MP's refresh cycle from now on

[Luba]

Generally, Mac's get updated based on when Intel updates their processors, but the MP skipped a generation (or was 2 generations?). Others have pointed out that Apple probably needed other aspects to fall inline which is why the nMP took so long to come out. Do you think henceforth the MP will update according to when Xeon processors get updated which I think is about every 1.5 years?

 

[ActionableMango]

That's a reasonable assumption, but nobody here knows for sure.

 

[knweiss]

I would be surprised if we didn't see a refresh in ~Q4/2014 as soon as Haswell-EP is available (because it is a very interesting CPU with features like AVX2, etc).

 

[ZnU]

Of course, there's an argument to be made that given the focus of this machine, it should really be updated in sync with AMD's GPU release cycle, not Intel's CPU release cycle.

 

[foidulus]

They are now dependent upon 2 roadmaps now

With the move to dual GPUs standard, mac prop updates are now subject to both Intel Xeon processor updates AND AMDs(maybe someday Nvidia?) product roadmaps, so basically I wouldn't expect to see a new one before the next point the 2 intersect.

Also the time it takes for a new model to come out may also depend on the thermal characteristics of the CPUs, say what you will about the old model, but it was capable of cooling off dual G5s, which was pretty impressive. Not sure what will happen with the new model if the thermal characteristics of the intel cpu/gpus change.

 

[MH01]

Of course, there's an argument to be made that given the focus of this machine, it should really be updated in sync with AMD's GPU release cycle, not Intel's CPU release cycle.

The problem is FirePro refresh cycles are not at common as Radeon . So either way it's not going to be regular

 

[AidenShaw]

With the move to dual GPUs standard, mac prop updates are now subject to both Intel Xeon processor updates AND AMDs(maybe someday Nvidia?) product roadmaps, so basically I wouldn't expect to see a new one before the next point the 2 intersect.

Also the time it takes for a new model to come out may also depend on the thermal characteristics of the CPUs, say what you will about the old model, but it was capable of cooling off dual G5s, which was pretty impressive. Not sure what will happen with the new model if the thermal characteristics of the intel cpu/gpus change.

And why doesn't this mean more frequent updates - a new Mini Pro bump when new GPUs are ready, and a new Mini Pro bump when new CPUs are ready?

Why continue to sell last generation GPUs just because the next generation CPUs aren't quite ready? And vice-versa.

 

[Cisco_Kid]

a new Mini Pro bump when new GPUs are ready, and a new Mini Pro

Whats a Mini Pro ?

 

[MH01]

Whats a Mini Pro ?

The new Mac Pro. Given its size compared to the old model. Just a nickname

 

[AidenShaw]

The new Mac Pro. Given its size compared to the old model. Just a nickname

And given its lack of drive slots, PCIe slots, memory slots, power supply watts, .....

 

[Cisco_Kid]

The new Mac Pro. Given its size compared to the old model. Just a nickname

Thanks mate, I was just being facetious

 

[phoenixsan]

I suppose.....

the refresh cycle on the nMP will be affected not only by offers from Intel, but for how well the new model perform in real world (for the users) and financially (for Apple).

Apple is trying to establish a new paradigm in design and performance. Making compromises after compromises. So, if this initial model goes well, plus another factors (as low prices of Thunderbolt peripherals), maybe the idea Apple is pitching us catches momentum. That will translate in interest from Apple to updates their hardware offers with more frequency. And that can be good, from some point of view, for consumers.....

Dreaming wildly, I can expect a yearly update.....

 

[deconstruct60]

.... Do you think henceforth the MP will update according to when Xeon processors get updated which I think is about every 1.5 years?

First, Xeons in general update on wide variety of schedules. Xeon E3's update yearly. Xeon E5 class (which the Mac Pro will probably use) is pretty close to 1.5 on "tock" ( microarchtecture updates ) and closer to 1.25 on "tick" ( process shrink focused ). Xeon E7 is on an even longer cycle.

If Apple gets disappointed in Mac Pro unit sales the none of the component parts is going to drive the update cycle.

