New Mac Pro thread (merged)

[ssls6]

"poor value for consumers" and "offers limitations to professionals"

funny. It would have been easier to type "I disagree". Time will tell if you are right or I am.

 

[blackhand1001]

Pretty sure TB has more bandwidth than PCIe.

No it does not. Thunderbolt 2 maxes out 20 gigabits per second. PCI express 3.0 has 16 gigabytes per second aka 128 gigabits per second. Even PCI Express 1.0 has more bandwidth. Thunderbolt is not even close to PCI express.

 

[Umbongo]

funny. It would have been easier to type "I disagree". Time will tell if you are right or I am.

No I was genuinely inquiring as to whether you had thought about what your 10 times number meant when Apple's desktop sales have been stagnant for 3 years and the desktop market is on the decline overall.

 

[handsome pete]

funny. It would have been easier to type "I disagree". Time will tell if you are right or I am.

The problem is that you're offering no reasons other than the new style for why you think it will outsell previous iterations by a factor of 10. All of your other reasons can be said about the old mac pros.

 

[kennyman]

Here is a complete spec of Mac Pro

Just checked the pictures of Mac Pro 2013. Let`s sum up on what we know so far;

1. CPU will definitely be single processor, at least from the one that they currently have on demo in Moscone West. If you look closer, in between the memory banks, you can clearly see the back plate of the socket and cpu (only 1 socket on it)

2. The back plate of the socket is clearly visible with 8 screws, which could be holding the cpu in place, meaning probably the cpu will not be soldered as some are assuming.

3. Chipset definitely looks like a C216 with 2 x DDR3 dual channel (hence 4 banks). As with the current Mac Pro, depending on the CPU, memory will either run at 1600 or 1866 as advertised (not sure what cpu is on the demo Mac Pro, Ivy Bridge E5 V2 which has not been announced yet). Also, memory looks like DDR3L sticks, low voltage, no need for heatsink.

4. It seems that a good amount of PCI Express lanes has been dedicated to the peripherals (TB2, USB 3.0 and Storage, which is not a M.2 BTW), Apple wanted to make sure to give enough options outside of the case itself for future expandability.

5. I am also sure Apple is using the four-channel Cactus Ridge controller. So they need to loop-back connection between discrete graphics and the motherboard itself. Hence they have now included 2 x powerful GPUs so that you can hook up to 4 x 4K, which will be powered by the AMD cards, very different from the implementation of TB on Air and MBP (which remember, need the GPU processing from CPU to work).

6. Lastly it is clearly aluminum, not plastic, I am afraid this Mac Pro will definitely be around 3000 USD, let`s see though

 

[theSeb]

No, the 12 core is the bto option. I am surprised how many people don't even bother to watch the keynote.

 

[Bear]

Just checked the pictures of Mac Pro 2013. Let`s sum up on what we know so far;

1. CPU will definitely be single processor with 12 logical cores,...

3. Chipset definitely looks like a C216 with 2 x DDR3 dual channel (hence 4 banks). As with the current Mac Pro, depending on the CPU, memory will either run at 1600 or 1866 as advertised (not sure what cpu is on the demo Mac Pro, Ivy Bridge E5 V2 which has not been announced yet)
...

1. CPU will be a CTO option, less expensive variants will be available.

3. Apple stated during the keynote that the memory subsystem is Quad Channel. So best performance would be 4 matched memory modules and I believe they also stated that it would be 1866 memory.

 

[thekev]

Pretty sure TB has more bandwidth than PCIe.

Thunderbolt's bandwidth is actually taken from available PCI lanes. It's not a native part of the chipset.

 

[kennyman]

1. CPU will be a CTO option, less expensive variants will be available.

3. Apple stated during the keynote that the memory subsystem is Quad Channel. So best performance would be 4 matched memory modules and I believe they also stated that it would be 1886 memory.

Agreed on the CPU, but I was just referring to the pictures of the Mac Pro that we have saw at WWDC. It is definitely a dual channel, quad channel means 8 memory banks/modules. I think you meant 1866MHz instead of 1886MHz

Or they are simply removing 1 memory bank from the quad channel....maybe it will yield more bandwidth and less latency, need to check afterwards when they start t0 sell it

 

[Bear]

Agreed on the CPU, but I was just referring to the pictures of the Mac Pro that we have saw at WWDC. It is definitely a dual channel, quad channel means 8 memory banks/modules. I think you meant 1866MHz instead of 1886MHz

Or they are simply removing 1 memory bank from the quad channel....maybe it will yield more bandwidth and less latency, need to check afterwards when they start t0 sell it

The Apple website says the new Mac Pro has a memory bandwidth up to 60GB/sec which is double what the old Mac Pro has.

