Click Here
home features news forums classifieds faqs links search
5752 members 
Amiga Q&A /  Free for All /  Emulation /  Gaming / (Latest Posts)
Login

Nickname

Password

Lost Password?

Don't have an account yet?
Register now!

Support Amigaworld.net
Your support is needed and is appreciated as Amigaworld.net is primarily dependent upon the support of its users.
Donate

Menu
Main sections
» Home
» Features
» News
» Forums
» Classifieds
» Links
» Downloads
Extras
» OS4 Zone
» IRC Network
» AmigaWorld Radio
» Newsfeed
» Top Members
» Amiga Dealers
Information
» About Us
» FAQs
» Advertise
» Polls
» Terms of Service
» Search

IRC Channel
Server: irc.amigaworld.net
Ports: 1024,5555, 6665-6669
SSL port: 6697
Channel: #Amigaworld
Channel Policy and Guidelines

Who's Online
76 crawler(s) on-line.
 13 guest(s) on-line.
 0 member(s) on-line.



You are an anonymous user.
Register Now!
 Steady:  18 mins ago
 Argo:  26 mins ago
 matthey:  1 hr ago
 AlfredOne:  1 hr 2 mins ago
 fishy_fis:  1 hr 9 mins ago
 agami:  1 hr 50 mins ago
 arthoropod:  1 hr 52 mins ago
 Bradsco:  2 hrs 3 mins ago
 MEGA_RJ_MICAL:  2 hrs 18 mins ago
 IRTheBorg:  2 hrs 43 mins ago

/  Forum Index
   /  Amiga General Chat
      /  Some Power related news
Register To Post

Goto page ( Previous Page 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 Next Page )
PosterThread
Rob 
Re: Some Power related news
Posted on 6-Jun-2020 18:17:50
#301 ]
Elite Member
Joined: 20-Mar-2003
Posts: 5915
From: S.Wales

@matthey

Quote:
Tabor may be able to beat that with a descent gfx card *if* the floating point problem doesn't affect performance too much. Tower 57 gave some hope of acceptable 3D performance for the bastard PPC core.


Dan's ports of Tower57 are purely software rendered.

 Status: Offline
Profile     Report this post  
AmigaBlitter 
Re: Some Power related news
Posted on 6-Jun-2020 23:30:46
#302 ]
Elite Member
Joined: 26-Sep-2005
Posts: 3433
From: Unknown

@thread

https://www.semiaccurate.com/2020/06/03/is-ibm-killing-off-power/

_________________
retired

 Status: Offline
Profile     Report this post  
matthey 
Re: Some Power related news
Posted on 6-Jun-2020 23:34:32
#303 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

Rob wrote:
Dan's ports of Tower57 are purely software rendered.


3D software rendering does more floating point calculations than 3D hardware rendering where the GPU offloads much of the work. Perhaps you are saying that it is only a psuedo-3D environment which could avoid most if not all of the fp use?

 Status: Offline
Profile     Report this post  
Rob 
Re: Some Power related news
Posted on 6-Jun-2020 23:48:30
#304 ]
Elite Member
Joined: 20-Mar-2003
Posts: 5915
From: S.Wales

@matthey

It's a 2D game.

 Status: Offline
Profile     Report this post  
matthey 
Re: Some Power related news
Posted on 7-Jun-2020 0:11:21
#305 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

AmigaBlitter wrote:
https://www.semiaccurate.com/2020/06/03/is-ibm-killing-off-power/


I have said it before. I would not be surprised if IBM did stop development of POWER. IBM will not say ahead of time they will be stopping as potential customers would be reluctant to buy existing POWER chips. Freescale didn't say they were abandoning PPC either until it was more than obvious. Some customers want more than one source of chips anyway. Rumors have it that Apple requested Motorola start producing PPC chips in addition to IBM before deciding to switch to PPC, that Motorola licensed the 68000 technology to Hitachi to satisfy customer requests for multiple vendors and that AMD ended up with x86 as an alternate supplier for Intel. Who else produces POWER? High end CPUs are crazy expensive to develop and have produced with the latest technology. POWER has barely made a dent in the server market and most of that is due to IBM reputation. IBM does get some big DARPA contracts but otherwise I wonder about the future of POWER. IBM has not been afraid to unload low margin businesses like they did with their foundries. I don't know if anyone would want POWER though. Margins would likely drop on POWER chip sales without IBM's reputation. My guess is that IBM will keep POWER around for DARPA contracts, big customer requests and sometimes as a loss leader but that development will slow and be downsized.

