The "I'll wait for Haswell Mac Pro" Thread

[scottrichardson]

So, I was all keen on buying a nMP. Super keen. For years. But now, after seeing some benchmarks and real world results over at Macworld, on the 8 core 3GHz system, I have decided to hold off. The improvements over my 2009 Mac Pro aren't great enough. Noting I have RAID 0 SATA 6Gb/s SSDs pulling real-world 1040MB/sec, faster than the Mac Pro and my Geekbench CPU score is around 17,000.

Anyone else here gone down the same path as me? WAS going to get one, but now will wait until the next one comes out some time in 2014? (we hope).

Scott

 

[resotek]

No worries if you decide to wait, but the HDD I/O is pretty much a wash between the two systems. I would argue that one system is doing it without jumping through RAID hoops (and issues). Also, the nMP 8 seems to be running ~10% faster before factoring in the heavy lifting GPGPU will provide (and not measured at all by Geekbench). In short, I would expect the same nMP to get faster and faster over time as more apps (including OS X) utilize GPGPU.

 

[MattDSLR]

I'm with you
I like the look of new MP mini, but is way to limited as to upgrade option and the increase in speed is way to small compering to my 5.1 and in a single core processing all my macs run faster. I'm dreaming that apple will re-introduce MacPro in 2015 with some more upgrade options and much faster processors, drives and more ram

 

[scottrichardson]

Are the Haswell XEON's purported to provide any substantial gains in CPU performance? I know they will be touting power savings, but I'm not really interested in that. I'd actually like to see the GHz go up a bit

 

[peabo]

Are the Haswell XEON's purported to provide any substantial gains in CPU performance? I know they will be touting power savings, but I'm not really interested in that. I'd actually like to see the GHz go up a bit

Unlikely, although there's always the chance of more cores per chip. CPU speed will not increase dramatically as you rightly said Haswell is more about power saving than speed increase. There will be new GPUs to look forward to though where there will likely by more of a boost in terms of performance.

 

[mrsavage1]

They might skip haswell. The previous mps were updated around every 2 years

 

[Umbongo]

They might skip haswell. The previous mps were updated around every 2 years

They only skipped Sandy Bridge-EP out of all the CPU updates since 2006. I doubt they will skip Haswell, that just gives the impression they aren't bothered again and that it isn't worth investing in OS X/Macs for professional work.

 

[Cubemmal]

I'm buying it more for the size, power, noise, Thunderbolt and USB3 than the performance. My 2009 (now relegated to BootCamp duty) still is plenty fast for what I do. However I don't think Haswell will offer you significant improvements, basically you'll get more cores for cheaper most likely. When Haswell comes out I'll probably upgrade to it, and give the 2009 oMP to my kid for gaming.

So sure, if you can wait, then wait. It'll probably be two years though.

 

[barkmonster]

No worries if you decide to wait, but the HDD I/O is pretty much a wash between the two systems. I would argue that one system is doing it without jumping through RAID hoops (and issues). Also, the nMP 8 seems to be running ~10% faster before factoring in the heavy lifting GPGPU will provide (and not measured at all by Geekbench). In short, I would expect the same nMP to get faster and faster over time as more apps (including OS X) utilize GPGPU.

PCIe flash is certainly using some form of RAID 0 with multiple NAND chips. The only difference between a dual 6Gb/s PCIe SATA card with 2 SSDs installed and a direct PCIe SSD is the direct one has the controller hardware integrated with the NAND on the PCIe card itself.

I'm quite happy that assuming they update the Mac Mini with Haswell, I'll be able to buy a faster equivalent to the current 2.6Ghz i7 model, then add 2 Samsung 840 Pros for comparable performance to the 900Mb/s+ of the PCIe flash in the 2013 Mac Pro and have comparable CPU power to an E5-1620.

Based on the current 2012 Mac Mini, it would be just under £1,200 but could be bought in stages with the 1Tb HDD the Mac Mini it comes with going in an old SATA enclosure for backing up.