But from a Xeon E5 supply perspective I doubt Apple is going to start on the "bleeding edge" release schedule. For one, Intel seems to have gone to slow-roll releases of E5. A subset of the product line up is rolled out in some limited volume and full ramp takes about a Quarter. Second, at least for this next update, Apple has to turn around a new socket/chipset change in a year ( as oppose to the other vendors which largley used E5 v1 (Sandy Bridge) designs in 2013 and could spend practically the whole year on prepping for E5 v3. ). If there are new GPU cards also that's 3 custom boards on a short design cycle. For a vendor that is shipping out subtsantive firmware fixes on day 1 for their new product, it seems doubtful Apple has fully ramped up the "version 2" design staff and resources at this point. Same short staffing that likely slid the Mac Pro into late 2013 largely prep further slides in the future (at least for next of a couple of iterations).

I would be surprised if we didn't see a refresh in ~Q4/2014 as soon as Haswell-EP is available (because it is a very interesting CPU with features like AVX2, etc).

I suspect that Intel will only effectively "soft release" Xeon E5 v3 ( Haswell-EP ... why the code name when the product name is trivial to figure out??? ) in Q4 '14. It would not be surprising at all to see a slide into Q1 '15.

Of course, there's an argument to be made that given the focus of this machine, it should really be updated in sync with AMD's GPU release cycle, not Intel's CPU release cycle.

Rather weak position. Far more grounded would be that they are on roughly equal footing. If there is a major generational updates in GPUs and none in CPUs there is little reason to hold back on a yearly update. Likewise if the GPU is somewhat stalled and there is a new CPU then move.

AMD cycle is partially dependent upon their fab partners who can flake out ( looking at TMSC as a rather prominent example ). if the GPUs and CPUs drifted 9-13 months out of phase then Apple could use just the one whose phase landed in that product release window to get a next generation Mac Pro out on a yearly basis. The problem is the phase gap is going to drift over time.

The footing isn't exactly equal since the I/O chipset and CPU are basically bundled together. Every tick/tock cycle Intel is going to drive substantive CPU + I/O board updates which will trickle across much of the Mac Pro. The GPUs are far more self contained by the generational changes.

 

[ZnU]

Rather weak position. Far more grounded would be that they are on roughly equal footing. If there is a major generational updates in GPUs and none in CPUs there is little reason to hold back on a yearly update. Likewise if the GPU is somewhat stalled and there is a new CPU then move.

You seem to be arguing for something like a roughly yearly update cycle rather than syncing the cycle with either GPU or CPU releases. That isn't really what I was addressing, and I expect updates may be a bit further apart than that.

 

[deconstruct60]

You seem to be arguing for something like a roughly yearly update cycle rather than syncing the cycle with either GPU or CPU releases.

Given Apple's hard line policy about not talking about new products actually taking action ( DOING something) each year is important. Disappear for random lengths of time is only going to cause problems. While exact dates are not strictly necessary, setting general expectations is.

One major screw-up in Mac Pro product management right now is crappy expectation setting of when new products will generally arrive.

I am not saying do yearly for yearly sake. But it is bone-head move to sit on the product for over a year when there major component(s) that can be relatively straightforwardly updated. If only to keep the whole "the Mac <blank> is dead... " talking to a tolerable minimum.

What I am saying that it is going to be relatively unlikely that in a year nothing has moved. Neither CPUs nor GPUs. Apple, Intel , AMD, and Nvidia all have to be asleep at the wheel for more than year. Even if "pro" CPUs and GPUs go multi year that have to be tied to roughtly the same length extended cycles although using entirely differnt design teams and to some extend different fab processes. What is the rational why they should all sync up and stay sync'ed up.

That isn't really what I was addressing, and I expect updates may be a bit further apart than that.

No. you were addressing tying them to a single component. That's just as goofy for GPUs as it is for CPUs. Even more so when they have different periods of update.

 

[ZnU]

No. you were addressing tying them to a single component. That's just as goofy for GPUs as it is for CPUs. Even more so when they have different periods of update.

