By: Groo (charlie.delete@this.semiaccurate.com), January 29, 2023 9:58 am
Room: Moderated Discussions
--- (---.delete@this.redheron.com) on January 28, 2023 10:37 am wrote:
> I think that seeing these things through a lens of "ethics" (which usually boils down
> to "I don't want to pay more for better stuff" is not useful or informational.
>
> A more useful lens is whether particular behavior grows the ecosystem.
> *That* is my concern with the details of how Intel is handling this.
> Every CPU that does not make functionality available to a developer is a CPU that is no longer growing
> the Intel ecosystem. Apple is running essentially the opposite strategy, giving every user pretty much
> everything (AMX, NPU, media engines, ...; though admittedly without an optimal level of documentation
> and APIs), and I think the Apple strategy is a lot more powerful for creating an ecosystem.
>
> If you want to engage in market segmentation, I think the trick is to do it by *capacity*, not by *capability*.
> Give everyone AVX512 (and BNNI, and the accelerators, and the rest of it) but either
> - at the low end make them work, but not as fast as at the high end. (AVX512 double- or even quad-pumped) OR
> - insert some sort of silly counter so that the user can essentially get fast
> "home use" of AVX512/accelerator for up to X billion instructions per day after
> which they are throttled, plus an On Demand key to end the throttling.
> Both of these grow the ecosystem in a way that simply not having the functionality does not.
> (And as for "exploiting" users, well, if AMD, or ARM..., run their market numbers and see that they're
> better off selling all users full capability 24/7 so be it. This is not an issue of "ethics" it's an
> issue of someone has to pay for this functionality, and let the best of these payment schemes win!)
>
Works great unless you have a competitor or two with quantifiably better products for less cost who also are not hated because of prior customer abuses. If Intel is in any of these situations, extorting customers in a very noticeable and ham-handed way is IMHO not the smartest way forward. Then again I have been specifically telling this to the people making those decisions since before the Cascade Lake launch and fat lot of good that did. :)
-Charlie
> I think that seeing these things through a lens of "ethics" (which usually boils down
> to "I don't want to pay more for better stuff" is not useful or informational.
>
> A more useful lens is whether particular behavior grows the ecosystem.
> *That* is my concern with the details of how Intel is handling this.
> Every CPU that does not make functionality available to a developer is a CPU that is no longer growing
> the Intel ecosystem. Apple is running essentially the opposite strategy, giving every user pretty much
> everything (AMX, NPU, media engines, ...; though admittedly without an optimal level of documentation
> and APIs), and I think the Apple strategy is a lot more powerful for creating an ecosystem.
>
> If you want to engage in market segmentation, I think the trick is to do it by *capacity*, not by *capability*.
> Give everyone AVX512 (and BNNI, and the accelerators, and the rest of it) but either
> - at the low end make them work, but not as fast as at the high end. (AVX512 double- or even quad-pumped) OR
> - insert some sort of silly counter so that the user can essentially get fast
> "home use" of AVX512/accelerator for up to X billion instructions per day after
> which they are throttled, plus an On Demand key to end the throttling.
> Both of these grow the ecosystem in a way that simply not having the functionality does not.
> (And as for "exploiting" users, well, if AMD, or ARM..., run their market numbers and see that they're
> better off selling all users full capability 24/7 so be it. This is not an issue of "ethics" it's an
> issue of someone has to pay for this functionality, and let the best of these payment schemes win!)
>
Works great unless you have a competitor or two with quantifiably better products for less cost who also are not hated because of prior customer abuses. If Intel is in any of these situations, extorting customers in a very noticeable and ham-handed way is IMHO not the smartest way forward. Then again I have been specifically telling this to the people making those decisions since before the Cascade Lake launch and fat lot of good that did. :)
-Charlie
Topic | Posted By | Date |
---|---|---|
NYT on SPR | --- | 2023/01/26 10:37 AM |
NYT on SPR | Chris G | 2023/01/26 06:02 PM |
NYT on SPR | me | 2023/01/26 07:44 PM |
NYT on SPR | Anne O. Nymous | 2023/01/27 01:09 AM |
NYT on SPR | Michael S | 2023/01/27 03:22 AM |
NYT on SPR | --- | 2023/01/27 10:31 AM |
Pat has been trimming the Intel product portfolio | Mark Roulo | 2023/01/27 01:29 PM |
NYT on SPR | James | 2023/01/27 02:00 PM |
NYT on SPR | Adrian | 2023/01/28 03:55 AM |
NYT on SPR | anonymou5 | 2023/01/28 04:03 AM |
NYT on SPR | Adrian | 2023/01/28 04:14 AM |
NYT on SPR | Groo | 2023/01/29 09:50 AM |
NYT on SPR | Groo | 2023/01/29 09:46 AM |
NYT on SPR | Brendan | 2023/01/29 01:00 PM |
NYT on SPR | Anon4 | 2023/01/29 04:06 PM |
NYT on SPR | Brendan | 2023/01/29 07:03 PM |
NYT on SPR | Groo | 2023/01/30 07:09 AM |
NYT on SPR | Groo | 2023/01/29 09:39 AM |
NYT on SPR | AnonSoft | 2023/01/30 11:01 AM |
NYT on SPR | hobold | 2023/01/30 12:39 PM |
NYT on SPR | AnonSoft | 2023/01/30 05:34 PM |
NYT on SPR | hobold | 2023/01/31 04:40 AM |
NYT on SPR | Jukka Larja | 2023/01/31 07:13 AM |
Heterogeneous CPU Cores With OpenMP | Mark Heath | 2023/02/01 04:45 AM |
Heterogeneous CPU Cores With OpenMP | Freddie | 2023/02/01 05:05 AM |
Heterogeneous CPU Cores With OpenMP | Mark Heath | 2023/02/01 06:42 AM |
Heterogeneous CPU Cores With OpenMP | Freddie | 2023/02/01 09:54 AM |
Heterogeneous CPU Cores With OpenMP | Mark Heath | 2023/02/01 04:45 PM |
Heterogeneous CPU Cores With OpenMP | —- | 2023/02/02 04:35 PM |
Heterogeneous CPU Cores With OpenMP | Freddie | 2023/02/02 04:39 PM |
Heterogeneous CPU Cores With OpenMP | --- | 2023/02/03 12:15 PM |
Heterogeneous CPU Cores With OpenMP | Freddie | 2023/02/03 03:46 PM |
Heterogeneous CPU Cores With OpenMP | Anne O. Nymous | 2023/02/03 12:57 AM |
Heterogeneous CPU Cores With OpenMP | --- | 2023/02/03 12:35 PM |
Heterogeneous CPU Cores With OpenMP | Anne O. Nymous | 2023/02/03 01:35 PM |
different big/little split.. | Heikki Kultala | 2023/02/03 02:33 PM |
Heterogeneous CPU Cores With OpenMP | Paul H | 2023/02/03 06:51 PM |
Heterogeneous CPU Cores With OpenMP | Jukka Larja | 2023/02/01 06:24 AM |
When heavily loaded, Threads run about equally fast on E-cores than P-cores | Heikki Kultala | 2023/02/01 02:08 PM |
NYT on SPR | Chester | 2023/01/27 09:30 AM |
use archive.org | anon | 2023/01/27 06:08 PM |
Bypassing paywalls | Doug S | 2023/01/28 02:05 PM |
NYT on SPR | Chris G | 2023/01/27 06:54 PM |
Intel On Demand | Chris G | 2023/01/28 04:24 AM |
Intel On Demand | me | 2023/01/28 06:24 AM |
Intel On Demand | Groo | 2023/01/29 09:53 AM |
Intel On Demand | rwessel | 2023/01/28 09:41 AM |
Intel On Demand | --- | 2023/01/28 11:37 AM |
Anit-waste bias | Paul A. Clayton | 2023/01/28 07:57 PM |
Intel On Demand | Groo | 2023/01/29 09:58 AM |
Intel On Demand | Andrey | 2023/01/30 05:04 PM |
Intel On Demand | blaine | 2023/01/28 03:07 PM |
Intel On Demand | me | 2023/01/28 03:25 PM |
Intel On Demand | me | 2023/01/28 03:33 PM |
Intel On Demand | Chris G | 2023/01/28 07:06 PM |
Intel On Demand | me | 2023/01/28 07:43 PM |
Intel On Demand - Validation, certification? | Björn Ragnar Björnsson | 2023/01/28 10:41 PM |
Intel On Demand - Validation, certification? | anonymou5 | 2023/01/29 02:49 AM |
Sapphire Rapids crippleware is a naked money grab | Chris G | 2023/01/29 04:44 AM |
Intel On Demand - Validation, certification? | Groo | 2023/01/29 10:05 AM |
Intel On Demand - Validation, certification? | AnotherAnonymousEngineer | 2023/01/29 10:33 AM |
Intel On Demand - Validation, certification? | Groo | 2023/01/29 11:16 AM |
Intel On Demand - Validation, certification? | dmcq | 2023/01/29 04:32 PM |
Intel On Demand - Validation, certification? | Brendan | 2023/01/29 08:01 PM |
Intel On Demand - Validation, certification? | Groo | 2023/01/30 07:17 AM |
Intel On Demand - Validation, certification? | Freddie | 2023/01/30 11:36 AM |
Intel On Demand - Validation, certification? | anon2 | 2023/01/30 07:41 PM |
Intel On Demand - Validation, certification? | anon2 | 2023/01/31 01:35 AM |
Crippleware | Chris G | 2023/01/31 05:47 AM |
Doctorow calls it "enshittification" (NT) | hobold | 2023/01/31 07:55 AM |
Crippleware | anon2 | 2023/01/31 10:51 AM |
Crippleware | Groo | 2023/02/01 02:06 PM |
Crippleware | anon2 | 2023/02/01 05:10 PM |
Crippleware | Chris G | 2023/02/01 05:52 PM |
Crippleware | anon2 | 2023/02/01 09:15 PM |
SPR Volume | me | 2023/02/02 04:47 AM |
SPR Volume | anon2 | 2023/02/02 07:04 AM |
Crippleware | Chris G | 2023/02/02 08:12 AM |
Crippleware | anon2 | 2023/02/02 08:42 AM |
Crippleware | anon2 | 2023/02/02 08:48 AM |
Crippleware | Charles | 2023/02/01 01:38 AM |
Crippleware | Chris G | 2023/02/01 02:59 AM |
language digression | Matt Sayler | 2023/02/01 04:53 PM |
Crippleware | me | 2023/02/01 06:27 PM |
Crippleware | Chris G | 2023/02/01 07:01 PM |
Crippleware | me | 2023/02/01 07:10 PM |
Crippleware | Chris G | 2023/02/01 09:32 PM |
Crippleware | Tony | 2023/02/01 11:18 PM |
Crippleware | me | 2023/02/02 04:27 AM |
Crippleware | anonymou5 | 2023/02/02 03:47 AM |
Crippleware | Chris G | 2023/02/02 05:59 AM |
Intel On Demand - Enshittification | blaine | 2023/01/30 12:13 AM |
Intel and mobile phones | James | 2023/01/29 09:09 AM |
Intel and mobile phones | Maxwell | 2023/01/29 02:25 PM |
Intel and mobile phones | Groo | 2023/01/30 07:20 AM |
Intel and mobile phones | anonymous2 | 2023/01/30 11:15 AM |
Intel and mobile phones | Doug S | 2023/01/30 12:51 PM |
Intel and mobile phones | Daniel B | 2023/01/31 07:37 AM |
Intel and mobile phones | Groo | 2023/02/01 02:03 PM |
SPR HBM | me | 2023/01/29 09:17 AM |
SPR-W | me | 2023/02/17 05:41 PM |
Accelerators on AMD/ARM | Chester | 2023/01/29 05:41 PM |