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.
Will there be a new Mac Pro in 2015?
- Thread starter Anto38x
- Start date
- Sort by reaction score
- Status
http://www.golem.de/news/grafikkarte-amd-benchmarks-sehen-r9-nano-vor-der-r9-290x-1508-115897.html From the view of performance the core clock should be around 850 MHz with 150W of power consumption. The power consumption is calculated however, only from the results, and 260W of power consumption for R9 290X.
Not bad, not great, IMO. Was hoping for more.
Anyways, counting the performance it will be around 7 TFLOPs of compute power from 175W of TDP.
Titan X has 250W of TDP and 6.7 TFLOPs of compute power.
Im starting to wonder how that will affect the Mantle-ish applications, where compute power and asynchronous shaders are the most important factor.
Things are getting interesting.
Not bad, not great, IMO. Was hoping for more.
Anyways, counting the performance it will be around 7 TFLOPs of compute power from 175W of TDP.
Titan X has 250W of TDP and 6.7 TFLOPs of compute power.
Im starting to wonder how that will affect the Mantle-ish applications, where compute power and asynchronous shaders are the most important factor.
Things are getting interesting.
Last edited:
.
Well, for the people who really want Apple to make a normal desktop tower, they'll obviously not be happy.
But as someone who is interested in the Mac Pro in the future, really just the updates themselves are almost more important than the contents. It sucks to get a computer with an uncertain future and support status. Obviously Intel has made upgrades more complex than they need to be, but Apple shoulders some blame too. Even if they just straight-up said "we expect to upgrade every two generations" so people know they are actively working on it.
Aside from that, for my own use cases better GPUs, TB3, and a second SSD slot inside the computer would be all I'm really hoping for (although the last one depends a lot on just how much bandwidth there is.) And then I'd love for my software to actually catch up to my hardware... (as I learn there is no multi-threaded rendering in After Effects 2015. Thanks, Adobe.)
Given that there only 40 PCIe lanes available on Haswell-E and the upcoming Broadwell-E, and current rumors are 48 PCIe 3.0 lanes on Skylake-E (not including those on the PCH), it seems like there are two possibilities:
1. Broadwell-E update sometime between Q4 2015 and Q3 2016:
2x GPUs occupying 32 PCIe 3.0 lanes
1x Thunderbolt 3 occupying 4 PCIe 3.0 lanes
1x SSD PCI-e controller occupying 4 PCIe 3.0 lanes
= 40 lanes
2. Skylake-E update sometime between Q1 2017 and Q4 2017:
2x GPUs occupying 32 PCIe 3.0 lanes
3x Thunderbolt 3 occupying 12 PCIe 3.0 lanes
1x SSD PCI-e controller occupying 4 PCIe 3.0 lanes
= 48 lanes
Which do you think is more likely?
For reference, the current Mac Pro:
2x GPUs occupying 32 PCIe 3.0 lanes
3 TB2 controllers x 4 PCIe 2.0 lanes (connect to the remaining 8 PCIe 3.0 lanes)
1. Broadwell-E update sometime between Q4 2015 and Q3 2016:
2x GPUs occupying 32 PCIe 3.0 lanes
1x Thunderbolt 3 occupying 4 PCIe 3.0 lanes
1x SSD PCI-e controller occupying 4 PCIe 3.0 lanes
= 40 lanes
2. Skylake-E update sometime between Q1 2017 and Q4 2017:
2x GPUs occupying 32 PCIe 3.0 lanes
3x Thunderbolt 3 occupying 12 PCIe 3.0 lanes
1x SSD PCI-e controller occupying 4 PCIe 3.0 lanes
= 48 lanes
Which do you think is more likely?
For reference, the current Mac Pro:
2x GPUs occupying 32 PCIe 3.0 lanes
3 TB2 controllers x 4 PCIe 2.0 lanes (connect to the remaining 8 PCIe 3.0 lanes)
Last edited:
http://ark.intel.com/products/75522/Intel-DH82C226-PCH
I think the most possible option at this point is update early 2016 with announcement in October and preorders for the computer in late 2015.
I have to say, I did not know that USB does not count for the CPU PCiEx lanes.
Also it turns out that one controller with PCiEx 3.0 4x will provide 2 Thunderbolt 3 ports. Low Power version of it will use 2x PCiEx 3.0 and will bring one port.
I think the most possible option at this point is update early 2016 with announcement in October and preorders for the computer in late 2015.
I have to say, I did not know that USB does not count for the CPU PCiEx lanes.
Also it turns out that one controller with PCiEx 3.0 4x will provide 2 Thunderbolt 3 ports. Low Power version of it will use 2x PCiEx 3.0 and will bring one port.
