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
84 crawler(s) on-line.
 11 guest(s) on-line.
 1 member(s) on-line.


 fishy_fis

You are an anonymous user.
Register Now!
 fishy_fis:  2 mins ago
 agami:  34 mins ago
 arthoropod:  36 mins ago
 Bradsco:  47 mins ago
 MEGA_RJ_MICAL:  1 hr 1 min ago
 IRTheBorg:  1 hr 27 mins ago
 bison:  1 hr 39 mins ago
 amisteph:  2 hrs 20 mins ago
 tygre:  2 hrs 44 mins ago
 matthey:  3 hrs 1 min ago

/  Forum Index
   /  Amiga General Chat
      /  Amiga hardware poll
Register To Post

Goto page ( Previous Page 1 | 2 | 3 | 4 | 5 | 6 | 7 Next Page )
Poll : What should happen to Amiga hardware?
Develop cheap 68k SoC for embedded, toys/games, retro, hobby
Develop expensive PPC SoC for desktop, laptop
POWER for desktop AmigaOS (no SMP or 64 bit addressing)
POWER for desktop with redesigned & incompatible AmigaOS
No more embarrassing Amiga hardware!
No opinion or pancakes
 
PosterThread
Hypex 
Re: Amiga hardware poll
Posted on 23-Jun-2020 17:35:20
#81 ]
Elite Member
Joined: 6-May-2007
Posts: 9974
From: Greensborough, Australia

@Raffaele

Quote:
Cross compilers do already exist and output code for any architecture...


Sure, but do they add the "__amigaos4__" tag so it loads on OS4? Does the startup code open dos.library? And does the main() routine have a working IExec interface pointer?

 Status: Offline
Profile     Report this post  
Hypex 
Re: Amiga hardware poll
Posted on 23-Jun-2020 18:20:42
#82 ]
Elite Member
Joined: 6-May-2007
Posts: 9974
From: Greensborough, Australia

@bison

Quote:
Apple announces Mac transition to Apple silicon


This announcement is rather strange. They refer to themselves in third person, is that normal for Apple?

Also it's rather vague. What does Apple silicon actually mean? Have they designed their own CPU?

Not once in the article did I see any mention of ARM. It's just implied. Neither did they directly say they were dropping Intel.

A few highlights:

Quote:
Developer Transition Kit (DTK), a Mac development system


Reminds me of the Amiga/DE SDK and OS4 x86 development system.

Quote:
Virtualization technology allows users to run Linux.


What LOM? Linux on Mac? MOL will be on the flip side!

Quote:
Developers can apply to the program at developer.apple.com, and the total cost of the program is $500.


Excellent. Let's raise a bounty. And prompty send it to Hyperion for porting OS4.

Quote:
Apple revolutionized personal technology with the introduction of the Macintosh in 1984.


Haha!

Another point on moving to another CPU. Apple is frequently used as an example. But there is a difference. The moved CPU like AmigaOS did. But, after they moved to PPC, they eventually dumped MacOS. And replaced it with OSX. OS4 is still at the MacOS stage because they haven't dumped it nor plan to do so. The Amiga Inc idea of OS4 on x86 to develop OS5 would be the sort of thing needed for AmigaOS to turn into a modern reincarnation. But then it might be like an AmigaOSX.

 Status: Offline
Profile     Report this post  
Hypex 
Re: Amiga hardware poll
Posted on 23-Jun-2020 18:28:12
#83 ]
Elite Member
Joined: 6-May-2007
Posts: 9974
From: Greensborough, Australia

@BigD

Quote:
I'm officially out once my current Mac dies!


Why? Do you like Intel Macs? Dislike ARM? They can't use the confusing PowerPC logo anymore. But they will be able to steal from the Amiga scene Intel Outside!

Quote:
BootCamp is the ONLY killer app left on macOS IMHO.


