On Monday, 10 September 2018 at 14:00:43 UTC, Iain Buclaw wrote:
> I've just got back from a conference where AArch64 was declared a disaster, and the future is now PPC64 and RISC-V.

References, please! :)

On Sat, 10 Nov 2018 at 16:25, Per Nordlöw via Digitalmars-d <digitalmars-d@puremagic.com> wrote:
>
> On Monday, 10 September 2018 at 14:00:43 UTC, Iain Buclaw wrote:
> > I've just got back from a conference where AArch64 was declared a disaster, and the future is now PPC64 and RISC-V.
>
> References, please! :)

Pub, Manchester. If you weren't there, you missed out. ;-)

-- 
Iain

On Saturday, 10 November 2018 at 15:27:31 UTC, Iain Buclaw wrote:
> Pub, Manchester. If you weren't there, you missed out. ;-)

;) Ok, until next time.

On Saturday, 10 November 2018 at 11:45:22 UTC, Rubn wrote:
> On Saturday, 10 November 2018 at 07:29:17 UTC, Joakim wrote:
>> On Monday, 10 September 2018 at 13:43:46 UTC, Joakim wrote:
>>> LDC recently added a linux/AArch64 CI for both its main branches and 64-bit ARM, ie AArch64, builds have been put out for both linux and Android. It does not seem that many are paying attention to this sea change that is going on with computing though, so let me lay out some evidence.
>>>
>>> [...]
>>
>> Two more interesting bits of news I saw lately:
>>
>> - The octa-core AArch64 CPU in the latest iPad Pro benchmarks better than last year's core i7 Macbook Pro, a bit behind this year's core i9 Macbook Pro:
>>
>> https://arstechnica.com/gadgets/2018/11/apple-walks-ars-through-the-ipad-pros-a12x-system-on-a-chip/
>
> Not really that impressive, my desktop i5-2500k from like 8 years ago is still faster than the CPU in last years Macbook pro. Intel has been stagnate with their 14nm processors. Even this years CPUs is still based on the same architecture as their CPU from 2-3 years ago.

No, benchmarks show the mobile i7 in last year's MBP as faster than your old desktop core i5, so the iPad Pro is faster than your desktop. :P Of course, the TDP of your desktop i5 is twice the mobile i7 and almost 10 times the A12X in the new iPad Pro, so Apple completely blows away Intel on the performance-to-power-dissipation ratio.

> With Ryzen we're expected to see a CPU next year with 64 cores and 128 threads at 7nm.

There are 64-bit ARM workstations and server cores if you want that:

https://www.anandtech.com/show/12571/gigabyte-thunderxstation-cavium-thunderx2-socs

> That GPU performance though, pretty impressive considering how large the heatsinks and amount of fans are on traditional GPUs.

Indeed.

On Tuesday, 13 November 2018 at 17:41:31 UTC, Joakim wrote:
> On Saturday, 10 November 2018 at 11:45:22 UTC, Rubn wrote:
>> On Saturday, 10 November 2018 at 07:29:17 UTC, Joakim wrote:
>>> On Monday, 10 September 2018 at 13:43:46 UTC, Joakim wrote:
>>>> LDC recently added a linux/AArch64 CI for both its main branches and 64-bit ARM, ie AArch64, builds have been put out for both linux and Android. It does not seem that many are paying attention to this sea change that is going on with computing though, so let me lay out some evidence.
>>>>
>>>> [...]
>>>
>>> Two more interesting bits of news I saw lately:
>>>
>>> - The octa-core AArch64 CPU in the latest iPad Pro benchmarks better than last year's core i7 Macbook Pro, a bit behind this year's core i9 Macbook Pro:
>>>
>>> https://arstechnica.com/gadgets/2018/11/apple-walks-ars-through-the-ipad-pros-a12x-system-on-a-chip/
>>
>> Not really that impressive, my desktop i5-2500k from like 8 years ago is still faster than the CPU in last years Macbook pro. Intel has been stagnate with their 14nm processors. Even this years CPUs is still based on the same architecture as their CPU from 2-3 years ago.
>
> No, benchmarks show the mobile i7 in last year's MBP as faster than your old desktop core i5, so the iPad Pro is faster than your desktop. :P Of course, the TDP of your desktop i5 is twice the mobile i7 and almost 10 times the A12X in the new iPad Pro, so Apple completely blows away Intel on the performance-to-power-dissipation ratio.

