That's on Windows. FirePro drivers vs Catalyst drivers are a whole other story. On the mac front, we never had Catalyst drivers in the first place. Which brings me to the question that why Apple is using FirePro at all? 4K costing GPU has almost the same internals as the 400$ one, the only difference is drivers. But on Mac we have the same exact drivers for all GPU's anyway, Apple's.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Mac Pro 6,1 Cylinder - Fact Gathering.
- Thread starter Tesselator
- Start date
- Sort by reaction score
That's on Windows. FirePro drivers vs Catalyst drivers are a whole other story. On the mac front, we never had Catalyst drivers in the first place. Which brings me to the question that why Apple is using FirePro at all? 4K costing GPU has almost the same internals as the 400$ one, the only difference is drivers. But on Mac we have the same exact drivers for all GPU's anyway, Apple's.
If it will be pcie connectors similar to what used in notebooks (MXM), then some adapters from MXM to PCIE X16 may arise. Making it possible to have docked videocard with pcie x16 connection.
Clue: it is a fairly safe bet that the GPUs in the Mac Pro are not going to cost $4000. The retail/OEM cost of a complete PCIe graphics card tells you naff all about the wholesale cost of the chipset to a large customer like Apple.
A successful Pro at an affordable price would promote OpenCL over NVIDIA's CUDA & push GPU-based computing into the prosumer arena, so it is in AMD's interests to give Apple a good deal. If - as you say - the cost of FirePro cards is more down to software than hardware then the deal might be more about licensing Apple to use the FirePro brand than offering them a bargain deal on the hardware.
When the Pros are available, and people start benchmarking them, we'll find out whether the "Workstation-class GPU" and "Firepro" tags are justified.
Honestly - all bets are off until we find out the price points and specs. A $1500 entry MacCan Pro would be a hugely different proposition to a $3500 entry MacCan Pro, and I don't think there's enough evidence yet to rule out either possibility.
I believe Apple said "Up to 7 displays"--the base model may not be able to do that. [Edit: just checked the product page, definitely says UP TO] That would mean the W5000 is a perfectly viable option for the base model. Moreover, a W7000 FirePro has an MSRP of $700 each--putting a dual card setup in the $1,400 range. Even with bulk purchasing, this is going to either eat up the nMP's profit margins, or, more likely, make it way more expensive. This seems unlikely.
But say you're right, the W7000, according to benchmarks I just looked at, is considered midrange... from 2012.
That's exactly what he meant by "they wont age well." Upgradability is a component of aging (in that respect the nMP is dead in the water), but, like you imply, the starting config is important to.
If the best card in the nMP is indeed equivalent to the 7970 (a W9000), that's a high-end computer being released with a card that's 2 years old. Nvidia solutions are already beating the pants off this. The only saving grace is that there are 2 of them, but I guess we'll see how the performance scales...
... But you can upgrade them, increasing their value significantly. That is one of the big reasons Mac Pros hold their value so well--they can be upgraded.
That's good that Apple will be taking this ability away from people! Surely a machine that can't be upgraded to a better machine will be worth so, so much more!
Total bologna. The ability to upgrade dramatically increases the longevity and therefore the value of old machines. That's why people are calling the iTrashCan disposable.
Scenario 1) Old computer with PCIe slots that can be upgraded to use the latest video cards.
Scenario 2) Old computer with non-upgradable video cards which weren't even top of the line when the computer came out.
What has more value? This is like renting a house and not being allowed to put your own furniture in.
Uhhh that's exactly what they've done in the past. The old Mac Pro page was touting the "cutting edge" 5870 all the way through 2013.
... Except that the ability to pick your own video card means you're not stuck with FirePros, you can upgrade to an NVidia offering either out of the box or when you get sick of your video card from January 2012.
You really addressed your own issue: Apple sells approximately 20 gazillion MBP's yearly, and iMacs in a similar order of magnitude. Even if the nMP sells beyond Tim Cook's wildest dreams, do you think it will even remotely approach that number?
... Except that everyone who works primarily with CUDA will skip out on the thing.
