i7 vs Xeon

[Trebuin]

I'm looking at the iMac & iMac pro this winter & wonder what I could lose performance wise when moving to the Xeon. I know that lower end xeons have a lower performance with single core which could impact gaming, but the multi core will improve encoding & rendering. I'm relatively new to the Xeon processor. I also encode in x265

 

[cynics]

Assuming the level of h265 you need is supported at a hardware level (like Kaby Lake CPUs) with the Xeon CPUs Apple uses in the iMac Pro you won't lose anything. The small lose in single core performance will likely be unnoticed for the most part. Oh, you'll lose some weight in your wallet.....

 

[Anonymous Freak]

The Xeon is really just a "renamed" Core i-series CPU. Although often the Xeon uses a design that is 1-2 generations behind the Core i-series, that is likely not true in this case; or at most 1 generation behind, as Intel just released new ultra-high-end Core i-series that are basically the reverse - they are what were previously called Xeons rebadged as Core i5, i7, and i9.

For HVEC (H.265, x265,) the Xeons that are almost assuredly going to end up in the iMac Pro will have hardware level support for encoding and decoding. And if they do use Xeons based on the latest Core i-series, their "Turbo" speed when using a single core will likely be the same as the Turbo speed of the lower-end CPUs in the non-Pro iMac.

The fastest a "configure to order" Core i7-based iMac goes is a quad-core 3.6 GHz "base" speed, 4.2 GHz "Turbo." (That means that if it has been running on all cores so long that it hits heat maximums, it will run at 3.6 GHz. If it isn't at heat limits, and is only using one core, it will run at 4.2 GHz.)

The iMac Pro's likely CPUs include the Xeon equivalent of the details-just-released Core i9 7980XE: 2.6 Ghz "base" speed, 4.4 GHz Turbo. That means if heat limited using all 18 cores, it will run at 2.6 GHz. And the maximum Turbo, which on this CPU covers when *TWO* cores are active, is 4.4 GHz. So for one or two thread applications, it will be faster than the non-Pro iMac. For three or four thread applications, it would probably be about the same (we don't know the exact 3/4 turbo frequencies of the new CPU,) and for anything that can use more than four threads, the Pro would be far faster.

There's even an 8-core variant (Core i7-7820X) that is faster in both base speed and Turbo speed than the highest-end current iMac. So it would be faster in *ANY* workload, even before you consider that it has twice as many cores! That's probably going to be the "base" level processor. The likely 10-core version is the Core i9-7900X.

Again, Intel just released "Core i" CPUs that have specs that previously would have been declared "Xeon" - in all likelihood these are the exact same CPUs that Intel will use, just rebranded Xeon again.

 

[joema2]

....For HVEC (H.265, x265,) the Xeons that are almost assuredly going to end up in the iMac Pro will have hardware level support for encoding and decoding...The iMac Pro's likely CPUs include the Xeon equivalent of the details-just-released Core i9 7980XE....

So far there is no basis for expecting this. No Intel CPU to date over four cores has Quick Sync. This includes all Xeons with over four cores also the new 18-core i7-7980XE -- none of those have Quick Sync, so no on-chip hardware acceleration for encoding and decoding video.

This is a major issue, esp with H.265, which is much more compute-intensive than H.264 thus needs hardware acceleration even more. Maybe Apple or Intel will somehow provide some solution but there is no announcement, leak or historical basis for expecting this.

 

[Anonymous Freak]

On chip acceleration for encoding and decoding video isn't limited to Quick Sync. Quick Sync (which uses the GPU portion) just does it faster (albeit at worse quality.)

At this point, Quick Sync doesn't do H.265 either. H.265 simply has no CPU-level "full hardware encoding" on any CPU, just the use of some vector instructions. Only the earliest of support for GPU encode at this point, either.

I should have phrased it better, though. "What hardware-level support for H.265 encoding there is exists in any feasible CPU that would be used."

Edit: Ah, it appears the very latest generation of desktop/laptop Core i "Kaby Lake" do finally have H.265 encode support. So, there is some chance of H.265 encode not being "fully optimized". Although I have to imagine 18 cores could do it even faster than Quick Sync...

 

[joema2]

On chip acceleration for encoding and decoding video isn't limited to Quick Sync. Quick Sync (which uses the GPU portion) just does it faster (albeit at worse quality.)

At this point, Quick Sync doesn't do H.265 either. H.265 simply has no CPU-level "full hardware encoding" on any CPU, just the use of some vector instructions. Only the earliest of support for GPU encode at this point, either....Edit: Ah, it appears the very latest generation of desktop/laptop Core i "Kaby Lake" do finally have H.265 encode support. So, there is some chance of H.265 encode not being "fully optimized". Although I have to imagine 18 cores could do it even faster than Quick Sync...