So not one, but two Thunderbolt 3 ports available in next gen Mac Pro.wikipedia said:
I hope you're right! It looks like the Broadwell-EP chips that would be a good fit for the new Mac Pro, assuming they're going to skip Haswell-EP, are confirmed for Q4 2015. This would line up well for an early 2016 release.
The question is, will Apple be satisfied dropping the number of thunderbolt ports from 6, to 2? Or, maybe they would try to hang two more Thunderbolt 2 ports off of the PCH given that there would be 4 remaining PCIe 2.0 lanes available there assuming they move the SSD controller to the CPU to get PCIe 3.0 speeds. So two Thunderbolt 3's, two Thunderbolt 2's... to me this seems great, but maybe apple would say this is confusing and chose not to update and wait till 2017.
As for USB, on the C600 chipset (currently used in the Mac Pro), USB 2.0 is built in, but USB 3.0 required an extra PCI-e 2.0 lane since it wasn't baked into the chipset. If Apple decides to upgrade this year or next with C610/Broadwell-E then this would not be an issue.
Only two tb ports.
How do you think this is going to impact monitor connectivity?
Are there going to be additional mini displayports directly from GPUs?
How do you think this is going to impact monitor connectivity?
Are there going to be additional mini displayports directly from GPUs?
I doubt they would release a machine with only 2 TB ports if that's the only way to do monitor connections. They could do:
2 TB 3, USB-C ports (from CPU)
2 TB 2 ports (from PCH)
2 Mini DP only ports (straight from GPU)
I think this would confuse people if they couldn't plug their Apple Thunderbolt display into the DP only ports, it seems like Apple wouldn't go this route.
Here's another possibility
Keep SSD speeds at PCIe 2.0 x4 speeds, leave it on the PCH controller. This would give them 8 lanes for:
4 TB 3, USB-C ports
To me this seems unlikely because they would be holding back SSD speeds compared to what's currently on the market.
Someone correct me if I'm wrong, but couldn't 3 thunderbolt 3 controllers with 6 USB-C ports share the same 8x PciE pool of bandwidth? While this wouldn't increase the collective pool of bandwidth available, it would increase the maximum bandwidth available to any individual controller.
Another question, does displayport count against the pool of bandwidth? If its a displayport monitor, I assume the port would fallback to displayport mode and receive its signal directly from the GPU. If its a thunderbolt display, wouldn't DP just be carried over thunderbolt without requiring any bandwidth from the CPU?
Another question, does displayport count against the pool of bandwidth? If its a displayport monitor, I assume the port would fallback to displayport mode and receive its signal directly from the GPU. If its a thunderbolt display, wouldn't DP just be carried over thunderbolt without requiring any bandwidth from the CPU?
While I agree in part, Apple's SSD speeds are still largely unparalleled in the larger PC industry, which is only just starting to move to m.SATA and are barely in the water with PCIe SSDs. They're really only competing with their own offerings.
Yes, 3 TB 3 controllers could share the same x8 lanes of bandwidth. The only issue is that if you saturated over four ports you would start to see degradation in bandwidth at each port. This wasn't an issue for TB 2 because it runs off of PCIe 2.0, meaning that there was extra bandwidth left over to support all 6 ports at specification.
Not sure exactly how it works, but I believe TB just passes the DP signal. However, if the TB controller is totally saturated and some of the 6 ports can't operate at their specified bandwidth, they might not be able to pass the DP signal properly. For example, if you have 6 ports that are supposed to pass DP, but the TB 3 controller only has the bandwidth to support 4 ports of DP signal.
While true, PCI-e 3.0 x4 M.2 slots that operate at 32gb/s are standard on new motherboards. These would outperform the current Mac Pro at PCI-e 2.0 x4.
The somewhat implicit artificial crisis here is that these are mutually exclusive. They are solutions from two different points in time. Apple could do Q1 2016 update followed by a Q1-Q2 2017 update. This whole update the Mac Pro randomly every 2-3 years product management isn't a good strategic plan. There is "surprise and delight" the customer and then there is 'lazy not really paying attention to customer needs' treating the product as a "ego hobby".
You are leaving out the 8 PCIe 2.0 lanes on the I/O PCH chipset. If take the x4 PCIe 2.0 SSD load off the chipset you would free up space for the TB v2 controller. one 40 Gb/s TB v3 controller + 1 20 Gb/s TB v2 controller = 60 Gb/s aggregated. 3 * 20 Gb/s --> 60 Gb/s aggregate.. It is a wash.
What Apple couldn't solve short term would be to go to TB v3 and provision two internal SSDs. I'm not sure they want to open the 2nd SSD option short or long term. It would likely open up the market a bit, but they'd need to shift to a more standards based slot for the 2nd SSD for that to be highly effective.