And from what I heard watching the keynote and based on the bandwidth, it is likely one bank of quad channel memory.

 

[Umbongo]

Agreed on the CPU, but I was just referring to the pictures of the Mac Pro that we have saw at WWDC. It is definitely a dual channel, quad channel means 8 memory banks/modules. I think you meant 1866MHz instead of 1886MHz

Or they are simply removing 1 memory bank from the quad channel....maybe it will yield more bandwidth and less latency, need to check afterwards when they start t0 sell it

No, quad channel is in reference to the memory controller which is per CPU. So this is 1 CPU with a quad-channel memory controller accessing 1 DIMM per channel. A dual CPU setup is still referred to as quad channel with it's 8-DIMMs over 2 CPUs.

 

[Tutor]

The Apple website says the new Mac Pro has a memory bandwidth up to 60GB/sec which is double what the old Mac Pro has.

And from what I heard watching the keynote and based on the bandwidth, it is likely one bank of quad channel memory.

Max memory bandwidth figure is based on CPU: compare max memory spec here[ http://ark.intel.com/products/52577/Intel-Xeon-Processor-X5675-12M-Cache-3_06-GHz-6_40-GTs-Intel-QPI ] for 2012 Mac Pro with [ http://ark.intel.com/products/64583...E5-2680-20M-Cache-2_70-GHz-8_00-GTs-Intel-QPI ] for 2012 Mac Pro that should have been. BTW - 60GB/s is about double 32GB/s. So memory bandwidth for Ivy must be about 10GB/s higher than Sandy, and Sandy is about 20GB/s higher than Westmeres.

 

[Tesselator]

They should have made it round like a beachball. Then it could spin every time the OS is too busy to answer a request.

Haha... that would be awesome...

Naw, it's all about number of screws, number of pieces, and number of solder pads. Shape doesn't matter much. In fact this tube shape is actually easier to work with - from what I know about FA (factory automation).

Apple pays to ship the parts to their assembly plants, then to a warehouse after completed, then to stores or airports, if the later then also to warehouses and/or stores when they land. You only have to pay for the very last leg of distribution (from it's stored location to your front door).

No, no. Again it's all about the number of pieces, number of screws, number of cables needing to be attached, and so on. Look again.

• Where are the wiring harnesses? Vastly reduced!
• Where's the drive backplane? Gone.
• Sleds? Gone,
• 6 SATA headers, 4 SATA power headers? All gone,
• GPU cards, No longer hand fitted. Also probably manufactured internally and without the need for the power connectors, cables, heat-sinks, or fans.
• ODD drive doors? Gone!
• ODD cage? Gone,
• Rear expansion slot covers and screw? All gone,
• CPU heat-sinks, GPU heat-sinks, Chipset heat-sinks? All morphed into one and set up so that a robot can do it FA style.
• The 3 different port-out PWBs? Became one - even the AC power looks surfaced onto that same singular PWB,
• The card edge headers? Gone. this reduces a lot of expense in just this single exclusion.
• Third party fans? Only one is needed now. There are four on my MP and all are purchased from a 3rd party supplier.

No, totally bro. They have streamlined both cost and production on this new system - maxed it out. As I said elsewhere the new MacPro is an engineering feat and a half! and almost all of that engineering was for Apple's own benefit == lower costs all around!

Yup, a lot of intelligence has been applied to this machine. Now the only thing is to wait and see if it ultimately pays off for them.

Yeah, that's a question I have as well. If they don't intend to offer a dual CPU configuration then why use Xeon? It's useful in this design for something I'm spacing off or?

From an engineer's point of view of which I am, it is a design and engineering feat. Reducing parts means reducing costs. And that includes cost on materials, cost on time, cost on labor. Now instead of using 5 people to fully assemble it, you can use just one because of less parts to put in.

Reliability. Less cable interconnection results to better reliability.

The thermal core is a good design, by using only a single fan it reduces the need for more energy to cool it down. Hot air always goes up, pulling cooler from below is a smart solution. There are some other things that should merit a whole new discussion.

Lastly, environmental impact. Less energy required, less material needed, less impact on the environment.

Yup, absolutely true!

 

[kazmac]

you're probably right...

A mini is for you.