Quick Sync is Intel's only on-chip hardware acceleration dedicated to video encode/decode. It does not use the GPU per se, but is totally separate fixed-function logic, IOW an on-chip ASIC. While the normal on-chip GPU logic is not used for Quick Sync, the presence of the on-chip GPU is currently required, probably for access to the bus and frame buffer.

Quick Sync does not encode/decode at worse quality -- the quality is similar to purely software methods, assuming the developers using Quick Sync know their job. Many developers in the past did a poor job of using Quick Sync and other hardware-accelerated methods as well. That is why it hardware accelerated transcoding still has a reputation for lower quality. But anyone who has actually used Quick Sync in FCPX knows the image quality is superb.

Kaby Lake has full encode/decode support for H.265, as can be seen here: http://www.anandtech.com/show/10610...six-notebook-skus-desktop-coming-in-january/3

I recently ran side-by-side tests of a 12-core Mac Pro D700 vs a 2015 iMac 27 vs a 2017 iMac 27, and both machines with Quick Sync were much faster than the 12-core nMP. On transcoding from H.264 to ProRes proxy, the 2017 iMac was TWICE as fast as the 12-core Mac Pro.

Despite having 18 cores, if the iMac Pro doesn't have Quick Sync or some other equivalent technology, it will likely be slower at some transcoding tasks than a 2017 iMac 27.

 

[BlueTide]

I'm looking at the iMac & iMac pro this winter & wonder what I could lose performance wise when moving to the Xeon.

I'm in the same boat, but as it's been for several years now, only time will tell for sure. I feel there are, in general, too many variable; what is the software being used and how that is coded, how does it work with GPU/CPU, what are the exact specs of the CPU, what are the turbo-speeds and other apps running in the background doing to perf, how is the cooling of the case working out, are you disk bound or IO bound... It just starts to compound.

So, personally, I think it's better to wait and see till there are benchmarks and first reports with the machine configuration AND the apps that you plan to use.

 

[Trebuin]

​

I'm in the same boat, but as it's been for several years now, only time will tell for sure. I feel there are, in general, too many variable; what is the software being used and how that is coded, how does it work with GPU/CPU, what are the exact specs of the CPU, what are the turbo-speeds and other apps running in the background doing to perf, how is the cooling of the case working out, are you disk bound or IO bound... It just starts to compound.

So, personally, I think it's better to wait and see till there are benchmarks and first reports with the machine configuration AND the apps that you plan to use.

I'm really thankful for all your inputs so far. I use handbrake for the bulk of my recodes & that uses Intel Quicksync as one of the video encoding options, though I have never targeted that on the Mac in the past. I can't bring it up on my 07, but this is too old anyways. Someone would have to look at a newer one (2017 does have h265 support) & see if handbrake uses it. Handbrake has both x265 & quick sync but I don't know the quality difference in them. I do know the speed of the 2017 iMac was very good.

The cooling is one of the supposed "pros" of the new model that I'm looking forward to. I'm assuming it supports h265 but I have yet to find it. So it seems that the single core processes should be fine, but the I7 may or may not out do the xeons (probably purley chips). The video card's 70-100% performance over the current highest end might be the only real selling factor. If anyone can find something on the Purley (intel scalable) chips & if they support hardware encoding of x265, I would appreciate that.

 

[Anonymous Freak]

Handbrake's Quick Sync option is only for h.264 video, not h.265. (At least, I don't see a Quick Sync h.265 option on my Kaby Lake Windows system.) And I've never seen Quick Sync on Handbrake on macOS. (Although my newest Mac right now is a 2012 Retina MacBook Pro - which has an "Ivy Bridge" CPU, which should support it, but it doesn't show up, so I don't know if Ivy Bridge is just too old or if it just isn't supported on macOS.)

 

[joema2]

....I've never seen Quick Sync on Handbrake on macOS. ...

That is a good point. Just because certain software uses Quick Sync on Windows does not mean that same software uses it on Mac. A good examine is Premiere Pro. It allegedly uses Quick Sync on Windows but not on macOS. Judging from the slow performance on H.264 files, I'm not sure how well it uses that in Windows, but it's definitely not used on Mac.

 

[Jack Burton]

The fastest a "configure to order" Core i7-based iMac goes is a quad-core 3.6 GHz "base" speed, 4.2 GHz "Turbo." (That means that if it has been running on all cores so long that it hits heat maximums, it will run at 3.6 GHz. If it isn't at heat limits, and is only using one core, it will run at 4.2 GHz.)