That's the problem with the little iBucket: The only things that are going to make it a viable product depend on it selling extremely well, and at the same time, it keeps slicing out entire groups of people from its customer base. All I'm getting from the fans of the thing is pure gobbledigook about how "oh, that group of users isn't so big anyway". So far we've heard that about enthusiasts, gamers, PCIe card users, >128GB RAM users, used Mac Pro buyers, internal hard drive users, dual processor users, optical drive users, and now CUDA users.
This is a great example of a self-fulfilling prophecy: Apple and its nMP ____-___ drinkers are telling us these users don't exist, and soon they wont!
Last edited:
How do you know that you cannot upgrade the GPU?
----------
As someone who did upgrade his Mac Pro's video cards more than 10 times, the first case is better. But upgrading video cards alone doesn't give you a much faster computer unless you are into cutting edge games constantly. Mac Pro isn't aimed at gamers and for professional users the workstation graphic cards have more life in them compared to consumer GPU's. A GT8800 won't run any games decently today but a Quadro you purchased in 2008 will be "ok" for running 3D modellers atm.
And I know that whenever I actually sold my old Mac Pro and bought a new one did I get an actual speed increase all around, not when I updated the GPU. So, to me, Mac Pro's always lacked upgrades since CPU and motherboard upgrading was out of the question, which are more important than GPU. If Apple could make a cheaper Mac Pro, I'd rather sell and replace the thing every year instead of buying a new GPU every 2 years.
Yes, because there are 7 physical display ports on the system. The entry level machine is STILL going to have 7 physical display ports on the machine. In all of Apple's write up the number of ports is constant. What you can use those ports for is variable. That is what drives the "up to" qualification. If you plug in 3 TB->Ethernet dongles to 3 Mac Pro TB ports, you many not be able to hook up 7 screens.
What you are purposing is that Apple systems a system with a physical display port that is "dead" when a user plugs in a display. That sounds like Apple?
Highly doubtful that it would either past Apple design criteria or even Thunderbolt certification. Display ports that don't output.
This has been covered. Those aren't the Apple made cards. The prices are do not have to be the same. The functionality will be equivalent, but that says nothing about price. Those cards actually would work in systems from multiple vendors. Apple's card will not. There is about zero reason they should be priced identically.
All Apple is buying from AMD is the actual GPU package (and probably some licensing rights/software ). They aren't buying AMD made (or contract made ) cards.
Additionally, it won't be separate drivers just for these cards. ( so no dual support/development queues ). The warranty will be typically Apple, not what AMD has on the cards.
What is very unlikely is that AMD is making and selling these cards.
Did AMD come out with a newer version of the W7000 in 2013?
Please. EVERYTHING ages. Systems with Nubus or PCI slots run out of options too. The slots themselves are not magically better "age well" criteria.
Apple has also announced no policies one way or the other on replacement of the video daughtercards. Even if not official by the times the systems are in the "out warranty zone" there will be a boneyard upgrade options.
2 years? Now shipping 1/9/2012
http://www.engadget.com/2012/01/09/amd-radeon-hd-7970-on-sale-shipping-now-newegg
Please..... TMSC's 28nm yields were crap in 2011 which slid both Nvidia and AMD designs into 2012
And frankly both Nvidia and AMD are all shipping 2 year old graphics by your criteria. Neither one has done an architecture upgrade in 2013 (yet).
If not priced right they won't make the design bake-off. And frankly beat the pants... as long as pick the right fight....
http://www.tomshardware.com/reviews/geforce-gtx-780-performance-review,3516-27.html
The whole Nvidia vs. AMD fanboy stuff is just pure misdirection from the real core issues.
The GPU daughtercards aren't removable?
It is not a tinker box. Workstation and tinker box are not completely overlapping categories. This whole thing about "control" rather than function ( usage during 99.9% of operational service ). Different folks are going to be put different value on things. For 'revenue generating' machines that 99.9% is far more criticial than the 0.1%.
What can you upgrade dramatically in terms of CPU and RAM? The dramatic performance improvements are going to come with socket changes. The move from DDR3 to DDR4 will be a RAM change next year when Xeon E5 make the shift.
4-6 years later the large jumps will see shifts in package/card sockets also. (even PCI-e slots. Go back and look at the Mac 3,1 x4 PCI-e v1 impact on the GPGPU performance a page or so back ).