Apple for quite a few years did precisely that 'goofy' thing with CPUs, and in fact has gone right on doing it with machines that didn't see the cycle disruption the Mac Pro saw. The original poster was asking if we could expect that to resume now that the re-designed Mac Pro had landed. I pointed out that there was an argument to be made that it made more sense to sync up with GPU releases. I did not take the position you appear to be arguing against, namely that syncing up with GPU releases is the obvious best solution in the universe and that Apple would surely do it.

Personally I would prefer approximately yearly updates, presumably with the exact timing tweaked by a few months here and there to catch convenient nearby CPU/GPU release cycles. But I'm not sure how likely it is that Apple will ever rev this system unless they have both new GPUs and new GPUs to put in it simultaneously. To the extent that they might ever do this, I think they'd be more likely to do with with GPUs, because the only circumstance under which I can see it happening is if CPU speeds really hit a wall (such that it's almost not worth bothering from one generation to the next) but GPU speeds don't. (Of course in a purely hypothetical sense the reverse could happen, with GPU stalling and CPU advancing, but that's not the way the next decade seems to be shaping up.)

 

[calaverasgrande]

I'd expect the next MP update to come after Intel has refreshed the TB spec. The current TB isn't really faster than TB1. It just took 4 10GB pipes and gave it the ability to bond those into 2 20gb pipes. Still same cumulative throughput. Same architecture inside the machine with only 4 PCIe lanes allotted to teh TB controller. They really should have called it TB 1.2.
I'd expect a TB3 to come out with 8 PCIe lanes which would really allow external GPUs and absurdly wide connections. Esp if it can likewise bond 4 10gb pipes into one 40gb while allowing another 4 pipes to exist as 10 or 20gb.

 

[Slow Programmer]

Depending on how it sales this could be the last one.

 

[Luba]

What's the E7 line? Even more heavy duty (stable, can run all-day, heat resistant, workstation durable) than the E5? If E7 updates take longer than E5, then it's very possible the E5 outperforms the E7, correct?

I take it Apple is not using the E3 because it's not heavy duty enough?

First, Xeons in general update on wide variety of schedules. Xeon E3's update yearly. Xeon E5 class (which the Mac Pro will probably use) is pretty close to 1.5 on "tock" ( microarchtecture updates ) and closer to 1.25 on "tick" ( process shrink focused ). Xeon E7 is on an even longer cycle.

If Apple gets disappointed in Mac Pro unit sales the none of the component parts is going to drive the update cycle.

But from a Xeon E5 supply perspective I doubt Apple is going to start on the "bleeding edge" release schedule. For one, Intel seems to have gone to slow-roll releases of E5. A subset of the product line up is rolled out in some limited volume and full ramp takes about a Quarter. Second, at least for this next update, Apple has to turn around a new socket/chipset change in a year ( as oppose to the other vendors which largley used E5 v1 (Sandy Bridge) designs in 2013 and could spend practically the whole year on prepping for E5 v3. ). If there are new GPU cards also that's 3 custom boards on a short design cycle. For a vendor that is shipping out subtsantive firmware fixes on day 1 for their new product, it seems doubtful Apple has fully ramped up the "version 2" design staff and resources at this point. Same short staffing that likely slid the Mac Pro into late 2013 largely prep further slides in the future (at least for next of a couple of iterations).




I suspect that Intel will only effectively "soft release" Xeon E5 v3 ( Haswell-EP ... why the code name when the product name is trivial to figure out??? ) in Q4 '14. It would not be surprising at all to see a slide into Q1 '15.




Rather weak position. Far more grounded would be that they are on roughly equal footing. If there is a major generational updates in GPUs and none in CPUs there is little reason to hold back on a yearly update. Likewise if the GPU is somewhat stalled and there is a new CPU then move.

AMD cycle is partially dependent upon their fab partners who can flake out ( looking at TMSC as a rather prominent example ). if the GPUs and CPUs drifted 9-13 months out of phase then Apple could use just the one whose phase landed in that product release window to get a next generation Mac Pro out on a yearly basis. The problem is the phase gap is going to drift over time.