<snip>

There's even an 8-core variant (Core i7-7820X) that is faster in both base speed and Turbo speed than the highest-end current iMac. So it would be faster in *ANY* workload, even before you consider that it has twice as many cores! That's probably going to be the "base" level processor. The likely 10-core version is the Core i9-7900X.

The fastest config of the iMac is 3.8 base, 4.2 turbo. Configurable to an i7 with 4.2 base 4.5 turbo. The 7820x is 3.6 base, 4.3 turbo, and 4.5 "turbo 3.0".

In real world tests (cinebench single thread, after effects, games that don't take advantage of lots of cores), the 7700k - and therefore the high end iMac - is still king in much of my workflow despite the promises of turbo 3.0. Perhaps software needs to be written to take advantage of it... but in multicore the 7820x is a beast.

The 7820x was the processor I was considering since it almost goes toe to toe with the 7700k in single thread and slays in multithread, but the heat generated from it gave me pause (not to mention it's hardly ever in stock for MSRP). I wonder how Apple is going to handle the heat unless intel gets it under control in the xeon line. Right now Intel advises liquid cooling even when not overclocking.

 

[wallysb01]

The Xeon is really just a "renamed" Core i-series CPU. Although often the Xeon uses a design that is 1-2 generations behind the Core i-series, that is likely not true in this case; or at most 1 generation behind, as Intel just released new ultra-high-end Core i-series that are basically the reverse - they are what were previously called Xeons rebadged as Core i5, i7, and i9.

For HVEC (H.265, x265,) the Xeons that are almost assuredly going to end up in the iMac Pro will have hardware level support for encoding and decoding. And if they do use Xeons based on the latest Core i-series, their "Turbo" speed when using a single core will likely be the same as the Turbo speed of the lower-end CPUs in the non-Pro iMac.

The fastest a "configure to order" Core i7-based iMac goes is a quad-core 3.6 GHz "base" speed, 4.2 GHz "Turbo." (That means that if it has been running on all cores so long that it hits heat maximums, it will run at 3.6 GHz. If it isn't at heat limits, and is only using one core, it will run at 4.2 GHz.)

The iMac Pro's likely CPUs include the Xeon equivalent of the details-just-released Core i9 7980XE: 2.6 Ghz "base" speed, 4.4 GHz Turbo. That means if heat limited using all 18 cores, it will run at 2.6 GHz. And the maximum Turbo, which on this CPU covers when *TWO* cores are active, is 4.4 GHz. So for one or two thread applications, it will be faster than the non-Pro iMac. For three or four thread applications, it would probably be about the same (we don't know the exact 3/4 turbo frequencies of the new CPU,) and for anything that can use more than four threads, the Pro would be far faster.

There's even an 8-core variant (Core i7-7820X) that is faster in both base speed and Turbo speed than the highest-end current iMac. So it would be faster in *ANY* workload, even before you consider that it has twice as many cores! That's probably going to be the "base" level processor. The likely 10-core version is the Core i9-7900X.

Again, Intel just released "Core i" CPUs that have specs that previously would have been declared "Xeon" - in all likelihood these are the exact same CPUs that Intel will use, just rebranded Xeon again.

FYI on the turbo speeds at specific core usage, per http://www.pcgamer.com/full-details-for-intels-core-i9-processor-lineup/:

Pretty damned good really. We'll see if the iMac Pro can keep the i9-7980XE (edit, rather its xeon equivalent) and its 165W cool enough to keep it running at 3.4GHz with 18 cores active, but if so, that's damn impressive. The internet seems full of people hating on Intel for the low "base" clocks on these compared to Threadripper (and its 3.4 GHz base 16 core), which seems odd since this typical behavior for intel to have "base" clocks that your CPU never really operates at.

 

[Trebuin]

So given the single core being slower, but the video card being close to double the speed of the high end 2017 iMac's, I suspect the pro will still crush the current iMac. The difference in the end is the extended theoretical life:
The GPU on the standard iMac models should match the lower end pro's in about 3 years, but the iMac pro will cost me about $2300 more due to memory & SSD bumps. I'll end up dropping about 3600 (4000 with taxes) on the 2017. I could pay around $2700 for a 2020 model, matching current pro specs, $6300 minus the sell of the 2017. In the end, probably the same about spent if I chose to go either way...the only difference is the hassle of selling the 2017 model, & knowing I get a new model in 3 years. That's given that video cards can actually advance through continued die shrinks. A tax free day would help put a pretty good dent in the cost.