actually agree on more than it might appear. But there is no doubt that your posts are more entertaining then mine!

Maybe someone could help me connect the dots here.

What do the new Haswell-E CPUs have to do with the Mac Pro? Mac Pros have always used Xeon chips, so don't we need to wait for Haswell-EP specifics? Or is there some performance connection between E and EP?

Intel's nomenclature is anything but crystal clear, but basically the Haswell-E, EP, EN and EX all share the same CPU core. So the CPU performance you see in the Haswell-E review is basically what you can expect from the EP that might end up in a Mac Pro refresh.

 

[Tutor]

E ~= EP with just fewer, but higher clocked cores, ignoring tweakability and QPIs

Maybe someone could help me connect the dots here.

What do the new Haswell-E CPUs have to do with the Mac Pro? ... . is there some performance connection between E and EP?

In sum, as VirtualRain points out, the CPU core in Es and EPs are essentially the same. Here's a little more detail.

As a self-builder and clock tweaker what I like about the Nehalems and Westmeres is that, when similarly spec'd, the Es and EPs are both clock tweakable to the same extent. Moreover, what you get in performance from an Nehalem/Westmere E is about as close to that you get from a similarly spec'd and clocked Nehalem/Westmere EP as is the performance delta between comparing two Es or two EPs similarly spec'd and clocked. You can run an E on an EP dual CPU motherboard (even a dual CPU pro/enterprise motherboard), just not two E's together properly because a QPI link is disabled; moreover, you can run an EP on an E motherboard (a single CPU enthusiast motherboard). Some E motherboards will even support an EP with ECC memory running properly. Motherboard bios and structure and OS compatibility are key to all of this compatibility. In fact, almost all CPU chips from the same family are essentially same, with distinctions being mainly derived from binning (chip selection - sorting for, among other things, performance related capability) and Intel's just shutting things down.

Clock tweakability has been severely limited by Intel in the EP versions of Sandy and Ivy Bridge Xeons. Of course, I don't like that. I've learned that the clock tweakability attribute of the Nehalem/Westmere EPs has, at least to some extent, been restored in, at least some, Haswell EPs, but I haven't yet had the opportunity determine the extents for myself.

Mac Pros have always used Xeon chips, so don't we need to wait for Haswell-EP specifics?

No. Futhermore, we no longer have to wait for the specifics. Again, here's the most accurate spec list for the Haswell EP lineup: Click this link [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon E5-2699 v3.html ] and scroll to the bottom of the page; then you can click those links for the individual Haswell EP chips to learn more about their specs. Some prices are here: http://www.cpu-world.com/news_2014/2014080502_Xeon_E5-2600_v3_CPUs_are_available_for_pre-order.html . These chips have already been advertised for pre-release ordering. Networking Hardware appears to have a few of most models for sale now - See, e.g.
[ See, e.g., http://www.ebay.com/itm/Intel-Xeon-...14-Core-230-GHz-BX80644E52695V3-/111444885728 ];
[ http://www.ebay.com/itm/Intel-Xeon-...107?pt=LH_DefaultDomain_0&hash=item1c452be9db ];
[ http://www.ebay.com/itm/Intel-Xeon-...292?pt=LH_DefaultDomain_0&hash=item1c452bcf3c ];
[ http://www.ebay.com/itm/Intel-Xeon-...948?pt=LH_DefaultDomain_0&hash=item1c452bb28c ];
[ http://www.ebay.com/itm/Intel-Xeon-...790?pt=LH_DefaultDomain_0&hash=item1c452b9a7e ];
[ http://www.ebay.com/itm/Intel-Xeon-...240?pt=LH_DefaultDomain_0&hash=item1c452b5dc0 ];
[ http://www.ebay.com/itm/Intel-Xeon-...002?pt=LH_DefaultDomain_0&hash=item19f2a12b1a ];
[ http://www.ebay.com/itm/Intel-Xeon-...953?pt=LH_DefaultDomain_0&hash=item19f29f2361 ];

 

[VirtualRain]

Again, here's the most accurate spec list for the Haswell EP lineup: Click this link [ http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon E5-2699 v3.html ] and scroll to the bottom of the page; then you can click those links for the individual Haswell EP chips to learn more about their specs.

I just got around to browsing this now... and I think it's very interesting and presents another twist on whether we might see a Haswell update for the nMP...

Did anyone notice the TDP ratings of these CPUs?

Many of the new 12+ core monsters in that table and even the 1600 v3's may be out of consideration for a nMP refresh simply based on having TDP > 130W. In fact, there's not a 6 or 8 core Haswell EP within the 130W envelope. This looks like a bit of a regression in thermal performance.

 

[Ender17]

clock for clock and core for core isn't where the big performance increase will come, it's the increased core count for the high end chips

 

[AidenShaw]

I just got around to browsing this now... and I think it's very interesting and presents another twist on whether we might see a Haswell update for the nMP...

Did anyone notice the TDP ratings of these CPUs?

Many of the new 12+ core monsters in that table and even the 1600 v3's may be out of consideration for a nMP refresh simply based on having TDP > 130W. In fact, there's not a 6 or 8 core Haswell EP within the 130W envelope. This looks like a bit of a regression in thermal performance.

WTF is the "130 watt envelope"?

Do you really think that the Apple engineers designed a system that was perfectly happy with a 130 watt CPU, but can't work with a 140 watt CPU?

If true - they should all be fired for making a system so close to the edge that it's impossible to upgrade to the next generation.

On the other hand, the existing system might take the v3 CPU with 8% more power draw without any issues - and just ramp the fan up a bit as the CPU temps warrant. Probably also a non-issue to add 10 watts to the power supply capacity.

VirtualRain - I really don't know what your agenda is with this thread. Either Apple updates the MP6,1 with Haswell E5-v3 CPUs, and makes everyone else happy - or they let it stagnate, which will make you happy.

 

[Tutor]

The GHz Ghosts May Be Becoming Legion And May Be Migrating

I just got around to browsing this now... and I think it's very interesting and presents another twist on whether we might see a Haswell update for the nMP...

Did anyone notice the TDP ratings of these CPUs?

Many of the new 12+ core monsters in that table and even the 1600 v3's may be out of consideration for a nMP refresh simply based on having TDP > 130W. In fact, there's not a 6 or 8 core Haswell EP within the 130W envelope. This looks like a bit of a regression in thermal performance.

I noticed all of those TDP ratings and I am a little concerned. The E series are said to run hot and to consume massive amounts of power - so much so, that EVGA's high end X99 motherboard has 2 eight pin power connect points to feed one Core i7-5960X up to 600 watts of dedicated power to fully over clock it [ http://www.newegg.com/Product/Product.aspx?Item=N82E16813188163 ] . My take from the reviews I read yesterday of the top end $4K+ Haswell EP model is that it's one hot chip, in more ways than one. It might be that a wise person who observed regarding the [Haswell] EPs, "... the word is that they are running hot ... " and said, "... running up against the laws of physics" might be either correct on both of these scores are, at least, headed in the right direction. That wouldn't bode well for the near future, whether the time clock ticks or tocks, unless the Ghosts can be banished or, at least, better managed. That would affect all brands, but most likely, the nMacPro sooner and more than most. So Apple would probably have to lead the way in finding a solution if Apple is fully committed to continuing the MacPro line in a cylinder with one fan. Obviously, I believe (1) that Apple's going back to a multi-fan cheese grater design is the first step in combating thermal issues and (2) that in the long run the old Mac Pro design is preferable to skipping CPU upgrades and downclocking dual GPUs and maintaining a single, down-clocked CPU system.