Now that I've come up for air and thoroughly read what this thing can do, this new Mac Pro is more computer than I would need in a couple of life times.

I would like to support the new technology and the U.S.A. builds, but since I am not designing or doing anything reliant on such power said prototype is not the machine for me.*

Here's hoping it is the machine for people who need such power and that Apple starts building more computers in the United States. Takes me awhile, but I think the design geek in me had fainted with glee yesterday.

Thanks for the suggestion. As much as I like the Mini concept, I am skittish in overpaying for older tech.

*Now if they release a bare bones version of this MP that would not be as much of a beast as this prototype is, I may look into it as crazy as this is. Famous last words.

 

[kat.hayes]

The biggest reason I love my current MP is the ability to store multiple drives in it. Now that this is no longer an option, I am open to getting an iMac and somehow connecting my barebones drives via external TB, since this is the same thing I'd have to do with the new MP.

Without having any real world tests, just based on the specs on paper.

1. Would you say that the speed and overall power of the new Macpro blows away the highest end iMac? Does this make the highest end iMac look like a toy?

Thanks.

 

[Topper]

Would you say that the speed and overall power of the new Macpro blows away the highest end iMac? Does this make the highest end iMac look like a toy?

Apples and oranges. The new Mac Pro is a workstation computer (please don't yell at me, people). The iMac is a consumer computer.
The new Mac Pro will perform better with professional applications.
The iMac will most likely out perform the new Mac Pro in gaming.

 

[kat.hayes]

I work with Illustrator, Photoshop and Premiere. Is the difference with the iMac and MP going to be shorter render times? Where else will the performance stand out between the two?

Thanks.

 

[jakesaunders27]

I work with Illustrator, Photoshop and Premiere. Is the difference with the iMac and MP going to be shorter render times? Where else will the performance stand out between the two?

Thanks.

i would have thought the mac pro would be better as it has the fire pro gpu's in which are designed for that kind of stuff whereas iMacs have laptop graphics in

 

[nanofrog]

Every reference design Intel has put out has had the TB controller attached to the I/O Hub.... all of them. If there is a cannonical place where it goes. Generally, that is where Intel tells you to put it.

For a single TB chip, and particularly on a consumer board (i.e. 20 lane PCIe controller based CPU), absolutely.

But that's not the case in this instance. 3x TB chips poses more of a challenge. Not impossible by any means, and even multiple ways to accomplish it (32x lanes for the GPU's, leaving 16 lanes total; 8x gen 3, 8x gen 2).

But here's the rub; the newer controller (DSL4510) does rely on gen 3 to obtain it's additional throughput (doubling the PCIe band per lane is how they were able to double TB's throughput). So placing the next gen TB chips on gen 2 lanes would in fact be a problem (still work, but would be stuck to 10Gbs, as the current crop are).

Thus leaving the 8x gen 3 lanes as the only logical candidate for this particular instance (once the I/O Hubs go to gen 3 PCIe lanes, that will change). And the only way to get 3x gen 2 TB chips on those lanes, would be via a switch (and no need for a buffered unit, though it's still possible if they were using the smaller parts <2channel, 1port>).

I doubt Apple is going to compromise on anything because they are going to want to keep the cost up. It is a 1/8 what it used to be phsycially. I doubt they are looking for 1/8 the price. The 2 GPUs just make the device tread water as far as price goes.

MSRP /= to BOM costs. As the old saying goes, "buy low, sell high". And this is precisely how you increase margins. Build it for as low a cost as reasonably possible, and sell it for the max the marketing data shows will generate the largest gross profit per unit of sales (i.e. quarterly). So it may not have the absolute maximum price possible, it will be priced at the point it will generate the right balance of sales where n sales units * MSRP = greatest income possible.

Now if they're careless with the BOM cost, the margins will decrease. But based on Apple's recent history, they're extremely fond of high margins, so I highly doubt this is the case. Rather, they've figured out the minimum BOM & other manufacturing costs to get the margins as high as possible.

So a simple PCIe switch would almost certainly be more attractive than a more expensive buffered version (though the ability to "change" say 8x gen 3 lanes for 16x gen 2 can be highly useful in certain circumstances).

TB controllers are a switch ICs. That is one of its primary jobs.

Of course. But they're not capable of switching incoming PCIe lanes between itself and any other TB chips that are connected to the same PCIe traces (no master - slave controller + PCIe switch to accomplish this). It requires the PCIe switch to handle the PCIe lane management before the TB chips are tied on. Helps keep the complexity and ultimately, the cost down (DSL4510 is a bit under $10 per in quantity).