Seriously? A program that lets you run Windows? I'm wondering where your honour lies here. Unless they fixed it up BootCamp is annoying. When I tested it didn't work with Windows. Apple did something so Windows didn't install out of the box which was a bit useless.

Quote:
. Playing Streets of Rage 4 and Sonic Mania has extended my use of my laptop to include entertaining my children and allowing me to skip the overrated PS4 console generation.


So you will get a PS5 instead?

 Status: Offline
Profile     Report this post  
bison 
Re: Amiga hardware poll
Posted on 23-Jun-2020 18:35:58
#84 ]
Super Member
Joined: 18-Dec-2007
Posts: 1637
From: N-Space

@Hypex

Quote:
Also it's rather vague. What does Apple silicon actually mean? Have they designed their own CPU? Not once in the article did I see any mention of ARM. It's just implied. Neither did they directly say they were dropping Intel.

Yes, it's a really weird announcement. They did make a passing mention of the A12Z, which is an indirect reference to ARM.

Apple A12Z

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

 Status: Offline
Profile     Report this post  
Hypex 
Re: Amiga hardware poll
Posted on 23-Jun-2020 18:39:59
#85 ]
Elite Member
Joined: 6-May-2007
Posts: 9974
From: Greensborough, Australia

@Raffaele

Quote:
MorphOS announced porting to AMD64 architecture.


Good point. They actually have it easier than OS4. Sitting under MorphOS is the QBox and the ABox running on top of it. Quark, the bottom end microkernel, forms the QBox. They just need this running on AMD64 and they are away. Next step is porting the ABox which forms the Amiga core of MorphOS. That would be harder, since it is designed for 32-bit big endian, so don't know how they will tackle it. ABox also has some quirks as it uses some form of API that duplicates the AmigaOS 68k ABI. And so closer than OS4 is to an OS3 API. But I haven't gone into it deeper.

So, when all is said and done, MorphOS x64 may put pressure on OS4 and hardware. But, even now, MorphOS is closer to the OSX idea than OS4 is. It just retains the OS. What is presents to the user is an Amiga like OS in form and function. It has good ideas in there separating system components from the user.

 Status: Offline
Profile     Report this post  
matthey 
Re: Amiga hardware poll
Posted on 24-Jun-2020 8:28:59
#86 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

bison wrote:
Yes, it's a really weird announcement. They did make a passing mention of the A12Z, which is an indirect reference to ARM.

Apple A12Z


Apple's A12Z SoC is a minor upgrade to the A12X SoC (probably the same chip rated different) which went in the iPhone XS, iPhone XR and iPad Pro starting in 2018. The A12X SoC was supposedly the first consumer device to use the 7nm process for a chip. There are 8 custom ARM cores at around 2.5GHz arranged in an ARM big.LITTLE configuration with 4 performance cores and 4 low power cores. It looks like Apple is bringing the same SoC to the desktop which will improve economies of scale and consolidate development efforts and support. Using smaller die sizes than x86_64 CPUs allows Apple to remain competitive enough in performance while helping to improve energy efficiency in the iPhone and iPad where it is more important. It looks like x86_64 CPUs will still have better single core performance, higher clock ratings and more powerful GPUs which is important for games and some sequential workloads but tasks where multi-core processing can be used will probably be close enough considering the Apple SoC uses a fraction of the energy (advertise the green apples). Apple could expand into other consumer products (like a console/TV device) to further improve economies of scale.

Performance of the A12X SoC is quite respectable. Benchmarks and comparisons to some other custom ARM CPUs are in the following link.

https://www.anandtech.com/show/13392/the-iphone-xs-xs-max-review-unveiling-the-silicon-secrets/4

An article found that the weak cores in the A12X SoC were close to the performance of the Cortex-A73.

