In general, I believe he means having more choice is better. Sure you were limited with processors before with the classic Mac Pro, but now you're even more limited with the design of the new Mac Pro.
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.
The Mac Pro is basically the Apple Cube
- Thread starter chupachup
- Start date
- Sort by reaction score
In general, I believe he means having more choice is better. Sure you were limited with processors before with the classic Mac Pro, but now you're even more limited with the design of the new Mac Pro.
Yep, and the PR/propaganda folks are out in force in this thread.
Fastest bestest ever, golly gee !!!
Followed up with a rational sounding diatribe extolling the virtues of fewer choices . Toss in some pedantic nit picking pointing out that 12 cores have only been an option since xxxx.
Have had 1 vs 2 CPU options since G4 days. Just stopped.
Of course having more choice is always better, all things being equal. But all things are never equal are they?
The reason we kept getting more cores is not because we needed them but because Intel was unable to produce faster running cores to begin with. It's because of technological limits or Intels inability, all new chips featured more cores instead of faster speeds. That was bound to become unnecessary at some point. Mathematically each core you add gives you less of a performance boost. So unless you were running 100% parallel processes all day long, you were bound to hit diminishing returns sooner than later. How many people here would jump through hoops to have a 96 core Mac Pro at some point?
This will only be solved when we finally get graphene processors running at 200-300 Ghz since with silicon they seem to have hit the 4Ghz ceiling and can't get passed it. Until then the biggest advances in CPU's will be about mobility and low power use. Not speed. Theoretically I can take 7 year old Xeons and build an amazing workstation around it just by using the newest tech everywhere else and nobody would notice a thing.
Unless one uses virtualization....
Unless one runs services with SSH connections....
Unless one runs vector-parallel apps....
HP's "socket tax" is much less
$3766 - HP Z420 - E5-2650v2 2.6 GHz 8 core
$3026 - HP Z620 - two E5-2609v2 2.5 GHz 4 core
Systems have same GPU, disks, amount of RAM.
These are prices at the HP online store. Resellers often discount quite a bit.
$3766 - HP Z420 - E5-2650v2 2.6 GHz 8 core
$3026 - HP Z620 - two E5-2609v2 2.5 GHz 4 core
Systems have same GPU, disks, amount of RAM.
These are prices at the HP online store. Resellers often discount quite a bit.
If you are talking about VT-d, yes you got me there. Not a performance related difference though.
And which apps are "vector-parallel" and how do they benefit from Haswell?
Yes, E5-2609v2 does not have Turbo Boost while the single 8 core boosts up to 3.4Ghz.
I'd rather pay the difference and get the 900Mhz faster processor, it's quite a gap.
Dont forget that you get an overhead by using 2 processors and that each needs to have its own memory pool compared to one processor. 2x4core does not run twice as fast as a single 4core. It can go up 20 to 30% speed loss. (According to my own expirience)
Multiple 8 cores yes, but at no point dual 4 cores and a single 8 core, or dual 6 cores and a single 12 core.
I don't think the performance gap is that big. Back when dual core was first introduced there was a slight performance advantage of it compared to dual single-core processors, but it wasn't as big as 20-30%, more like 10% at most.
I have no idea what the situation is today.
Also VT-x(EPT).
And the performance difference running virtual machines is huge - that's the whole point of hardware-assisted virtualization!
Any app that's using vectors (long lists of numbers). AVX in Haswell has 256-bit registers that can work on 8 32-bit numbers at once, or 4 64-bit numbers. That means that math-intensive calculations can be 4 or 8 times faster - because you can do 4 or 8 calculations in parallel.
In Ivy Bridge, AVX is mainly for floating point data. Haswell has AVX2, which basically adds support for common integer operations as well.
It seems odd to be concerned with both Turbo and high core count - since if you're keeping all of the cores busy you won't get much turbo boost.
You need to have memory on each socket, but it's all one pool of memory.