Which benchmarks did you look at, cause I have mine overclocked past 4 GHz.

Not faster than my desktop where it counts, at least for me (the gpu). Nothing really needs that much more power for CPU, and single thread performance hasn't gone up that much.

>> With Ryzen we're expected to see a CPU next year with 64 cores and 128 threads at 7nm.
>
> There are 64-bit ARM workstations and server cores if you want that:
>
> https://www.anandtech.com/show/12571/gigabyte-thunderxstation-cavium-thunderx2-socs

Don't think it would compare in terms of performance. As even that new fancy ipad pro CPU isn't anywhere near a current-day desktop CPU.

On Tuesday, 13 November 2018 at 21:10:08 UTC, Rubn wrote:
> On Tuesday, 13 November 2018 at 17:41:31 UTC, Joakim wrote:
>> On Saturday, 10 November 2018 at 11:45:22 UTC, Rubn wrote:
>>> On Saturday, 10 November 2018 at 07:29:17 UTC, Joakim wrote:
>>>> [...]
>>>
>>> Not really that impressive, my desktop i5-2500k from like 8 years ago is still faster than the CPU in last years Macbook pro. Intel has been stagnate with their 14nm processors. Even this years CPUs is still based on the same architecture as their CPU from 2-3 years ago.
>>
>> No, benchmarks show the mobile i7 in last year's MBP as faster than your old desktop core i5, so the iPad Pro is faster than your desktop. :P Of course, the TDP of your desktop i5 is twice the mobile i7 and almost 10 times the A12X in the new iPad Pro, so Apple completely blows away Intel on the performance-to-power-dissipation ratio.
>
> Which benchmarks did you look at, cause I have mine overclocked past 4 GHz.

The same Geekbench site as in the iPad Pro article usually has the MBP i7 ahead:

https://browser.geekbench.com/v4/cpu/search?utf8=✓&q=2500k
https://browser.geekbench.com/v4/cpu/search?utf8=✓&q=7820hq

> Not faster than my desktop where it counts, at least for me (the gpu). Nothing really needs that much more power for CPU, and single thread performance hasn't gone up that much.
>
>>> With Ryzen we're expected to see a CPU next year with 64 cores and 128 threads at 7nm.
>>
>> There are 64-bit ARM workstations and server cores if you want that:
>>
>> https://www.anandtech.com/show/12571/gigabyte-thunderxstation-cavium-thunderx2-socs
>
> Don't think it would compare in terms of performance. As even that new fancy ipad pro CPU isn't anywhere near a current-day desktop CPU.

Did you read the link? That's not an iPad, but a workstation with an AArch64 server CPU, similar to the Ryzen you mentioned.

On Tuesday, 13 November 2018 at 17:41:31 UTC, Joakim wrote:
> No, benchmarks show the mobile i7 in last year's MBP as faster than your old desktop core i5, so the iPad Pro is faster than your desktop. :P Of course, the TDP of your desktop i5 is twice the mobile i7 and almost 10 times the A12X in the new iPad Pro, so Apple completely blows away Intel on the performance-to-power-dissipation ratio.

While the last statement may hold some truth, I sure as hell don't trust Geekbench numbers. And all ARM/x86 comparisons seem to be based on that one little, highly controversial benchmark suite.

Their multicore results suggest a Ryzen 1950X outperforms an i7 7700K by only 50%. A small testsuite of 8 real-world apps (incl. raytracing, video encoding, compression) shows a min gain of 80% and max gain of 200%, definitely averaging above 100%: https://www.computerbase.de/2018-08/amd-ryzen-threadripper-2990wx-2950x-test/2/#diagramm-blender-280
As with any benchmark, your milage may vary, but a dedicated multicore score apparently not scaling well at all makes me wonder.

Unless someone gives me hard numbers that their phone/tablet can compile something like LLVM/LDC in a time anywhere near their desktop CPU, I don't waste a second believing that a ~5W CPU can match a 65+ W desktop one, even a 5-years old one like mine. Laptops would surely already be shipping with these chips if that was the case.