What you are talking about is incremental updates using older stuff. That can be done using old stuff. Well the Mac Pro 2013 will still be able to use "better" 2013 parts in 2018 much better than any of the other Mac models that will be sold in 2013. Could it be incrementally higher with some more standardization, perhaps. But there is little current indication as to what the drop off is going to be. What is true though is that 2013-2014 parts in 2018 aren't going to deliver "dramatic" increases in performance.
Whatever. It isn't. Not even close. It is meant to be more modular component from an over all deployed system point of view. ( plug in new CPU/GPU/RAM module into system that has storage and some I/O external ).
The Mac Pro never was about being a single box solution. If it was an all-in-one it, would be an iMac. It is now. It won't be when the new system design is released.
No. One because your furniture is old ( not newest bleeding edge stuff). Second, there is no performance aspect in your analog example. What you are talking about is looking for a new home for your old stuff. It is really a discussion about sunk costs and becoming fixated about sunk costs.
Uhhh that's exactly what they've done in the past. The old Mac Pro page was touting the "cutting edge" 5870 all the way through 2013.
No it isn't. In fact the opposite. Apple makes hyperbole claims but they static. There is no "new this month" or "configs changing constantly with bleeding edge " claims.
The Mac Pro doesn't have to equal the number. The primary issue is growth not volume. If from 2013 to 2014 the MBP's go from 5M to 5.15M (3%) , iMacs from 2M to 2.06M , and MP go from 100K to 103K (3%), then that would be just fine.
I don't think Apple even expects them too. The whole entire workstation market is less than a 1M systems per year. Even if Apple took over 100% of the workstation market the Mac Pro couldn't equal MBP sales volumes. That is not a "streach goal" (to match MBP). That is a smoking drugs goal that is completely disconnected from reality.
If that isn't a very growth market then it is a non issue long term. It isn't. CUDA exclusivity is shrinking. (the links to evolution of Adobe PPro are only one example of that. )
The fundamental flaw here is where was this selling extremely well in the old form factor? Appealing to the exact same set of cusumers that failed to drive the current form factor forward at sustainable rate is going to work better?
There is little to indicate that this one is worse than the old.
When get past the hand waving but does it really slice out folks from what has always been the targeted customer base.
Enthusiasts --- tinker base?
Because the whole Mac platform is optimized for games.
Folks who want cards only or folks who want the functionality the cards provide? there is a difference.
There is a thread somewhat in last month or so about what folks actually had installed in slots . There was not alot outside the bounds of USB 3.0 , eSATA , PCI-e based SSD, and < or equal to 2 GPUs cards
https://forums.macrumors.com/threads/1614365/
Dual processors or folks that need core count? Incrementally over time the core count is coming to the new form factor.
again.... the vast majority folks are under 64GB. The >128GB RAM market is hardly high growth or highly populated.
https://forums.macrumors.com/threads/1598539/
Frankly, the >128MB often has high overlap with one of the above categories ( highly specialty PCI-e cards, dual processor (not cores) , etc. ) which means this adds little to increasing the number of folks "lost" with the new device.
Used Mac Pro buyers are typically cost constrained. You just finished arguing how these new Mac Pros will be cheaper in their old age.... That is going to stop them from buying more used Macs?
Optical drives .... well see at end of Q4 whether ODD less iMac actually when backwards in sales (even accounting for the upgrade avaialbility hiccup). It is doubtful. And frankly there is a simple solution for those that need one (and for those with more than one Mac an single external ODD is actually likely cheaper since the drive is intermittently used. ) Not really talking about ODD usage, but higher than average ODD usage.
Not really. Even if those subgroups add up to 30% of the old Mac Pro customer base if Apple finds new blood ( out of general PC market , non- Mac Pro users with growing workload demands, and current Mac Pro users with plateaued workload demands (e.g., 6 cores and 64GB likely enough) ) then it is wash. If they find 40% of old Mac Pro customer base it will actually be growth which is one of the primary root cause problems with long term product sustainability.
The huge flaw here is totally not counting any of the new folks that may trickle in with the different/smaller form factor and denser computational increase.
Nope. Telling you that you have trouble counting. You can't just myopically just count what you want to count. Have to count what is actually out there in the real world.
There is a spin in these threads that has much more to do with multiple vendors selling solutions than actually being able to physically upgrade the GPU. It isn't the same thing. One is about market creation/sustaining (and perhaps quantity of parts availability ) while the other is about evolutionary bumps in performance.
