Separate names with a comma.
Discussion in 'Mac Pro' started by MacVidCards, May 5, 2014.
This is reassuring, as I've just got my hands on a 4.1 2x2.26ghz to go down the upgrade path.
Two observations - as I asked in another post, do you have any idea if there are issues with using the higher power drawing (130w vs 95w) 3.33/3.46 cpus?
Secondly - I see you and Barefeats went with dual Radeon GPUs. As a future/potential customer (we've traded emails about the 780), are these dual GPUs a preferable option? Were they chosen to max out the GPU performance?
Really? Clockcleaning? I realize your business is centered around selling video cards, but out of the 8 tests the cMP beat the nMP in 5, they had a dead heat in 1 (29.9 vs 30 is a stasticial dead heat) and the nMP won in 2.
The "worst" defeat (percentage) was Resolve 1X Noise reduction where the cMP beat the nMP by a whopping 12.7%.
We have all known that the Ivy Bridge processor isn't that much faster than Westmere especially since the cMP can push faster clocks. I would actually assume that the cMP would win in this exact scenario since the Radeons aren't under clocked like the D700's are. In fact if anything, this actually shows how well the tiny nMP does compared to a full sized tower. Many believed it wouldn't be able to hold its own against a full sized tower because it wouldn't be able to dissipate heat well enough.
Guess that isn't the case eh?
Edit: The more you exaggerate (i.e. clock cleaning in this case) the more respect I lose for you. Acting like a 13 year old is not a way to run a business.
Has anyone actually bothered reading the article (OP included) ? I mean except the small differences shown in those cute colorful bars.
The article concludes that the cMP of the comparison is pricier, way larger, noisier it has some installation caveats/tricky parts in order to achieve this performance and - of course - still missing latest technologies like usb 3.0 and Thunderbolt 2.0.
It is faster, indeed. But that's hardly an "ouch" case.
True enough but the two it did win were in FCPx which is pretty irrelevant in a professional machine.
And the worst the nMP did was 12% less than the cMP. Neither exactly is taking the world by storm really. As already stated, the cMP lacks PCI3.0, USB 3.0 and Thunderbolt.
When you take the percentage difference on all 8 tests, sum them up and divide by 8, the cMP in these tests only wins by an average of 1.9%.
Clock cleaning /= 1.9%
Edit: Oh yeah and let's also not forget, that 1.9% will actually cost you MORE than buying a fully decked out nMP (per Barefeats) and at this point would have no warranty.
The lack of performance progress is pretty amazing. Release a new machine that is slower or similar in performance to the 4-5 year old predecessor? Brilliant!
Whoa, maybe you missed this ENTIRE article, GPU's are now playing a huge part in processing of video/photos/etc. In fact 6 of the 8 tests were both GPU and CPU tests. Guess what? Both are running modern (current) video cards. Therefore the cMP is NOT the same cMP released 3-4 years ago, but rather a hybrid of Modern GPU with an aging CPU. What you also are missing, is that x86 processors (especially workstation since they are now a generation behind) aren't gaining much processing power (clock for clock) anymore. All X86 processors have virtually plateaued and instead more is being concentrated on GPU's.
i would imagine a fully upgraded dual cpu machine would certainly best an off the shelf single socket machine..
for all intents&purposes, those things are running equal (as in 18 seconds vs 20 seconds is the same thing when you're sitting there)
if i didn't see the results, i would of guessed the oMP in that configuration and in those tests would have truly clock cleaned the nmp.
for all sorts of reasons, if someone looks at this comparison and decides to get an old mac instead of a new one, they're making a mistake.. thing is, those that agree with me on this already agree with me and those that don't agree, never will so there's not much use in discussing it.
HEAT - The Zombie of the GHz Wars. Current Solution - Disaggregation of functions.
Paul, you get my thumb's up here. I have a slew of Dell Workstation 450s from the early to mid 2000's that have 3 GHz Xeons. Those systems are just not mega cored and don't have the latest subsidiary enhancements to help speed things along. For what is approaching a decade, CPU speeds are just creeping along, if at all. GPUs have mega farms of parallel processing cores (and I've got plenty of them - fast approaching 100,000) vs. the currently low CPU core allotment. I built EVGA-SR2 systems in 2010 and 2011 with dual X5680s (3.33 GHz) that I underclocked, i.e., I configured them to run at 2.4 GHz idle but turbo boost up to low 5 GHz in the case of 1 core/CPU, with incrementally decreasing numbers of cores involved along the way ( http://browser.primatelabs.com/geekbench2/500630 ). They have 13 total turbo stages/bins. They still run like champs. Were it not for heat, I could get those CPUs to run all of their cores at 5+ GHz at idle.
All I'm saying is that - as an owner of several loaded 2010 Mac Pros since its release - is that it'd be nice if there was an option to upgrade to something faster while remaining on the Mac platform instead of migrating to Windows.
Nowadays there really isn't anything noteworthy that is Mac-only, and cross-platform file compatibility is quite good for most apps, so it's not the end of the world. Just a bit disappointing.