On Tuesday, 13 November 2018 at 22:39:34 UTC, kinke wrote:
> On Tuesday, 13 November 2018 at 17:41:31 UTC, Joakim wrote:
>> No, benchmarks show the mobile i7 in last year's MBP as faster than your old desktop core i5, so the iPad Pro is faster than your desktop. :P Of course, the TDP of your desktop i5 is twice the mobile i7 and almost 10 times the A12X in the new iPad Pro, so Apple completely blows away Intel on the performance-to-power-dissipation ratio.
>
> While the last statement may hold some truth, I sure as hell don't trust Geekbench numbers. And all ARM/x86 comparisons seem to be based on that one little, highly controversial benchmark suite.
>
> Their multicore results suggest a Ryzen 1950X outperforms an i7 7700K by only 50%. A small testsuite of 8 real-world apps (incl. raytracing, video encoding, compression) shows a min gain of 80% and max gain of 200%, definitely averaging above 100%: https://www.computerbase.de/2018-08/amd-ryzen-threadripper-2990wx-2950x-test/2/#diagramm-blender-280
> As with any benchmark, your milage may vary, but a dedicated multicore score apparently not scaling well at all makes me wonder.

These broad benchmarks are indicative, but the best is obviously to try it out with your own software.

> Unless someone gives me hard numbers that their phone/tablet can compile something like LLVM/LDC in a time anywhere near their desktop CPU, I don't waste a second believing that a ~5W CPU can match a 65+ W desktop one, even a 5-years old one like mine. Laptops would surely already be shipping with these chips if that was the case.

As my first post notes, 64-bit ARM laptops are now shipping and more are on the way:

https://www.anandtech.com/show/13498/samsung-unveils-galaxy-book2-12inch-snapdragon-850-with-x20-lte-20-hrs

https://wccftech.com/snapdragon-8180-new-details-octacore-soc/

On Tuesday, 13 November 2018 at 23:20:08 UTC, Joakim wrote:
> As my first post notes, 64-bit ARM laptops are now shipping and more are on the way:
>
> https://www.anandtech.com/show/13498/samsung-unveils-galaxy-book2-12inch-snapdragon-850-with-x20-lte-20-hrs
>
> https://wccftech.com/snapdragon-8180-new-details-octacore-soc/

We'll see where that goes, the Windows ARM adventure with x86 emulation had to fail of course.

If the power efficiency of ARM was really much better than x86, I wonder why it hasn't exploded in the server market yet, where software compatibility shouldn't play that big a role. Judging by the performance of the AArch64 boxes provided by Shippable/Packet (using -j16 leading to a performance similar to -j3 for an x86 CI service, for the LDC CI suite, but requiring obviously a lot more memory for that throughput), I guess it's not that much better when the x86 chips are tweaked for throughput (server CPUs: lower frequencies, lower voltages, much more cores).

On 15/11/2018 12:27 AM, kinke wrote:
> On Tuesday, 13 November 2018 at 23:20:08 UTC, Joakim wrote:
>> As my first post notes, 64-bit ARM laptops are now shipping and more are on the way:
>>
>> https://www.anandtech.com/show/13498/samsung-unveils-galaxy-book2-12inch-snapdragon-850-with-x20-lte-20-hrs 
>>
>>
>> https://wccftech.com/snapdragon-8180-new-details-octacore-soc/
> 
> We'll see where that goes, the Windows ARM adventure with x86 emulation had to fail of course.
> 
> If the power efficiency of ARM was really much better than x86, I wonder why it hasn't exploded in the server market yet, where software compatibility shouldn't play that big a role. Judging by the performance of the AArch64 boxes provided by Shippable/Packet (using -j16 leading to a performance similar to -j3 for an x86 CI service, for the LDC CI suite, but requiring obviously a lot more memory for that throughput), I guess it's not that much better when the x86 chips are tweaked for throughput (server CPUs: lower frequencies, lower voltages, much more cores).

Some are applying ARM for servers. Cloudflare is a great example of this[0].

But they do have to invest quite a bit in R&D to get everything working right. Which may or may not have something to do with it.

[0] https://blog.cloudflare.com/neon-is-the-new-black/