Apple doesn't offer a CPU/RAM daughtercard upgrade service or parts and yet some folks are doing them on 2009-2010 Mac Pros now.
If Apple and customers could come to a meeting of the minds on how much folks will pay and how frequently they'll upgrade GPU daughtercards perhaps Apple could create service to support this. That will never happen as long as most of the discussion primarily conflates this market making/control/upgradability possible issue. The FUD will guarantee it never happens.
That reminds me of the "Airport Extreme Enabler" back in 2007. Certain Macs came with 802.11n hardware, but were limited in software to 801.11g until such date that the Airport Extreme Enabler software was released (for $1.99) to turn 802.11n speeds on.
Wait, now I have a question: If "intense OpenCL work" can max out 8GBps, what do you say to those who talk about nMP setups with 20+ Video cards in little daisy-chained PCIe boxes?
If OpenCL is even higher bandwidth than gaming, wouldn't that mean it would do better with PCIe-level bandwidth rather than thunderbolt2?
... How hard would it be to make a W5000 that supported 4 monitors instead of 3? ... How hard would it be to have the HDMI port shut down if all six TB display ports were active (other AMD cards do this)? ...
At a cost difference of possibly hundreds of dollars a unit, I'm sure Apple will think of something. We shall see though.
That's true. That could definitely save Apple some money (provided Apple can have the economy of scale to beat the PC price). Here's to hoping it's not as ridiculously overpriced as the PC offering.
Yep, I remember the "video personality card" upgrade for the PowerMac G3. You could buy one, it only cost $800.
You're right though, we don't know the options to upgrade yet, I shouldn't presume to think Apple wont have at least 5-6 options for video cards in their initial offering--something they have never done. How great to think that 100% of your upgrade options are tied to a single vendor responding to a tiny market.
And when is the nMP shipping exactly? Today? Or is it roughly 2 years after that date you just posted?
Very true. And to think, I never knew I wanted to run a "Binomial Option Pricing Algorithm."
I like how you found the one area of the entire article where NVidia didn't dominate (OpenCL Performance)... and by the way, with protein folding (on the exact same page), they did. The article suggests this is mainly just neglect on the part of NVidia's drivers towards OpenCL (something that could later be fixed through drivers).
Nvidia regrettably still appears to treat OpenCL as a second-class API.
Also, the article suggests that apps ported from CUDA to OpenCL (such as F@H) tend to favor NVidia. This may not bode well for Adobe on AMD in the future.
Second, the F@H benchmark proves that porting a CUDA application to OpenCL is not as trivial as it may sound.
Meanwhile, in DirectX and OpenGL performance (Maya/AutoCAD), NVidia again dominates. Not to mention, AMD can't even run CUDA applications, putting this huge list out of reach of the nMP.
In all, the article was extremely favorable to the GTX780 over the 7970.
This isn't an NVidia Vs AMD debate, the Titan/GTX780 are simply much newer cards. It's been 20 months since the 7970 was released. When the nMP comes out, it'll be more like 24 months (AKA 2 years).
Finally, let's not forget that the nMP comes out in 4 months. By that time, the Quadro K6000 will be out. What if I put 2 of those into my 5,1 Mac Pro? How is that going to compare?
Even if you want to stick with AMD: When is the next FirePro being released? 2014? How will that compare to the W9000? Having the ability to upgrade those video cards sure is a wonderful thing
You can't dramatically upgrade the CPU and RAM in the old Mac Pro, you're right. You can, however, replace the GPU's out which significantly increases the capabilities of the machine. You can also add functionality that it doesn't have: SATA, USB3, etc. The nMP may be able to do some of these things, but definitely not as cheaply.
I just learned yesterday that you can get a $50 four port SATA III PCIe card working and booting in OS X without adding extra drivers.
What do you think the chances of a manufacturer pricing such a device at $50 over thunderbolt within the next few years? What if I want USB 3.1? What if I want to take advantage of the falling prices of SSD over PCIe? TB can technically do it (well, up to 2GBps, anyway), but how much will it cost, how immature/buggy will the tech be, and how long will TB users have to wait for it to come out?