 

[flat five]

So, I was all keen on buying a nMP. Super keen. For years. But now, after seeing some benchmarks and real world results over at Macworld, on the 8 core 3GHz system, I have decided to hold off. The improvements over my 2009 Mac Pro aren't great enough. Noting I have RAID 0 SATA 6Gb/s SSDs pulling real-world 1040MB/sec, faster than the Mac Pro and my Geekbench CPU score is around 17,000.

Anyone else here gone down the same path as me? WAS going to get one, but now will wait until the next one comes out some time in 2014? (we hope).

Scott

that's not 'waiting for haswell'... more like using your 2009 for its intended lifespan instead of wasting money on replacing too soon.
so yeah, good decision

 

[Rich.Cohen]

The improvements over my 2009 Mac Pro aren't great enough.

If I had a 2009 that met my current needs, I'd hold off simply because I like to avoid the first round of a new design. I've got a Mac Pro 1,1 that I was going to replace last year, but put off because the 5,1 wasn't that big an improvement. Different circumstances, different choice.

 

[Macsonic]

So, I was all keen on buying a nMP. Super keen. For years. But now, after seeing some benchmarks and real world results over at Macworld, on the 8 core 3GHz system, I have decided to hold off. The improvements over my 2009 Mac Pro aren't great enough. Noting I have RAID 0 SATA 6Gb/s SSDs pulling real-world 1040MB/sec, faster than the Mac Pro and my Geekbench CPU score is around 17,000.

Anyone else here gone down the same path as me? WAS going to get one, but now will wait until the next one comes out some time in 2014? (we hope).

Scott

Scott. I think it's a prudent decision on your part as sometimes the processor performance increase may be incremental in relation to purchasing costs until we get hands on experience. I am using a 2009 4.1 8 core 2.66ghz and was able to compare its speed on a specific task with my client who has a 5.1 2012 Mac Pro, 2.66ghz at 12 Cores. I used After Effects to render a short movie and used the same file on both machines and used a stopwatch to time the completion. It's not a side by side comparison as the 12 core was at my client's studio. In rendering time, the 12 core 2.66ghz was faster but roughly only by 10% compared to the 8 core 2.66ghz. The speed difference was not day and night. Both machines had 24g ram though the 12 core Mac Pro's ram is at 1333Mhz We know the Geekbench of the 2009 8 core 2.66ghz is around 16,000+ while the 12 core 2012 2.66ghz is around 21,000+ .

To be clear, I am not suggesting we ignore Geekbench stats. I know there were postings in this forum that Geekbench only reflects CPU performance and not the whole computer in general performance. And also the speed test I did is not conclusive nor may not be a general reference for every other pro users. Certain softwares respond to CPU, GPU, ram differently for different tasks and users' workflow varies. For the benefit of the doubt, best to gather a variety of info aside from Geekbench when we are considering buying a new machine and wait for the new model to be released and shipped out and test it out.

Your 2.93ghz 8 core is still a capable machine because I had experiences using that model and I am not just looking at specs on paper.I will also be waiting for version 2 with or without Haswell cpus and see how things go from there.

 

[macines]

If you don't need one, don't buy one.

End of story.

 

[scottrichardson]

Yep, you're all right.

I can do EVERYTHING I need to do on my 2009 Mac Pro - in fact it does all the things I need to do REALLY well. Except for when Adobe CC bugs out, like today!

 

[wallysb01]

Unlikely, although there's always the chance of more cores per chip. CPU speed will not increase dramatically as you rightly said Haswell is more about power saving than speed increase. There will be new GPUs to look forward to though where there will likely by more of a boost in terms of performance.