 Status: Offline
Profile     Report this post  
matthey 
Re: Some Power related news
Posted on 7-Jun-2020 1:09:53
#306 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

Rob wrote:
It's a 2D game.


Many modern games like Tower 57 use 3D gfx. This one is old school then. The gfx are impressive if they are all hand drawn instead of 3D rendered.

 Status: Offline
Profile     Report this post  
Hypex 
Re: Some Power related news
Posted on 9-Jun-2020 16:49:36
#307 ]
Elite Member
Joined: 6-May-2007
Posts: 9974
From: Greensborough, Australia

@matthey

The short answer is no. Can you give some insight into the long answer?

 Status: Offline
Profile     Report this post  
matthey 
Re: Some Power related news
Posted on 9-Jun-2020 23:49:49
#308 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

Hypex wrote:
The short answer is no. Can you give some insight into the long answer?


In reference to POWER?

POWER is being beaten by x86_64 in the high end CPU market where x86_64 CPUs have better single core single thread performance which is better for general purpose and gaming use. POWER CPUs were close to being competitive in overall performance and energy efficiency for the server market by improving multi-threading efficiency but I believe they struggled to lower the price enough to be competitive, especially with the margins IBM expects from the past. Various ARM offerings have tried to compete in these markets before exiting as well.

PowerPC is losing market share in the high end embedded market to ARM. High end embedded ARM designs have been getting more powerful and are cheaper than custom IBM designs. ARM's ISAs look to me to be superior to the PPC ISA for embedded use (RISC-V compressed ISAs as well). POWER was likely a loss leader in recent years for custom PPC designs but both POWER and PPC appear to be in decline. This is not a cyclical decline but a systematic problem that is likely to get worse over time. The embedded industry has been one of the fastest growing markets (especially IoT and custom SoC designs) which IBM has been investing in. I expect they will offer a different ISA than POWER/PPC which will be another sign of the end for POWER/PPC. IBM is already "collaborating" with ARM.

https://newsroom.ibm.com/2015-09-03-IBM-and-ARM-Collaborate-to-Accelerate-Delivery-of-Internet-of-Things

IBM will likely have PPC cores available for custom SoC designs for some time. The PowerPC 476FP is competitive with NXP PPC cores.

https://www-03.ibm.com/press/us/en/pressrelease/28399.wss

Maybe IBM has been forced to lower prices due to lack of PPC demand and the core could be licensed at a discount compared to a few years ago. The problem for the Amiga is lack of mass production though. Good luck at making PPC popular enough to mass produce again too.

 Status: Offline
Profile     Report this post  
Rose 
Re: Some Power related news
Posted on 10-Jun-2020 17:54:10
#309 ]
Cult Member
Joined: 5-Nov-2009
Posts: 705
From: Unknown

Meanwhile in reality....


https://www.phoronix.com/scan.php?page=news_item&px=GCC-POWERXX-IS-POWER10

Coming of POWER10 isn't surprise to anyone who follows industry, it was announced already last october. There are few niches where Power9 is great and I don't see any reason why Power10 wouldn't be even better.

 Status: Offline
Profile     Report this post  
matthey 
Re: Some Power related news
Posted on 11-Jun-2020 7:13:34
#310 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

Rose wrote:
Meanwhile in reality....




IBM hasn't been doing so well in the server market. The competition is mostly using cheap x86_64 hardware as you can probably tell by the names. IBM develops it's own products (very expensive) instead of using commodity hardware and free software. They have been forced to cut costs to remain competitive.



IBM's server percentage includes z/Architecture mainframes (System/3x0 CISC architecture modernized to 64 bit, multicore and 5+ GHz frequencies) which I believe accounts for more than half of their server percentage. That doesn't leave much market share for POWER.

The embedded market has been growing substantially faster than the server market yet PowerPC use has been declining. I couldn't find any pics to show but just look at some 2019 embedded poll results.

https://www.embedded.com/wp-content/uploads/2019/11/EETimes_Embedded_2019_Embedded_Markets_Study.pdf

See the question, "Which of the following 32-bit chip families would you consider for your next embedded project?"