Again I don't think they are looking for a more cost effective SSD. They could have used a single SATA lane and reused the SSD from the much higher volume rMBP in here.

Given the throughput claims, this isn't likely (iirc claim 1.2GB/s for the SSD), which means more than a single gen 3 PCIe lane. So I suspect it's more likely connected to 4x gen 2 lanes via the I/O Hub.

Designed in the USA and fabricated in the USA doesn't really scream Foxconn to me.

Foxconn has a plant in operation in Santa Clara, CA. It's where they're currently manufacturing the Google Glass, and according to other articles, there are other plants in the works, such as an LCD panel facility in Detroit.

2012 articles:
Foxconn Planning to Build Manufacturing Plant in the US
Foxconn Planning To Build US-Based Plants, Will Train American Engineers In Taiwan Or China

2013 articles:
Google Glass To Be Built In The United States By Foxconn
Foxconn comes to America to make Google Glass

all Intel Chipsets, even the C600 one has HD Audio built in.

Yes, but I'm under the impression it still uses one of the gen 2 PCIe lanes to transfer that data.

I'll dig a bit later, as I am curious.

I'm a bit skeptical that is the case. As I said Intel really did not increase the aggregate bandwidth with TB v2.0 .

Double the PCIe band is what allowed them to double the TB throughput. Can't shove 20Gbs over the same gen 2 lanes. Either they'd have had to double the lane count on gen 2, or double the band (gen 3 PCIe) to accomplish it, and they went with the latter.

So I don't see it as so much as a reshuffle, but rather just taking advantage of the gen 3 specifications they'd already been working on (doubled bandwidth per lane).

If there is no heavy video traffic that the larger 20Gb/s headroom allows for better congestion control ( and likely latency ) on 12 device chains with varied PCI-e traffic.

True, but they intentionally went with this for the 4K video capability (1x per TB chip). And if 3 monitors are being used along with another TB device, particularly one that pushes quite a bit of data itself such as high speed storage, users will likely notice some choking going on.

 

[telequest]

In contrast to many posters here - shooting from the hip - I've been holding back for half a day, thinking about this.

(snip)

What I find heartening is that Apple's dead serious about the (new) Mac Pro (i guess we'll start referring to it as the "nMP"), they give a sneak preview to build the hype, they've designated it to be built in the USA - in short, they are not going to give up on the Pro segment unless the sales "do the cube".

In summary, many will surely be disappointed (especially those who expected to get these CPU's in the current chassis type), while many will just be relieved, that Apple's not killing the pro lineup.

Sums it up for me. Worst outcome would have been killing off the MP. And in the short run, a real upgrade to the current MP box would have been ideal for many of us. But I'm guessing an incremental upgrade wouldn't ever boost sales enough to make the product interesting to Apple, which would then lead to its eventual demise.

So we have something pretty revolutionary. Time will tell if it's a success for Apple and for Pro Users. Meanwhile, I've got my 2010 Hexacore running full time, and options to boost it with more RAM and SSDs. So I'll wait for early adopters of the new MP to give it a run, and then think about the 2nd generation of the new breed after the kinks get worked out.

Oh no. That sounds like we'll have to start a new thread on how long we'll have to wait for the new Mac Pro to get updated. Sigh.

 

[Mr. Retrofire]

The iMac will most likely out perform the new Mac Pro in gaming.

I doubt that.

 

[theSeb]

I doubt that.

If we consider the baseline Mac Pro will definitely not have the GPUs mentioned in the keynote (up to blah blah blah), then it is certainly feasible that an iMac with a 680mx could be faster for gaming than a base Mac Pro.

 

[ElderBrE]

If we consider the baseline Mac Pro will definitely not have the GPUs mentioned in the keynote (up to blah blah blah), then it is certainly feasible that an iMac with a 680mx could be faster for gaming than a base Mac Pro.

Sorry to say but, what are you smoking?

 

[theSeb]

Sorry to say but, what are you smoking?

I am smoking the good facts and chasing the magic dragon with pragmatism. Considering what we are likely to end up with in the base Mac Pro, then my statement is correct. Unless the base price has been bumped up heavily. We can only speculate until we see all of the configurations.

 

[Umbongo]

Sorry to say but, what are you smoking?

Two W5000s or W7000s working in Crossfire would be better than the 680mx. In the dual Radeon 7850/7870 range.