The footing isn't exactly equal since the I/O chipset and CPU are basically bundled together. Every tick/tock cycle Intel is going to drive substantive CPU + I/O board updates which will trickle across much of the Mac Pro. The GPUs are far more self contained by the generational changes.

 

[Luba]

Did a little reading and it's quite complex. E7 is Enterprise class, but the E5 workstation class CPU, so Enterprise is more demanding than workstations?

E5 supports AVX (Advanced Vector Extensions), but the E7 doesn't, even though it's a higher (better) level CPU?

 

[AidenShaw]

What's the E7 line? Even more heavy duty (stable, can run all-day, heat resistant, workstation durable) than the E5? If E7 updates take longer than E5, then it's very possible the E5 outperforms the E7, correct?

I take it Apple is not using the E3 because it's not heavy duty enough?

I love this urban legend around here that Intel deliberately makes Core i7 CPUs unstable and error prone, and wants you to buy a Xeon if you want a stable CPU.

That's nonsense - the only practical difference in the chips is that Xeons support ECC memory, so they'll fix single bit errors and blue-screen with a memory error code on multi-bit errors. The Core i7s won't see any errors, but may produce wrong results or randomly crash as the result of wrong data.

If your memory isn't having errors, there's no difference.

The E7 is just an E5 that will work in quad-socket configurations. It has an extra inter-socket link.

----------

Did a little reading and it's quite complex. E7 is Enterprise class, but the E5 workstation class CPU, so Enterprise is more demanding than workstations?

E5 supports AVX (Advanced Vector Extensions), but the E7 doesn't, even though it's a higher (better) level CPU?

The current E7 chips are Sandy Bridge - a tick (or tock) behind the E5s.

 

[deconstruct60]

.....

That's nonsense - the only practical difference in the chips is that Xeons support ECC memory, ...

If your memory isn't having errors, there's no difference.

Not really. Xeon E5 support higher memory capacity configs. They also disable overclocking and tweaker knobs. There are more features than just ECC, but ECC is the major one.

The E7 is just an E5 that will work in quad-socket configurations. It has an extra inter-socket link.

Horse hooey. E5 work in quad configrations. The 4600 series does.

The current E7 chips are Sandy Bridge - a tick (or tock) behind the E5s.

More Horse hooey. There never was a Sandy Bridge E7. E7 is jumping straight from Westmere to Ivy Bridge (E7 v2 ) because needed processes shrink to load up on that "larger than E5" core count increase. At this point E7 is not following the normal Intel tick/tock cadence... it is just "tick" (process shrink along with the 'saved up' micro archtecture update "tock")

E7 are more Itanium ( and Power , Alpha , Sparc ) system targeted replacements than. Up to 4 sockets with fewer NUMA hiccups because incrementally better inter-package CPU communication and more RAS features.

 

[deconstruct60]

What's the E7 line? Even more heavy duty (stable, can run all-day, heat resistant, workstation durable) than the E5? If E7 updates take longer than E5, then it's very possible the E5 outperforms the E7, correct?

On a wide range of personal computer apps the E7 doesn't outperform the most members E5 group ( and all of the 1600 subset ). On a $/performance ratio on the same apps it is even a worse debacle. It isn't about "durablity" is it far more about being able implement reliability , available, and servicabilty (RAS).

I take it Apple is not using the E3 because it's not heavy duty enough?

Not using them because can't particularly do any better than an iMac if use them. It is the same class implementation as is used in iMacs. Apple already sells iMacs. For Mac Pro probably going to be looking for something that does something substantialy different over most of the product configurations.

Mac Pro is oversubscribed on PCIe lanes with an E5 with 48 ( 40 E5 8 chipset ) You never could get a practical 3 TB controller system implementation out of using a E3 at the core (24 16 E3 8 chipset ). It is almost half the I/O capability even beyond being capped at 4 x86 cores.