NUMA in a dual-socket IvyBridge/Haswell system is basically a non-issue. As was discussed in the several threads on one vs. two vs. three vs. four memory channels on the MP6,1 - the huge caches on these systems mask the large theoretical memory bandwidth difference for almost all applications.
The same thing applies to NUMA issues. The relatively small difference between "near" and "far" memory is irrelevant to most useful apps. Are there some synthetic "bandwidth virus" benchmarks that show a hit of a few percent - sure, but don't worry about it.
This example was trying to do low end to highlight the cost uplift inherent to the second socket.
I'll now give an example of some non-lowend systems. (And again, these are the HP online prices, not the actual price at resellers.)
First, let's do a single 12 core in the two systems:
- $6166 - Z420, E5-2697v2 12 core (2.7 GHz/3.5 GHz)
(not actually offered by HP, but they offer the 8core E5-2650v2 in both, and the 12core is +$2400 over the 8 core on the Z620) - $6476 - Z620, single E5-2697v2 12 core (2.7 GHz/3.5 GHz)
So, HP has a $310 "socket tax" to get the dual-socket system. (The 12-core MP6,1 is $6999 at Apple.com - a bit higher but not outrageously so.)
So, what happens if we build the Z620 with dual six-core CPUs and compare it to the twelve-core Z420?
- $6166 - Z420, E5-2697v2 12 core (2.7 GHz/3.5 GHz)
- $4096 - Z620, two E5-2630v2 6 core (2.6 GHz/3.1 GHz)
- $7496 - Z620, two E5-2643v2 6 core (3.5 Ghz/3.8 GHz)
So, save $2000 and get comparable multi-thread performance - or pay $1300 more and get a system with much better single and multi-thread performance?
You are looking 100% in the wrong direction indeed. A computer is software that runs on hardware. The hardware part that you guys keep on going on about isn't the problem here, it's the software part that is. The software has to support the hardware properly in order to make it run properly. That means manufacturers have to support OS X. And there's your problem.
To quote you:
That would make a success rate of 33% but it is very inaccurate due to the very small batch. Or in other words: it doesn't say anything about it working properly. And even if it did it would mean that it won't work in 66% of the time aka the majority of the time. So don't talk nonsense about things being upgradable if you give numbers that show it doesn't work the majority of the time. Never assume it will if it won't work in the majority of time. If you do it will bite you because you are promising someone something that you can't keep.
Yes and that's the problem. If you have to do research it'll become technical. Having to do research is a very high wall, hitting technical stuff is another high wall. What do people do when they hit a high wall? They won't bother. Too much effort.
And some people need to know what users are exactly. The problem is knowing that you can upgrade. Most users have no clue about that and those who do generally also have the knowledge and will do the research and might even do the upgrades themselves. Most users however will only come to you and ask why their machine is so slow. In reality most users are not technical at all, they don't know nor understand computers. Therefore we cannot expect that any manufacturer that is aiming at the general public will bring in those features that we techies want. Take a good look at Android vs iOS.
There are lots of explanations, the problem is that some of you do not wish to see them and are trying their utmost to battle them. If your business depends on it I can imagine that you have even more reason to do so.
Same here but we're talking about a reseller who is in the Apple business since the 90s and who also do Apple sysadmin stuff. They know their stuff and it shows in their product portfolio (meaning that whatever you buy there is guaranteed to work; they did the proper research). Unfortunately (or maybe not) they are not that known, only within a very small circle of people (it's probably why they still exist: very small but loyal customers).
Yep, that's how most manufacturers differentiate their products. They aim at certain users and use cases. Upgradability has never been part of that at Apple. The fact that they use certain components is what makes it possible but there is no official support (which is why you need to do research).
I don't mind people sharing their point of view, you can learn from it. In this case people need to learn that things are not always the way they think they are. Companies handle new computers differently as do ordinary users that make up about 90% of the entire userbase. They are not into upgrades and components like we are. When they upgrade they upgrade to a new machine.