There will likely be 2 more cores almost across the board with Haswell (so probably no more 4 cores, maybe just 1 specialty 2600, and the top end is up to 14 cores) and maybe small GHz bumps (i.e. .1). The die shrink on Haswell, Broadwell, will bring the big core counts though, where we’re going to get up to 18 cores on a single CPU. The top GHz is going to continue to be stuck at about 4 though.

I think this is what Apple was looking ahead to when deciding to drop to a single socket machine. The number of folks that will need >18 cores is going to be pretty small. Of course you could always get better performance from a machine with 2 10-cores than a machine with a single 18 core CPU, but still.

 

[thekev]

Are the Haswell XEON's purported to provide any substantial gains in CPU performance? I know they will be touting power savings, but I'm not really interested in that. I'd actually like to see the GHz go up a bit

It's probably more an increase in core count at the lower level configurations. At least the slides that have shown up on tech sites suggest this, and they have been a fairly good indication in the past. I suspect you are looking at more like 2015 though. Intel hasn't exactly maintained a 12 month refresh on those cpu lines.

They might skip haswell. The previous mps were updated around every 2 years

That is a ridiculous assertion based on false information. It will probably be more than a year before you have Haswell EP Xeons shipping in quantity. Even beyond that Apple ships when they ship. What makes it more ridiculous is that it shortens the number of cycles they get out of a given logic board configuration. Beyond that Haswell is expected to go to a hex core for whatever they call the E5-1620 v3. Right now that spot is occupied by a quad core. Ignoring a 50% increase in core count would be silly.

 

[peabo]

Well we already have Haswell E3 Xeons as detailed here:

http://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors#Xeon_E3-1xxx_v3_.28uniprocessor.29

Compare them to Ivy Bridge E3s to see what kind of improvements we're likely to see with the Haswell E5s when they surface.

 

[Macsonic]

Yep, you're all right.

I can do EVERYTHING I need to do on my 2009 Mac Pro - in fact it does all the things I need to do REALLY well. Except for when Adobe CC bugs out, like today!

I am still using the old Adobe CS 5 and no plans yet in migrating to CC. There's a growing dissatisfaction with Adobe CC from subscribers in this DISCUSSION. Though I think other users will still continue using CC.

 

[deconstruct60]

It's probably more an increase in core count at the lower level configurations.

In the Xeon E5 product line ( which the Mac Pro will probably continue to stick with).

That is a ridiculous assertion based on false information.
It will probably be more than a year before you have Haswell EP Xeons

Xeon E5 ( "ep" not sure why need code names when the product name isn't a secret) will go to v3 ( Haswell). However, there precedence that the Xeon E7 line will skip Haswell microarchitecture and go straight to Broadwell. Exactly what happened on this last iteration where E7 skipped Sandy Bridge all together. E7's are in the ultra high core count area and there isn't much of a die shrink when optimize at a fixed process size. Xeon E5 v2 ("Ivy Bridge") and E5 v3 ("Haswell") share the same fab process size. There may be marginal die shrinkage due to marginally better layout, but it isn't a "die change" transition of tick/tock. It is a architecture major change ( "tock") move.

Pragmatically have gotten to stage where E7 need "tick" transitions to pack in the additional core count. (Plus the major competitors: Sparc , Power, ... just don't move that fast. )

What makes it more ridiculous is that it shortens the number of cycles they get out of a given logic board configuration.

this 2013 Mac Pro's logic board is a dead end. v2 (Ivy Bridge) is a "tick" (and completes the tock/tick cycle in this space where the chipset stays the same). Another reason doubtful Apple will have another Mac Pro in 2014 given just barely have the current one out the door in any substantive quantity.

Beyond that Haswell is expected to go to a hex core for whatever they call the E5-1620 v3. Right now that spot is occupied by a quad core. Ignoring a 50% increase in core count would be silly.