Is Resolve updated to take advantage of the dual GPU in the nMP? I guess not. So how is the analysis fair? In an real editing room, fast storage is essential which the cMP doesn't provide.
We preach what we sell I guess. OP has nothing good to say about the nMP even before it was released because there isn't any third-party options for the nMP GPU which he could make a living outta.
Biased review. Trashed.
I will agree that if Apple had chosen to go to a dual proc workstation they could have pushed the core count to 24, but the future in workstations is not CPU, but GPU. CUDA and OpenCL (especially since it is system agnostic), is really the future of computing. Any simple mathematical repetitive process is way better completed via a GPU (with hundreds of "cores") than a general purpose CPU. Look at all the Coin Miners. They use cheap Sempron processors and spend all their money on the GPU's. Nothing like putting a $50 CPU in a case with $3K+ GPU's eh?
If the comparison of this article had put a 12-Core 2010 Mac Pro still running dual Radeon 5770 (or a single 5870), then nMP would have crushed the Mac Pro. All this story did was prove that CPU's aren't a whole lot faster in 2013 than they were in 2010.
What's best for you depends on what you do, but I might not do what you do.
Me too, for some things, if the fully upgraded dual cpu machine was a cMP with X5680 or X5690 Westmeres and the off the shelf single socket machine was the 12-core nMP.
True, but if you're not setting there because you are running a 50 or 100 or 500, etc. hour 3d rendering job, then those seconds quickly began to add up.
Me too, for some things, but it's not a clock cleaning for low duration activity - different degrees of faster performance for some, and a dead heat or slower to varying degrees for others.
I agree totally if they haven't fully assessed their true needs.
But with this I respectfully completely disagree. If people who you posit might be in either camp let down their guards/prejudices just long enough to think about what they spend or would spend much of their system's time doing, they might bolt from their current camp and free themselves from it's a cMP vs. a nMP world view. Discussing the matter might lead some to freedom from their own mental shackles and provide a basis for them making the best informed purchasing decision based on their true needs.
Can you point out where this is proven?
If 1 CPU from 2013 had similar performance to 2 CPUs from 2010, then CPU performance has literally doubled.
Did YOU actually bother to read the article? You can get the cMP decked out as seen there for $7200 (>$2 grand less than nMP) if you don't want to pay for the $2500 PCIe expander and want to go with an external PSU instead, which there are multiple guides on this forum on how to do it ( I wrote one). Plus you have room for internal hard drives, Optical drives, and highly economical /reliable PCIe expansion, where you can add *yet another* video card if you were so inclined (unclear if this will affect performance in these apps, but if it does, woohoo!).
Also realize: This is just the beginning. When they get the R9 290X working with OS X, it's going to tear it up-- about 20% faster than the 280x (d700) at opencl.
nMP are going to be stuck with the D700 for a long time (my guess: forever, or until they swap out their whole computer), cMP users are just waiting for drivers to unlock tons of new amazing hardware that blow the D700 away. That's not to mention future video cards, beyond the 290x.
The 7970 (basically the same chip as D700) was released 2.5 years ago. How much longevity did anyone think these things were going to have?
Agreed. It did however show the usefulness of having dual sockets
Users should purchase the machine that fills their needs best. Depending on the tasks at hand, that may be the oMP. What sets the oMP apart from the nMP are the cost to upgrade and the upgradability itself.
Careful review of the relative costs of the new and old MP's is interesting. Note that Rob has included an external PCIe cage/power supply that costs $2500. Although I smiled at his characterization of the skills required to install a auxiliary internal power supply in the oMP's unused optical bay as "McGiver" -like, it's not really that hard.
And those who sneer at the fact the oMP in the test uses current gen GPUs (two of them) miss the point: the oMP's video card CAN be upgraded, whereas (at least at present) the nMP's cannot.
But many Mac Pro users do not require heavy-duty video processing. If the CPU dictates one's workflow, other factors come into play.
Need fast storage? With the oMP, simply add SSDs, or add an inexpensive internal SATA3 PCIe card and stock SSDs, or an mSATA PCIe adaptor (like the cool new Additronics 4x cards), or use PCIe storage for max speed. The nMP requires expensive external Thunderbolt-based expansion.
Need lots of storage? Put your SSD boot drive in the optical bay and add HDs to the oMP's four internal drive trays. You'll need to use external HDs with the nMP.
Need USB3? A card is under $30 for the oMP. And no adaptor is required for Firewire or USB2.
The oMP is big and loud? Mine sits in a machine room, so I wouldn't know.
The real message in the Barefeats article, as I see it, is one can choose the platform that best fills one's needs. Depending on those needs, the best choice may be the oMP.
It's core to core count. 12 cores vs 12 cores right? Whether the system has 1 process or 2 doesn't matter right? Case in point, back in early 2000s, if you had a Dual Core system (two cores on one proc) or dual single procs it's the same right? So how is that different today if we run 12 cores spread across two processors vs 12 cores on 1? A processor is technically a Core right? You have a 12 core processor technically means you have 12 CPU's right? They might share a few components (cache, memory controller, etc. etc.) but overall the OS treats each Core as its own processor right? It assigns a task to a core. Core to core there has been very little increase in processing power. We have simply reduced the heat/power requirements of each Core so we can jam more cores into one package.