When I was an intern we wanted to upgrade one of the HP workstations from single cpu to dual cpu. The machine was only 6 months old. We contacted HP and they told us the machine wasn't supported any more. There were no cpu upgrades. Now, we could do some research, buy a cpu and risk it not working but why would we want to upgrade a machine that the manufacturer no longer supports? There are more risks so it is better to replace it with a new one.
Something similar applies to machines that are 3 years old: yes you can upgrade a component but do keep in mind that the other components are also 3 years old (at least). They were top notch 3 years ago but they are outdated now. If you want a fast machine you eventually end up upgrading so much components that a new machine is a better idea. A lot of decisions are made on time, effort and money. Just to give you an idea how sysadmins deal with these and other things.
A hobbyist won't think that way. He'll go "gees, that new Mac Pro costs me another 3k, that video card only 300; am definitely buying the video card!". It's quite different when you have to fork out all that money yourself instead of via some kind of financial trick.
That's why Apple gets away with doing things as they do. Most of their userbase simply doesn't care. I just like that they use Thunderbolt. It gives you some of the pro things we techies like. It also gives me components that have proper OS X support which means I don't have to spend huge amounts of time doing research, getting things to work and/or troubleshooting problems because things are guaranteed to work. For my users that means the support costs are lower. If they work in a group the group can buy such a component and share it amongst their members. That saves money as well. We have a lot of issues with the PCI components and none with the USB versions. USB is something that you can pass on to a vm easily with any virtualisation software. PCI passthrough however... If possible we recommend people to get the USB version. That way they are not as dependant on the hardware. I find this to be the only downside to Thunderbolt: you can't pass it on to a vm.
It was and still is a cute little machine but it wasn't always a good choice indeed
Did you have a hard time upgrading your Mac Pro?
I sure didn't. Sure I used to build computers as a hobby about 10 years ago before I took over my marking firm. But my business is full of Mac Pros. None of them are hard to upgrade. They are easier than a comparable windows machine to upgrade. No you have less GPU options to choose from, but that is good and bad. OSX supports less GPU options but when they work they work well. The same cannot be said for a comparable windows machine. Apple makes sure what is supported has no problems and is fully supported.
Tight integration between software and hardware is what it's about. Windows users do not have that luxury. This is what sets Apple apart. It just works. You seem not able to grasp that fact. Or do and just are ignoring it.
The 2010 Mac pro is still a good investment. You say it's better to just buy a new machine. You can upgrade a 2009 to 2012 8 core Mac Pro to a 12 core monster, all SSD, USB 3, new GPU etc for very little cost vs new. And it will outperform all but the highest configuration nMP. My IT dept upgraded two 2010 models for 800 each. Now they perform on par with the highest config nMP from my office. I bought four new models of Mac Pro. It was cheaper to upgrade our existing machines. All machines except are older than three years.
USB3, SSD, CPU were all plug and play. How much easier do you want? No research required. Most people who have workstations know their way around a computer or have someone that does.
My IT team had no trouble finding upgrades or installing them. fully software supported.
What windows machine from 2009 or 2010 can you still upgrade to a 12 core machine with just plug and play? No wires etc.
There is more than a dozen CPU's that fit in a 2009 to 2012 mac pro. What windows machine is comparable to that from those years?
The Aluminum Mac Pro is a work of art and precision in my opinion. We also have the new Mac Pros and they are impressive as well. Not as much as the old aluminum ones. Nor as upgradeable. Less options.
I fail to see your point here. The old Mac Pro is very upgradeable. No problems, just plug and play in my opinion.
What problems did you have? Or do you have with yours?
Last edited:
Every. Single. One. Of. Them. (Assuming that you're comparing multi-socket workstations, that is.... None of the $299 Best Buy specials can go 12 core.)
Except that there would be many more than "a dozen" CPUs to upgrade.
I'm not sure they started shipping the G4 as a dual CPU box for the best of reasons... The reason they did it was because when Intel broke 1 ghz they were having trouble getting over the 500 mhz mark. So it was more of a "Look, we have 1 ghz total too!" move.