The 1650/1660 might go to 8 for a 6 , 8 , 8 line up in the < $1,050 CPU range. Apple wouldn't need to use a 1680 ( which is priced in the E5 2600 upper half range )

 

[deconstruct60]

No worries if you decide to wait, but the HDD I/O is pretty much a wash between the two systems.

if a new update brought 2 PCI-e SSD that would be a big leap if aggregated them together. Whether the new 610 chipset brings more bandwidth for that to happen is up in the air (as far as CPU bandwidth it is the same E5 v2 and v3 will have 40 PCI-e v3 lanes. No changes). The current is a bit over subscribed. If Apple can uncork that with better matched chipset support there will be more headroom.

Also, the nMP 8 seems to be running ~10% faster

Can get into qualms about how big significant is but this is if running up to the limits of the machine. If average utlizitation is in th30-80% range then no it isn't.

Intel's move aren't going to be broadscale improvements much better than this. It is a 2-3 decade old instruction set. It has been optimized. There isn't much low hanging fruit. There are lots of apsects that all have to move forward to get bigger jumps. Every 1.5 year 10% boost cadence adds up to over 40% 6 years which is pretty big jump.

However, that is broad spectrum performance. Over narrower ranges ( e.g., crypto , SIMD/SSE/AVX) the short term jumps are far more substantive. E5 v2 on modern hand optimized AVX instructions is a huge gap on the older 3500/3600/5600 CPUs.

. In short, I would expect the same nMP to get faster and faster over time as more apps (including OS X) utilize GPGPU.

As noted above don't even have to leverage the GPU. Just not old code libraries. AVX support in E5 v3 will be more well rounded and allow compilers to do better job.

I don't even think Apple has even started to touch Haswell architecture's transaction memory TSX features at all.

the GPGPUs will play a roll but it is also more than abit overblown that is all that is being new computationally in the new (and upcoming) systems.

 

[LongSticks]

So, I was all keen on buying a nMP. Super keen. For years. But now, after seeing some benchmarks and real world results over at Macworld, on the 8 core 3GHz system, I have decided to hold off. The improvements over my 2009 Mac Pro aren't great enough. Noting I have RAID 0 SATA 6Gb/s SSDs pulling real-world 1040MB/sec, faster than the Mac Pro and my Geekbench CPU score is around 17,000.

Anyone else here gone down the same path as me? WAS going to get one, but now will wait until the next one comes out some time in 2014? (we hope).

Scott

Scott a couple of queries on your OP

Was this the article, you are referencing - dated 20.12.2013 - http://www.macworld.co.uk/mac/reviews/?reviewid=3455533

If so, is this the section re real world results? -

New Mac Pro 2013: benchmarks
We haven't been able to test a Mac Pro ourselves yet, but we have seen another system with a single 8-core, 3.4GHz Xeon E5-2687W processor. This is a Windows-based workstation, but we've seen comparable scores between Macs and Windows PCs in the past using the Cinebench benchmark – which is based on Maxon's Cinema 4D 3D animation suite.

Running Cinebench's 3D rendering test – which is almost exclusively a measure of CPU performance – we saw a score of 14.04 points (a measurement that only applies to Cinebench scores and has no wider context). This is actually 6.8% slower than the 12-core Mac Pro we reviewed back in 2010, which obtained a score of 15.07– though we expect the 12-core Mac Pro to be significantly faster. It's also 47.5% slower than the 26.78 score that the same Windows workstation with two Xeon E5-2687W chips installed.

If so this real isn't real world testing, these are implied results from a windows system by my reading?! We don't know the architecture, system memory etc of the windows system as it's not quoted?

If your OP is from this article dated 19.12.2013 - http://www.macworld.com/article/2082022/the-new-mac-pro-first-impressions.html

And this section -

But in the meantime we can say that the first time we ran GeekBench 3, the eight-core Mac Pro got a single-core score of 3599 and a multi-core score of 25997. For comparison, a 12-core 2012 Mac Pro (two 2.4GHZ Xeon processors) had GeekBench 3 scores of 2142/22886, and a 2013 27-inch quad core iMac with a 3.5GHz Core i7 processor GeekBench 3 scores of 3859/14705.