You're missing the point.
I really don't care if they put a million cores on one socket or a million sockets with 1 core.
The bottom line is that in multi-threaded CPU oriented applications, top end Mac Pro system performance has not increased in 4+ years, while it could have doubled, simply by sticking with a 2-socket design.
CPU performance hasn't even come close to nearly doubling between the 2010 and 2013 MacPros. For self builds it has and for Windows and Linux system users it has. So I do get your point. But it was done by ratcheting down core speeds and increasing core count - disaggregation, plus a smaller amount (than I apply) of underclocking with increases in turbo bin stages.
Currently the CPU in the top of the line MacPro6,1 is a 12 core E5-2697 that runs at a base speed of 2700 MHz/ 2.7 GHz and now sells for about $2.5K+. There were not (and are not now) any 12-core Xeons for any previously released MacPros; however, there were (and are) 2009, 2010 and 2012 MacPros with dual 6-core Xeons, the highest performing/price of which is the X5690 (sold for around $1.7k originally and sells for about $1,5K now).
As indicated above, faster for what should have importance more in the context of the particular application(s) that one spend(s) the most time running or otherwise has the greatest importance to the user. My guess is that for each of us our needs vary. But if I were to put it in the context of running a benchmark, here are some things to think about:
1) http://browser.primatelabs.com/geekbench2/500630 - the settings of the processor matter;
2) http://browser.primatelabs.com/geekbench2/search?utf8=✓&q=MacPro5,1+x5690 - not every processor of even the same model no. performs the same;
3) http://browser.primatelabs.com/geekbench3/search?dir=desc&q=macPro6,1+E5-2697&sort=multicore_score - core count, core speed, QPI speed and instruction set have more to do with whether processors are fully and truly comparable and
4) [ http://browser.primatelabs.com/user/9077 ] - a machine that has a Geekbench 2 performance no. of X will have a Geekbench 3 performance of, at least, X x 1.15 to 1.2.
BTW - I learned from a little experiment that I did over the weekend that the current OS is capable of running up to 32 cores, whether it's 32 real cores or 16 real real cores and 16 hyper-thread cores (but much slower cores): compare - http://browser.primatelabs.com/geekbench3/557286 with 16 real cores + 16 hyper threaded cores vs. http://browser.primatelabs.com/geekbench3/558590 with 32 real cores and no hyper-threading. Also see posts #'s 1022 to 1030 here: http://forums.macrumors.com/showthread.php?t=1333421&page=41 .
Here's an example:
The 12-core nMP's there list between 827 and 850.
The 12 core 2010 2.93 Ghz Mac Pro is 785. So a bit slower if you disregard the 3.06 and other faster available ones.
The dual socket 24 core machines running 1600-1650ish as one would expect.
So you lose roughly half the performance by going to the single socket design vs simply updating the previous design to the new chipset.
That's my Takaway!
nMP = TB
oMP = Expandability
Make your Choice and Pay your Money.
of course, you're right..
i was more taking a jab at some personalities but when it comes down to it-- being just as hard headed as the 'faults' i was speaking against.
or something like that
So I suppose AMD sends a free video card to all cMP users whenever they release a new one, or hold on, two.
If you are willing to spend more money, obviously you can get a faster computer, just like I'm going to sell my nMP whenever there's a new nMP and buy the newer one, therefore getting a faster computer all aboard by spending around the same money someone would spend on two latest generation GPU's.
Or hell, maybe Apple will even provide an upgrade path for the GPU's in the nMP, who knows. It's way too early for speculation.
One thing we all know is that cMP's will never be supported, they may even lose the ability to run the latest OS from Apple and since Apple is not supporting them anymore, their resale value is going to plummet.
Everyone should make the switch to PC's today if the nMP isn't going to cut it for them. Buying an oMP is plain stupid.
have you seen anybody running maxwell with 2 nMP 12cores yet?
i don't think it would be entirely practical or cost effective to do that but i'm just curious if anyone has done it yet..
it's maybe $4000 more to get 2x 12core nmp than it would be if apple redesigned the mac pro as a dual socket machine with 24cores.
in a way, the stance could be "if you're relying on multi cpu as your main means of performance enhancement then we're still offering that.. you just gotta pay even more for it these days"
it appears as if apple is breaking away from ghz race (and more importantly, the core count race) and challenging developers to do the same.. get rid of that 1999 code which relies on hardware improvements to double the speed of their not-so-efficient algorithms every 1.5yrs.. it's a big bet and it's going to be a couple of years (as in- at least 2yrs) for us to see how well they've played their cards.. in the meantime, i'm optimistic about their decision.
[EDIT] and a bonus- a real bonus for users/consumers if this pans out as apple is expecting and the developers start writing smarter, more efficient code which doesn't focus solely on cpu is that the software is going to run a lot better on laptops etc where it's impractical to stuff more than a handful of cores in there.