The core problem for the Mac Pro 2013 is that it is a 2013 model. Time is an issue. Can Apple squat on a 2013 product for another 16-20 months and remain competitive? That would be an extremely dubious position to put themselves into. This story is rebranded Dell sales marketing but illustrate of Apple is competing against.
http://www.cnet.com/news/dell-tries-stealing-creative-pros-away-from-apple/
The longer Apple plays at "Ostrich with head in the sand" the far more effective these Dell ( and HP ... they have one too) campaigns get. Maybe the Mac Pro is being product managed by a bozo, but there are only a couple of more effective ways to shoot the Mac Pro product line in the head other than squat and do nothing for almost 4 years.
It's certainly interesting to see Dell and HP move in so aggressively. Realistically I think this is good, because if you aren't served by Apple you have people working on getting you a deal and getting you up to speed on another platform. I'm wondering how effective a long-term business strategy it can be, though. Focusing on creatives and pros worked for Apple to a degree, but I'd say that for a good chunk of that history Apple was clearly dying, and the world isn't like the PC landscape of the 90s and early 2000s—desktops are increasingly irrelevant for a large part of the public, and workstations themselves following the same trend. For now, they're the only place where there's still a strong technical reason to demand new hardware—I can do all the page layout, spreadsheets and word processing I want on a PowerMac if I wanted to.
According to this report, the workstation market is essentially <1 million units. [1] It's a damn small slice of pie to keep fighting over and to rely on heavily. Apple is insulated from their own failings and the larger whims and malaise of the PC market. Will be interesting to see how HP and Dell fare.
According to this report, the workstation market is essentially <1 million units. [1] It's a damn small slice of pie to keep fighting over and to rely on heavily. Apple is insulated from their own failings and the larger whims and malaise of the PC market. Will be interesting to see how HP and Dell fare.
Yeah I agree, it's ridiculous that they don't update every CPU generation, but they've proven that they don't intend to do this time and time again.
Right, I mentioned in a subsequent post that they could do:
2 TB 3, USB-C ports (from CPU)
2 TB 2 ports (from PCH)
and still be able to pull 4x PCIe 3.0 lanes for an SSD from the CPU.
However, what you're suggesting is that they could use something like a PLX chip for the TB 3 controller to share bandwidth from both the CPU and the PCH? This sounds not possible. The bandwidth of DMI 2.0 is 2 GB/s.
If want to invite switch problems you can dilute the bandwidth almost done to zero. 2 * 4x PCIe v3 --> 12 PCIe v3. So have potential demand cuoncurrent x12 on x8. Plus each stage down TB chain of 2-3 is another set of switches of a different type. I'm not sure what the rational motivation for that is. Especially in the performance ( get most out of what you got without straying into overclocking/out-of-spec) zone. The more modern PCH chipsets for the Xeon E5 v3 and up have USB 3.0 built in. There is about zero good reason to bleed PCIe lanes off the CPU subset to provision USB 3.0 when you have already bought it from Intel in the chipset. It would be a redundant implementation with extremely little added value if solely just providing USB 3.0. Finally, TB v3 has defacto has a USB controller built in. There is limited need another/different USB-C ports if have TB v3 ones.
The available flexibility on TB v3 is higher. You can do more with fewer ports with TB v3 if using modern equipment. (e.g., two 4K streams on one cable ). The Mac Pro could still easily drive 6 'normal' , no-HDI montors off of a just a pair of TB v3 ports. if had 2 TB v 3 ports and 2 TB v2 ports then don't need any legacy dongles for TB v1-2 equipment and still have as much PCI-e bandwidth as before.
If have 4-6 USB 3.0 sockets then don't have to use the TB v3 ones for USB. There isn't a good rational to load up TB ports if you don't have to. The premise of having to need 3 pairs of TB v3 sockets on a Mac Pro because have to load of all of the external connectivity onto those 6 ports is fundamentally flawed. One socket type for everything on a back panel the size of a Mac Pro is drinking tons of Type-C kool-aid. Having some USB TYpe A and mini-DisplayPort ports too would be a good thing given the amount and depth of peripherals that Mac Pro owner ( both old and new ) already own.
Further out into systems downstream two things will happen. First, there will be more PCIe v3 lanes ( whether from I/O PCH chipset and/or directly from CPU package ). If it is a display output issue the USB 3.x with Type C probably will drive a wide set of external monitors too. The GPU output doesn't all have to go out the TB ports. ( there is already a HDMI port, so not exclusive now anyway. )
It doesn't increase anything. It enables single tracking usage of a just limited subset of ports appear to work but actually are not adding any bandwidth at all. It is an exercise in dilution not provisioning more.