24% increase in geekbench score, I would say is significant? I agree the blackmagic results are surprising, but then again we haven't seen the 1tb results and there are rumours that the results scale with size?
I'm really not being provocative. I have posted on here before....as much as I love the nMP my first Pro may still be an over spec'd 5,1 from Scrumpymacs!
What's cursing my decision as well is the USB 3 and TB2 out of the box and future proofing my needs for the next 3 years!

Just trying to understand results and feelings between the old and the new before I jump!

 

[scottrichardson]

Scott a couple of queries on your OP

But in the meantime we can say that the first time we ran GeekBench 3, the eight-core Mac Pro got a single-core score of 3599 and a multi-core score of 25997. For comparison, a 12-core 2012 Mac Pro (two 2.4GHZ Xeon processors) had GeekBench 3 scores of 2142/22886, and a 2013 27-inch quad core iMac with a 3.5GHz Core i7 processor GeekBench 3 scores of 3859/14705.

24% increase in geekbench score, I would say is significant? I agree the blackmagic results are surprising, but then again we haven't seen the 1tb results and there are rumours that the results scale with size?
I'm really not being provocative. I have posted on here before....as much as I love the nMP my first Pro may still be an over spec'd 5,1 from Scrumpymacs!
What's cursing my decision as well is the USB 3 and TB2 out of the box and future proofing my needs for the next 3 years!

I guess it comes down to how much of a jump in performance I will honestly see?

With the iMac posting a HIGHER single core score, it makes me wonder how often we're 'feeling' the speed of that single core speed, and how often we're feeling the effects of multiple cores. In general 'use', surely the snappiness etc is more correlated to that single core frequency than multiple cores? I'm not sure how far developed Grand Central Dispatch is yet, so I'm making a lot of assumptions here.

Sure, Photoshop will feel 'smoother' using more beefy GPU's. And if Photoshop can make use of dual D300s, D500s and D700s, then that would be interesting.

I think I really need to take one of these nMPs for a spin at an Apple store with Photoshop running, and a bunch of other apps, with a couple of screens connected, before I can make a final judgement call.

 

[theSeb]

I am honestly surprised how many people are willing to make purchasing decisions based on one very synthetic benchmark and it seems to have become the gospel around here with multiple threads on the topic. At best GB only measures your computer's sprinting ability using very specific criteria. It does not produce a realistic workload , which is more like a marathon in comparison.

 

[LongSticks]

I guess it comes down to how much of a jump in performance I will honestly see?

With the iMac posting a HIGHER single core score, it makes me wonder how often we're 'feeling' the speed of that single core speed, and how often we're feeling the effects of multiple cores. In general 'use', surely the snappiness etc is more correlated to that single core frequency than multiple cores? I'm not sure how far developed Grand Central Dispatch is yet, so I'm making a lot of assumptions here.

Sure, Photoshop will feel 'smoother' using more beefy GPU's. And if Photoshop can make use of dual D300s, D500s and D700s, then that would be interesting.

I think I really need to take one of these nMPs for a spin at an Apple store with Photoshop running, and a bunch of other apps, with a couple of screens connected, before I can make a final judgement call.

Totally agree. I'm one of the weird ones who has to run Windows in my workflow and with the Autodesk software I use - it's all single core performance. I've even started looking at alternative software that uses multicore so that I can fully utilise the nMP. I'm so loathed to take logical move back to window machine as I never have driver, virus, software issues inside Bootcamp.

I think im looking at this decision process over the machine - order one, set it up using my workflow, run it and then keep or return within 14 days depending on my real world experience. It's playing games with Apple ordering - but only way I'm ever gonna know for sure with such a radical system change.

 

[d-m-a-x]

no way dude, i've been waiting since 200 fricking 9 for a good upgrade. At that rate i''ll be retired before MP 9,1 comes out