I guess we'll have to agree to disagree. Apple definitely implies or outright says they're using cutting edge hardware, even when their product hasn't been upgraded in years.
LOL? How does that work. If Apple sold 1 Mac Pro this year and then sold two nMP next year, would they suddenly say "100% growth year over year! Let's dump half our profits into this windfall!"
Profit margin * Price * Quantity sold = The Bottom Line.
Also, when picking R&D, Apple must factor in Opportunity cost. Making a matching PCIe chassis for the nMP might be a neat thing to do, and having the software in place to run a GPU through it could be a boon. However, how much does Apple stand to profit off of that investment Vs having the same employees design a new mouse for their iMac, or a new power-saving technology for their MacBook Air?
Welcome to How Not To Run A Failing Business 101
Regardless of how investors look at things (admittedly growth-focused), the rules of Economy of Scale are the same: you have to have Scale in gross terms. They don't call it "Economy of Scale Relative to the Previous Quarter"
So, again, I ask you: How is Apple going to justify a huge R&D budget for nMP accessories if the sales aren't much higher (or *gasp* are lower) than the current Mac Pro? How many early adopters are going to buy a nMP without the promise of a plethora of fairly-solid hardware? This is the Pro market, after all--they're not going to form lines outside of the Apple store to buy the latest incarnation of a product simply because it adds the letter "S" after the name.
I'm sure the Mac Pro sold a lot better in 2006 and 2008, before the performance to price ratio started dropping like a stone. Sure, it's not to the level of the consumer products, but there are a heck of a lot more consumers than professionals.
I can't say what's right or wrong for Apple, but I'm sure having a professional computer filling roles in some capacity helps create a synergy with its other products. People only want to learn one OS, buy one version of software, etc. If Apple wants to destroy the professional line, they would certainly have more information that I do on their market. However, that does not make them impervious to making mistakes.
I would think that the "targeted user base" is "anybody with money." I'm sorry that "everyone but Pixar employees" just aren't "professional" enough to fit into Apple's "target."
Yes, because things that are priced lower always well more. You realize that I was conjecturing about the low resale value of the nMP due to DEMAND, not supply, right?
Your comprehension of economics is bad, and you should feel bad.
Oh I count them, I just don't think the new form factor will attract as many users as it detracts. If you have, you know, data, showing otherwise, I'd like to see it.
I see, so you have some kind of data about who buys and uses mac pros? Let's see it.
Last edited:
CUDA Renderers that i can think of:
V-Ray (dropping OpenCL support)
Octane
RedShift
Arion
iRay
Splat
FinalRender
New preview extensions in Pixar PRMan
OpenCL that i can think of:
Lux
Indigo
Still, most big production renderers are CPU only leaving the beloved Mac Pro at a distinct disadvantage.
Oh okay, that was my understanding as well. I thought you were suggesting though that it was getting more exclusive.
Well I suppose technically it is. V-Ray is dropping it. No one is currently migrating towards it that I am aware of and most new ones are CUDA only.
I suppose a requirement for getting substantially more exclusive is that there would need to be substantial development work being done in OpenCL rendering in this space to begin with, which there isnt.
You seem fairly knowledgable about CUDA, any idea of the bandwidth requirements? Someone suggested that it could actually be higher than high-FPS gaming in certain tasks. Has anyone tried benchmarking CUDA applications through 4x vs 8x PCIe, etc? (My Google-fu is failing)
cuda & openCL need very fast/wide bandwidth (well, it doesn't need it.. it's just that it's possible for it to utilize it.
thing is, it's not bandwidth between the gpu and and the rest of the system which is so important or beneficial (or your anti-thunderbolt argument in this case).. it's the speed between the gpu and video ram.
basically, a gpu can be thought of as a specialized cpu and the vram as specialized GDDR3 with a 128-bit bus between them.. that's cuda's bandwidth limits
It depends entirely on the application and how much data needs to be shuffled back and forth from system memory to VRAM and how frequently. It is possible that a given application would not show any appreciable throttling when forced to run over T-bolt.
OpenCL Benchmarks are actually a likely candidate as they typically don't replicate real world usage scenarios in terms of bandwidth needed. Luxmark for example. Copy one tiny scene into VRAM and then run a low-bandwidth but high brute force GPU calculation.