On a motherboard piling PCIe switches on top of other PCIe switches isn't a good thing. Plus all of these additional switches consume space and power.
I know, but that's what we said with the Quadra 840AV, 120MB SCSI drives, 640x480 video, and 16bit/48khz audio. Onward and upward! (Something about Moore's Law should go here.)
I agree that there isn't necessarily a need for 3 thunderbolt controllers and 6 USB-C ports. I think the rationale for 3 controllers and 6 ports with shared bandwidth is simply because the ports share double duty with displayport. On the current mac pro, if you want to run 3 displayport monitors, that leaves you with 3 TB ports and 2 TB controllers for peripherals.
However, I don't see Apple offering USB-C and miniDP on the same machine. I can't think of a case where Apple has increased the variety of ports. So a combination of USB-C/TB3 and mini-DP/TB2 doesn't seem likely. You may be right though that its not feasible to offer 3 TB3 controllers. But if they cut down on ports, that would mean the machine is more dependent on dongles. Of course, Apple loves dongles, so maybe its not that far fetched.
Since we already have USB-C on the macbook, I foresee USB-C becoming standard across the line, with everything above the macbook getting USB-C/TB3, replacing miniDP and the magsafe connector. This would come with the new set of adapters that would make every USB-C hub something that combines power, USB and displayport into one dongle.
Apple may have jumped off the cycle of following every Tick/Tock cycles with the Mac Pro product line. So may end up with every 2 years cycle. As long as it is a regular two year cycle I think the vast majority of potential Mac Pro users will readily adjust to that. They need not to buy frequently either, so it is just a matter of aligning with expectations. In that context jumping to the tail end of the cycle "tick" (the 'shrink') is super conservative and safe since the chipset for this class tends to spend the whole tick/tock cycle. So dealing with relatively safe, bug well known components.
However, skipping a whole tick/tock cycle is completely lacking in reasonable motivation. Especially for Xeon E5 which is not on a 12 month pace. Makes even less sense when consider the GPUs are increasing important to computational performance. The likelihood that they are also stagnant for a full Xeon E5 tick/tock cycle is about zero. Even if the CPUs didn't move something else major probably has.
If anything it looks like Apple may have going to a 'tail-wags-dog' philosophy and have coupled the Mac Pro to Thunderbolt update cycle.... which IMHO is delusional from a strategic perspective. The CPU/GPUs are out of kiter because using TB to align the upgrades.
Not suggesting that at all. There is just one each of the TB controllers ( one TB v3 and one TB v2). That's four ports. Implicitly was saying you can get rid of the PLX altogether. If move a x4 SSD to the CPU then have two x4 consumers. There is no need for any PLX switch. If you move one TB v2 down to the I/O PCIe lanes vacated by the "old" x4 SSD then have just two TB controllers. They are on different sources.
Counting the total/aggregate TB bandwidth available on the back panel of the Mac Pro this 4 TB port model with a mix of of v3 and v2 have just as much bandwidth as the old 6 port model. The current one isn't getting the absolute max out of the x8 PCIe v3 lanes. Short term, it isn't a backslide in terms of TB bandwidth to fall down to four TB ports. TB bandwidth is not PCIe bandwidth.
2 GB/s --> 16 Gb/s Which is enough to cover the defacto x3 PCI-e v2.0 ( 15 Gb/s bandwidth ) the TB v2 controller consumes. There are lots of deployed one TB v2 controllers systems out there sitting on a DMI 2.0 link. It works.
Yep, I completely agree, there's no way we'll see TB 2 and TB 3 on the next Mac Pro. 2 TB 3 controllers, 4 USB type-C TB 3 ports, with dongles if you want more displays than that.
I can see a mix of TB2 and TB3 ports as much more likely than USB3.1 t1 and USB3.1 w/ Type C connector. Apple aggressively drops old tech but not *that* aggressively (we got ADB on El Capitan, we got FW400 on the PowerMac G5s—and early Mac Pros, etc.) Insofar as keeping a legacy port I think a TB2 would be far more useful for most pros then the old square USB.
For those who beat the dual-processor train, you can point to the math we're doing as one of the problems with the current single-processor design.
On the other hand, using that in your assumptions means that they would have needed to design the nMP as a dual-processor standard piece of kit from the get-go.
Yeah that is a good point about FW400 and FW800 sticking around for a while.
I imagine the old square USB will stick around because of ubiquitousness. It's already implemented on the chipset, so why not include it? If Apple really wanted to get rid of it for the Mac Pros they would have done so during the 2013 redesign.
A question I have is whether it is even possible to have the Thunderbolt 2 controller use PCIe lanes from the PCH rather than directly from the CPU.
- Status