Xilinx Virtex-5 (with PowerPC 405) 5%
NXP MPC5xxx 3%
Freescale/NXP PowerPC 55xx 3%
Xilinx Virtex-4 (with PowerPC 405) 2%
Freescale/NXP PowerPC 5xx, 6xx 2%
Freescale/NXP PowerPC 7xx, 8xx 2%
Freescale/NXP 68k, ColdFire 2%
Freescale/NXP PowerQUICC 1%
AMCC PowerPC 4xx 1%
IBM PowerPC 4xx, 7xx 1%

PowerPC has one bright spot in the Xilinx Virtex FPGA where 1-2 PowerPC 405 cores can be integrated with custom FPGA logic. Nice flexible setup for embedded use allows up to 400MHz PowerPC 405. Could a PPC Amiga use it? Not practical. The Virtex is the high end Xilinx FPGA line costing many times what the low end Altera Cyclone FPGAs used in the Amiga Vampire hardware costs. As I recall, Gunnar did some testing in a higher end Altera FPGA around 400MHz as well so if anyone wants to buy a Virtex with PowerPC 405 core to compare, I'd love to hear the results, but my money is on the Apollo Core.

The other PPC chips above have been losing market share and are close to being just blips on the radar. Even the 68k/ColdFire is right there with them. The IBM PPC chips are really old. IBM was more careful about producing off the shelf processors after the PPC 970 (G5) problems. Custom designs for customers would not be shown in the poll and are too expensive for all but the largest embedded projects. One thing that is clear from the poll, ARM is dominating the embedded market.

 Status: Offline
Profile     Report this post  
AmigaBlitter 
Re: Some Power related news
Posted on 11-Jun-2020 15:07:07
#311 ]
Elite Member
Joined: 26-Sep-2005
Posts: 3433
From: Unknown

@thread

https://pbs.twimg.com/ad_img/1267495468717666306/ljdtmVy_?format=png&name=small

_________________
retired

 Status: Offline
Profile     Report this post  
AmigaBlitter 
Re: Some Power related news
Posted on 1-Jul-2020 15:53:45
#312 ]
Elite Member
Joined: 26-Sep-2005
Posts: 3433
From: Unknown

@thread

https://www.nextplatform.com/2020/06/30/big-blue-open-sources-the-core-inside-bluegene-q-supercomputers/

_________________
retired

 Status: Offline
Profile     Report this post  
AmigaBlitter 
Re: Some Power related news
Posted on 1-Jul-2020 16:23:31
#313 ]
Elite Member
Joined: 26-Sep-2005
Posts: 3433
From: Unknown

@AmigaBlitter

From tnp article:

"
Here’s the fun bit. Implemented in 45 nanometers, the Power-A2 integer core was 2.9 square millimeters in area and generated 0.9 watts at 2.3 GHz. Projecting out to 7 nanometer processes (presumably implemented at Samsung, IBM’s current fab partner for Power chips), this area will drop to 0.17 square millimeters for the integer core and power will be 0.15 watts at the base 3 GHz and 0.5 watts at the top speed of 4.2 GHz.

In addition to opening up the Power-A2 core, IBM is getting requests to open up other cores and chips, including some of its older 32-bit designs, believe it or not. People are also interested in getting the game console chips IBM created for Sony and Microsoft opened up – we think the “Cell” variant of the Power line, famously used in the “Roadrunner” supercomputer at Los Alamos National Laboratory from 2009 to 2013, would be interesting to play around with, too.

And there is always a chance that Power8, Power9, or even Power10 could be opened up at the core or even fully. Anything is possible. We have suggested – and Furmanek did not say this was stupid but she also made no commitments – that Power10 should be the last chip that IBM designs by itself and that Power11 should be the first one that involves a community of chip designers all helping out. This would be very hard to do, we think, sort of like building an elephant by committee. But there should be a way to get a more varied Power chip line created than what IBM is delivering today, and opening up the core on the Power-A2 is a good second step. The PowerISA being open was never enough, but was also a good first step."