Almost all software couldn't use dual processors. The operating system back then wasn't really built for dual processors. And the memory bus really couldn't feed dual processors.
It's a little bit misleading to claim Apple has been a serious dual processor company since the G4, when the G4 was anything but a serious dual processor machine. It was a token machine to prove Apple could hold it's ground against he P4. At no point was it ever a serious competitor to the multi-processor Intel boxes, it could barely hold it's own against a single Intel processor. And the G4 didn't start as a dual processor line, they only made that changeover when things started sucking.
The G5 was really the first multi-processor box that I'd consider a serious effort, and not just a "we're trying to make the box look better because the G4 is sucking" feature. But for the same reason I also don't buy that Apple has been a serious multi-processor machine company. Excluding the pathetic G4 boxes and the rare Power Mac 9600 DP, they were only in the multi-processor business for a fraction of the time they've been in the workstation market. Apple, especially with machines like the G3 has always traditionally been in the low end pro/high end prosumer market, and the nMP seems to be a realignment to their more traditional strategy.
Apple only really started to care about dual processor boxes when they were trying to get into scientific computing/clusters, and that died along with the Xserve.
So very wrong. From the introduction of the dual quad core (8 core) Mac Pro in 2007 to all Mac Pro systems from 2008 to 2010 there were dual quad core systems. They introduced dual 6 core (12 core) server models in 2010 too and then dual 6 core (12 core) models in their non-server line up in 2012.
Sounds like someone is pissed that Apple threatens his lucrative (side) business with GPU upgrades...
I bet this bird sings another song once someone figures out how to upgrade the nMP graphic cards.
Sounds like someone is pissed that Apple threatens his lucrative (side) business with GPU upgrades...
I highly doubt that we'll ever see a GPU upgrade for nMP developed by a third party or Apple.
Sounds like someone is pissed that Apple threatens his lucrative (side) business with GPU upgrades...
When you can't dispute the facts, dispute the messenger.
Propaganda 101
Nice work.
I tried to discuss politely and fact-based, but you preferred to defame that as "diatribe" and vilified everyone not being your opinion as "PR/propaganda folks" (a behaviour which is not limited to this thread, besides), so I learned from the best.
----------
Don't underestimate the tinkerers! People thought similarly about GPU upgrades for some of the older Macs and the tinkerers went to great lengths to adopt standard (PC) parts, once those got cheap enough.
The SSD's in the rMBP's are also sporting proprietary connectors, yet there are already two professional vendors offering replacement products (at least for the SATA-based rMBP's so far).
Once the nMP's are getting into the private market in bigger numbers (i.e. second-hand), a demand will build up. Question is when the number will be high enough to make the venture worthwhile. I expect it to take longer than with the rMBP's - simply due to the smaller market - but it will happen eventually.
I suggest you have a look at the iFixit teardown pictures. to develop a GPU upgrade for the nMP has nothing to do with tinkering anymore. this would mean serious business with R&D costs way too high. not a single off the shelf PC part can be used. no, I don't see that this will happen ever. it's not the same as it was with the G4 Cube where you could use an AGP graphics card as long as it would fit and as long it didn't draw too much power.
Didn't know this. But I'm confused now, it seems like VT-x was supported back in 2005 even.
I asked for examples of apps. I'm a mathematician, I rarely use Maple or Mathematica so it'd be nice to see that kind of a performance increase when I get a haswell or broadwell machine in the future. And since we don't have any Haswell Xeons at the moment, this advantage belong to the future chips.
If you are keeping all cores busy at all times, yes. But unless the machine is just a number cruncher or a server with no user sitting in front of it, this won't be the case.
----------
Agreed, if your job requires you to run 100% multithreaded processes almost all day long, getting the cheaper dual cpu machine will be better.
But back to topic, Apple never offered such an option to begin with.
----------
Not to mention anyone attempting to manufacture a GPU upgrade for the new Mac Pro has to run extensive tests to make sure that the cooling inside the cylinder will be enough since the card cannot have a fan, which wasn't a problem before.
Last edited: