Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)
- Thread starter Mago
- Start date
-
- Tags
- macpro 2016
- Sort by reaction score
TDP went up along with computing power. Ivy (130W), Haswell (140W) and Broadwell (140W). And it wasn't just TDP, but power consumption on average rose with Haswell a lot and made it hotter chip than Ivy on Xeon EP series. Haswell was Intel's' failure on Workstation/Server side (also incorporated some bugs), the production samples were actually better than the final chip.
It really could be the story of the infamous tube.. it's thermal limitation were too much of a problem to shoehorn Haswell Xeon in it. Both failed, Apple and Intel.
Last edited:
Same here, I've got an 2010 cMP, so I'm not really in a hurry to buy a new system because there are many well known options to upgrade the system. And I expect some years of service from my 2010 cMP … but not everyone is in this comfortable situation where we can lean back and see what Apple's going to offer next.
There are folks out here with older cMP that are really getting old—as is "may fail any time without prior notice". These systems start to get a liability, because they can fail anytime and there is no reasonable replacement system. It's not that I think the nMP is a bad replacement, but this thread really gives me the creeps.
I like the nMP as a general design, but being forced to buy a system with such a potential issue, I'm glad I can stay a little longer on my old 5,1 cMP which does not lock at random.
The point is, some of us are lucky to have a recent cMP, but many others are in a quite dire situation. If my cMP would fail right now, I would rather consider a rMBP or an iMac than a nMP.
AVX2 is a full set of arithmetic operators (both integer and float) using 256-bit registers.
It can do 8 32-bit or 4 64-bit operations per instruction.
Think of it as loop unrolling done in hardware - a powerful optimization suitable for lots of mathematical operations. An AVX-2 aware compiler could use AVX-2 instead of (or in addition to) traditional unrolling.
AVX-2 also has a fused multiply-add, so it can do even more arithmetic per instruction.
Last edited:
Sigh. The rise in TDP has been covered for years on this forum, it is puzzling how the same propaganda still keeps popping up. Starting with Haswell Intel integrated a Voltage regulator into the CPU package. This is far more a 'move' than an 'increase'. Ivy had voltage regulators too. They just were not on the chip package. So that contribution of TDP went to the overall system. There is more localized TDP budget with Haswell but it isn't an overall net increase.
In overclocking contexts ( which Apple has no design requirements for ), the integrated voltage regulation isn't gong to scale as well. For Macs, it is a non issue unless want to run the hardware completely out of spec.
Failure? The Server (workstation) division of Intel is what is keeping the profits afloat for the last couple of years.
The "I want to hot-rod my tower" and the workstation market are not the same set. There is no large, dominant intersection of those two sets either. The hot-rod crowd tends to be more vocal in forums like these, but they aren't the market drivers they think they are.
Haswell (and no Broadwell) is too hot so Apple hasn't been able to update the Mac Pro 2013 is just pure hooey. It is a contrived excuse for Apple's inaction and/or rationalization why it isn't worth upgrading the MP you have ( 2010-2013 model).
Some quotes about desktop Haswell:
---
"That’s a shock really! At full load the Haswell-based (Core i7-4770K) system needs almost 30 watts more power than the Core i7-3770K configuration. So, even though low-voltage Haswell-based CPUs for mobile gadgets are very economical, their desktop cousins are a completely different story. Intel has optimized the new microarchitecture for ultra-mobile applications but the desktop Haswell is no good in terms of performance per watt. Not all Haswell-based CPUs are energy efficient, as we can see.
There is one thing that should be taken into account, though. The Haswell’s power consumption and heat dissipation go up rapidly at full load which is generated by our LinX-AVX utility." - x-bit labs
---
"Intel Xeon E3-1240 V3 Power Consumption
Power consumption we saw is a bit of a trade-off with the Haswell generation. In our Intel Core i7-4770 review, we saw the Haswell generation provide noticeable idle power consumption and also a bit higher load power consumption. This chart highlights a few key aspects to this Haswell generation. First, min to max power consumption has a larger delta than in previous generations. We recorded a 76w difference with the Haswell part but only a 60w difference with the Ivy Bridge part." - serverthehome.com
"The Haswell CPU core temperatures are seriously higher than those of the previous generation processors. And although most every-day tasks do not cause the CPU to heat up so dramatically, we should base our conclusions primarily on specialized stability tests, which create heavy but nevertheless quite realistic load." - x-bit labs
---
"Retail versions of Intel Haswell processors are hotter and more power-hungry than pre-production chips, and can’t be overclocked to the same speeds, British PC manufacturers have told PC Pro.
Four companies have told us that systems built using retail versions of Haswell chips haven't been able to match the speeds seen on pre-production models, which are given to manufacturers for testing before the official launch. All four PC makers spoke to PC Pro on condition of anonymity." - alphr.com
---
"Another manufacturer claims that retail Haswell parts are proving too hot to handle, stating that even while running at stock speeds the chips reach higher temperatures than the pre-production engineering samples - by around 15 degrees Celsius, according to the unnamed company's tests - or even the retail models of Intel's last-generation Ivy Bridge chips." - bit-tech.net
[doublepost=1471798415][/doublepost]So in addition to the increased power consumption and heat, Haswell could have delivered up to 7% more computing power, but in most cases 4-5%.
For Apple it would have require a new motherboard and some tweaking for cooling and PSU. For few percent increase in computing capabilities and the design cost I'm sure Apples saw that they couldn't make profit with Haswell EP. Their tube design was betting that both Intel and AMD will offer lower power chips, but both failed. So Apple skipped the Haswell.
[doublepost=1471798758][/doublepost]About the bugs in Haswell:
http://www.anandtech.com/show/8376/...rratum-found-in-haswell-haswelleep-broadwelly
[doublepost=1471800945][/doublepost]Of course this doesn't take away the thing that Apple should have lowered Mac Pro's price last winter. Just to give a sign that they don't think Mac users are idiots. Now we have the impression that they do...
Last edited:
It is actually weird, that there's hardly any E5-1600 v3 reviews out there. Especially ones that compare v2 with v3. Suppose Intel didn't send any for reviewers... I wonder why?
Last edited:
Ok, after a lot of research, it seems that during the process, Haswell E has managed to improve the not so good situation of Core i7-4770 and the worst of the series 4790. In the end Haswell E gives sometimes better sometimes worse total system power consumption figures than Ivy Bridge E did. So as deconstruct60 mentioned, it shouldn't be that big issue.
Last edited:
macOS DB7 and PB6 finally out.
[doublepost=1471951643][/doublepost]PCIe 4 specification coming out sometime next year, currently at v0.7. With at least 300W of power.
But I guess CPUs with PCIe 4 are still too far out.
Double the data rate would help with all the SSDs and TB3 controllers.
[doublepost=1471951643][/doublepost]PCIe 4 specification coming out sometime next year, currently at v0.7. With at least 300W of power.
But I guess CPUs with PCIe 4 are still too far out.
Double the data rate would help with all the SSDs and TB3 controllers.
Even if you have a lot of cores they seem better spread out over two processors. Looking at processors for the 6,1 for example, the 6-core has a much higher clock than the 12-core, so two 6-cores would be better than a single 12-core despite them having an equal number of cores. Not to mention twice as many memory slots and more PCIe lanes.
Last edited:
Yup. On the other hand it would mean twice the power consumption over single CPU.
Just idly wondering if currently, a dual 6 core @ xGHz is indeed equivalent to a single 12 core @ xGHz (barring the fact that dual 6s are usually clocked much higher). Back when SMP started gaining momentum in the enthusiast community (RIP 2CPU.com), the common wisdom was that it was better to have the cores on a single CPU, due to overhead between 2 CPUs talking to each other. Of course Northbridges, QPI, etc. have come a long way since then and I'm not up to speed.
With two CPUs, although you get twice the aggregate memory bandwidth, you introduce additional memory latency. Also, on a single CPU the cores will share the top level cache, and thus with two CPUs maintaining coherency between those caches has an additional cost. Whether the dual socket system is faster or slower than an equivalent (i.e. same number of speed of cores) single socket system will depend on memory access patterns of the application.
There may be some issues with dual cpus, but many people would appreciate the extra PCIe lanes available... and the huge amount of RAM also.
And of course as we step back into reality, and not the "assuming equal cores at equal speeds", its MUCH cheaper to get the same cores at roughly the same speeds with duel CPU set ups. 2x 2630V4 would cost you about $1300 and get you 20 cores with max turbo up to 3.1 GHz, while 1x 2698V4 also gets you 20 cores, up to 3.5 GHz, but costs $3200.
I disagree. Intel doesn't manufacture Xeon processors for laptops and SFF systems. They're designed for workstation class systems. If power consumption / heat dissipation is the reason why Apple has not updated the 6,1 Mac Pro then that's all on Apple. Pre 6,1 Mac Pros can be upgraded with processors which consume more power / dissipate more heat because they were built to accommodate increases in power consumption / heat output.
[doublepost=1471980311][/doublepost]
In the example provided by Wallysb01 the two processors (170 watts) consume 35 watts more power than the single processor (135 watts)
Last edited:
Here you go: http://ark.intel.com/products/family/88210/Intel-Xeon-Processor-E3-v5-Family#@Mobile
And, for example: http://www.dell.com/us/business/p/p...1015us_4&model_id=precision-m3510-workstation
For a SFF system using higher core counts available in Xeon E5v4, pick one of these: http://ark.intel.com/products/family/91287/Intel-Xeon-Processor-E5-v4-Family#@Embedded
and use a mini-ITX motherboard such as: http://asrockrack.com/general/productdetail.asp?Model=EPC612D4I#Specifications
Last edited:
Does those CPUs have higher core clocks than single 20 core? No.
That was exact point of the sentence someone made. That dual 6 Core Mac Pro would have higher single threaded performance than single 12 core one. Yes, it would, with double the power consumption.
In that example, dual 10 core, 85W CPU are not boosting as high as single 20 core, 140W CPU. So whole point is completely destroyed, and only 3 things for dual CPU setup here are price, amount of RAM available, and amount of PCIe lanes.
I always thought the number of PCIe lanes routed to the PCIe slots is defined by the southbridge chip. The southbridge chip is connected to one or more CPUs via QPI interconnects. At least on the cMP the PCIe bandwidth was always 36 lanes because of the X58 chipset did not allow more. I'm not aware that this differs for the single socket or dual socket systems.
Edit: For the current generation of Xeons, this is no longer true, as the PCIe part has moved into the chip. However I did not find any documents which explained whether one CPU could access the PCIe devices connected to the other CPU. I assume some kind of arbitration mechanism must be there to avoid conflicts when both CPUs want to access the same PCIe device.
Last edited:
Assuming, for the sake of argument, these processors are intended for laptops and SFF systems doesn't their existence negate the point Zarniwoop was attempting to make.
[doublepost=1471982945][/doublepost]
It doesn't destroy my post. What I posted is factually, objectively, and verifiable as correct.
[doublepost=1471983029][/doublepost]
Newer processors contain PCIe lanes directly off the CPU. A system board manufacturer can add more through the chipset if they so choose.
The problem is you answer to post without understanding the context. What is destroyed is the context.
Dual 6 core, 140W CPU with 3.6 GHz Base clock, 3.8 GHz would be faster than single 2.8 GHz 12 core CPU, with 140W TDP. Yes that is true. The problem is that it would consume twice the amount of power.
dual 10 core, 85W CPU, would consume only 35W more than single 20 core CPU with 135W, yes that is true. But it also would be slower, than that single CPU.
That point is completely destroyed.
I did no such thing. I merely stated the power consumption of the two CPUs he used. I made no other statements. Unless those power numbers are in error my post remains 100% intact.
Read the sentence AFTER the bolded in the quote. Context is destroyed. Yes, your post is valid. But context of the posts is destroyed.
You understand right now?