_________________
retired

 Status: Offline
Profile     Report this post  
Hypex 
Re: Some Power related news
Posted on 1-Jul-2020 17:16:32
#314 ]
Elite Member
Joined: 6-May-2007
Posts: 9974
From: Greensborough, Australia

@matthey

Quote:
In reference to POWER?


Yes, since the article is above my pay grade and gives no details.

Quote:
Note: The following is for professional and student level subscribers.


Thanks for the information.

Quote:
Maybe IBM has been forced to lower prices due to lack of PPC demand and the core could be licensed at a discount compared to a few years ago. The problem for the Amiga is lack of mass production though. Good luck at making PPC popular enough to mass produce again too.


Well the Amiga demand wouldn't meet any mass production numbers I think. It relies on the crumbs left behind. I don't know if open sourcing the Power ISA will help in this regard.

 Status: Offline
Profile     Report this post  
matthey 
Re: Some Power related news
Posted on 2-Jul-2020 1:40:24
#315 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

AmigaBlitter wrote:
From tnp article:

"
Here’s the fun bit. Implemented in 45 nanometers, the Power-A2 integer core was 2.9 square millimeters in area and generated 0.9 watts at 2.3 GHz. Projecting out to 7 nanometer processes (presumably implemented at Samsung, IBM’s current fab partner for Power chips), this area will drop to 0.17 square millimeters for the integer core and power will be 0.15 watts at the base 3 GHz and 0.5 watts at the top speed of 4.2 GHz.


It's not difficult to make low power cores. It is difficult getting performance from low power cores. First of all, this is only a core probably with register files and maybe the 16kiB L1 ICache and 16kiB L1 DCache. A 16kiB L1 ICache will likely have more ICache misses than on a 68060 with 8kiB ICache as every 25%-30% improvement in code density is like doubling the ICache size. These are the same cache sizes as the PPC 603e which was barely adequate for desktop use. The core is in order superscalar which has typically given weak performance for RISC cores although this is not uncommon for older "throughput" cores. It has 4 threads per core which helps the core do more work per core but each process can actually take more time as it switches between threads and shares the small caches.

The core is designed for edge of network embedded applications for IoT devices where less code is used as workloads are similar (packet processing and encryption/decryption). ARM throughput cores are the Cortex-A53 (Raspberry Pi 3 cores), Cortex-A55 and now the Neoverse E1.

https://community.arm.com/developer/ip-products/processors/b/processors-ip-blog/posts/arm-neoverse-e1-platform-empowering-the-infrastructure-to-meet-next-generation-throughput-demands

The Neoverse E1 core gained OoO and SMT but is still less than 4W.

Quote:

In addition to opening up the Power-A2 core, IBM is getting requests to open up other cores and chips, including some of its older 32-bit designs, believe it or not. People are also interested in getting the game console chips IBM created for Sony and Microsoft opened up – we think the “Cell” variant of the Power line, famously used in the “Roadrunner” supercomputer at Los Alamos National Laboratory from 2009 to 2013, would be interesting to play around with, too.


CELL cores are also designed for throughput. The cores are mostly in order with a deep pipeline for high clock rates but integer performance is disappointing. For example, a PPC G5 core at 1.8GHz outperforms a CELL core at 3.2GHz in 7-Zip compression and decompression benchmarks.

https://www.7-cpu.com/

Higher core clocks means electricity can't travel as far in one cycle. Dividing the core into more but smaller stages solves this problem at a cost of more transistors. Cache accesses take more cycles unless cache sizes are reduced to make them faster (why we have multi-level caches). Even with muli-level caches, cache accesses often become more cycles increasing the load-use latency with RISC. Branch misprediction penalties increase so more transistors are required for improved branch prediction. These cores are difficult to program and instruction scheduling becomes very important to avoid all the bubbles.

CELL core
Branch misprediction penalty = 24 cycles
Integer instruction latency = 2 cycles
Integer MUL latency = 11 cycles
FPU 64-bit instruction latency = 10 cycles
L1 load latency = 5 cycles (normally a 6 cycle load-use penalty but likely a few tricks to reduce?)

It's a waste to clock the core up and then increase the cycles of everything. IBM probably did this to clock the SPE units up for more vector throughput. It looks good on paper and can be for multimedia heavy workloads but is not what you want for general purpose use.

Quote:

And there is always a chance that Power8, Power9, or even Power10 could be opened up at the core or even fully. Anything is possible. We have suggested – and Furmanek did not say this was stupid but she also made no commitments – that Power10 should be the last chip that IBM designs by itself and that Power11 should be the first one that involves a community of chip designers all helping out. This would be very hard to do, we think, sort of like building an elephant by committee. But there should be a way to get a more varied Power chip line created than what IBM is delivering today, and opening up the core on the Power-A2 is a good second step. The PowerISA being open was never enough, but was also a good first step."


It is no coincidence that the first core opened was in order and avoids most side channel attacks like Spectre. I doubt much of modern POWER cores will be opened as it could expose security vulnerabilities in hardware people are still using. IBM must be getting desperate to even consider this. We can see they are wanting partners to spread out the development costs.

 Status: Offline
Profile     Report this post  
MEGA_RJ_MICAL 
Re: Some Power related news
Posted on 3-Jul-2020 1:25:04
#316 ]
Regular Member
Joined: 13-Dec-2019
Posts: 326
From: AMIGAWORLD.NET WAS ORIGINALLY FOUNDED BY DAVID DOYLE

ZORRAM

_________________
I HAVE ABS OF STEEL

 Status: Offline
Profile     Report this post  
AmigaBlitter 
Re: Some Power related news
Posted on 6-Jul-2020 13:53:53
#317 ]
Elite Member
Joined: 26-Sep-2005
Posts: 3433
From: Unknown

@MEGA_RJ_MICAL

https://www.theregister.com/2020/07/03/open_chip_hardware/

_________________
retired

 Status: Offline
Profile     Report this post  
bison 
Re: Some Power related news
Posted on 6-Jul-2020 18:48:58
#318 ]
Super Member
Joined: 18-Dec-2007
Posts: 1637
From: N-Space

@AmigaBlitter

Even with all the restrictions they could probably fab something more useful for Amiga than the P1022.

_________________
"Unix is supposed to fix that." -- Jay Miner

 Status: Offline
Profile     Report this post  
matthey 
Re: Some Power related news
Posted on 6-Jul-2020 22:16:35
#319 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

bison wrote:
Even with all the restrictions they could probably fab something more useful for Amiga than the P1022.


The 130nm process is old being introduced around 2001. However, the 68060 was introduced in a 600nm process and there is more die size reduction from 600nm to 130nm (470nm reduction) than from 130nm to 5nm (125nm reduction) today. Older processes have some nice advantages like low cost, high yields, less current leakage concerns, etc.

Since this is the POWER news thread, AmigaBlitter was probably pointing out the IBM open core. The other two news stories may have more significance to POWER though. The IBM core by itself really isn't that useful or interesting. The whole open source SonicBOOM RISC-V core has good performance (claimed highest performance open source core) and nicely documented high tech features like predicated short forward branches and instruction and data prefetching (features of the Apollo Core too). The micro op OoO still has the bottleneck in memory as much energy in wasted moving loads sooner when calculating an EA can take several cycles longer than ISAs with more complex addressing modes. At least the core supports RISC-V compressed ISA extensions for the extra instructions needed which most Linux distros for RISC-V are using. This gives RISC-V better code density than ARM AArch64 and leaves PPC in the dust. It's getting easier and cheaper to make custom SoCs where POWER/PPC further looses out.

I was home this weekend and asked my brother (an engineer with a MBA) what he would do based on the 1st chart of IBM server revenues I showed above. He said he would introduce new more competitive products or exit the market. Anyone who thinks POWER has a guaranteed future with IBM doesn't know much about business.

Last edited by matthey on 23-Jul-2020 at 04:02 PM.

 Status: Offline
Profile     Report this post  
AmigaBlitter 
Re: Some Power related news
Posted on 14-Jul-2020 14:58:49
#320 ]
Elite Member
Joined: 26-Sep-2005
Posts: 3433
From: Unknown

@thread

Interesting power blog:

https://www.riscyslack.org/?Blog


_________________
retired

 Status: Offline
Profile     Report this post  
Goto page ( Previous Page 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 Next Page )

[ home ][ about us ][ privacy ] [ forums ][ classifieds ] [ links ][ news archive ] [ link to us ][ user account ]
Copyright (C) 2000 - 2019 Amigaworld.net.
Amigaworld.net was originally founded by David Doyle