"What did surprise me a lot was seeing just how well Apple’s small cores compare to Arm’s Cortex-A73 under SPECint. Here Apple’s small cores almost match the performance of Arm’s high-performance cores from just 2 years ago. In SPEC's integer workloads, A12 Tempest is nearly equivalent to a 2.1GHz A73."

https://www.anandtech.com/show/13453/apple-iphone-xs-review-addendum-small-core-and-nn-performance

The Cortex-A73 is the successor to the Cortex-A72 which is used in the Raspberry Pi 4. Ironically, Cortex-A72 cores were often used as the performance core in ARM big.LITTLE setups where Cortex-A53 cores were used for the low power cores. Cortex-A53 cores were used in the Raspberry Pi 3. The same article estimated the powerful cores in the Apple A12X SoC were 3-4 times the performance of a Cortex-A73 which can be up to 30% better performance than a Cortex-A72 like used in the Raspberry Pi 4. Of course the Raspberry Pi 4 is anything but state of the art (SoC is 28nm die size) but this hasn't stopped them from selling 30 million units. Die shrinks make performance and energy efficiency look good.

I believe ARM Holdings designed top performance core is the ARM Cortex-A78. It would be interesting to see how it stacks up against Apple's A12 high performance cores. The earlier ARM Holdings designed Cortex-A76 is outperforming the highest performance RISC-V CPUs.

https://sifive.cdn.prismic.io/sifive/d1aafa10-9a98-46bb-8661-c4b6e241f416_SiFive+U8+Takes+RISC-V+Out+of+Order.pdf

The SiFive U84 RISC-V 10 stage pipeline allows higher clocks but now has a 10 cycle branch misprediction penalty (better branch prediction needed) and a 4 cycle load/use penalty meaning there will be 3 cycles after a load before an instruction can use the data or there will be up to 3 cycles of bubbles (more aggressive and expensive OoO needed). The higher performance ARM Cortex-A76 in the comparison is using micro-ops like x86_64 which requires more logic and power. It looks like RISC-V is trying to compete with reduced area and power instead of performance. The RISC-V ISA is a simple throwback to pure RISC ISAs avoiding some of the mistakes and adding some simple modern features. ARM AArch64 is a feature rich ISA with complexity rivaling CISC ISAs (more so than PPC). The 68k was criticized at the time by hardware designers for its complex addressing modes but AArch64 added many of them including a sign or unsigned extended and shifted index register and pre and post indexed addressing.

RISC-V addressing modes
lw x1,(x2) // Register Indirect
lw x1,4(x2) // Register Indirect with Immediate

PowerPC addressing modes
lw r1, (r2) // Register Indirect
lw r1,4(r2) // Register Indirect with Immediate Index
lwu r1,4(r2) // Register Indirect with Immediate Index (update base)
lw r1,(r2),r3 // Register Indirect with Index
lwu r1,(r2),r3 // Register Indirect with Index (update base)

AArch64 addressing modes
ldr W1, [X2] // base
ldr W1, [X2, #4] // base plus immediate
ldr W1, [X2], #4 // post indexed
ldr W1, [X2, #4]! // pre indexed
ldr W1, [X2, X3] // base plus register offset
ldr W1, [X2, X3, lsl #3] // base plus register scaled
ldr W1, [X2, W3, sxtw] // base plus register with sign extend
ldr W1, [X2, W3, sxtw #3] // base plus register with sign extend and scaled

68k addressing modes
move.l 4,a6 // Absolute
move.l (a0),d1 // Address Register Indirect
move.l (4,a0),d1 // Address Register Indirect with Displacement
move.l (a0)+,d1 // Address Register Indirect with Postincrement
move.l -(a0),d1 // Address Register Indirect with Predecrement
move.l (a0,d0.l),d1 // Address Register Indirect with Index
move.l (a0,d0.l*8),d1 // Address Register Indirect with Index (scale)
move.l (a0,d0.w),d1 // Address Register Indirect with Index (sign extend)
move.l (a0,d0.w*8),d1 // Address Register Indirect with Index (sign extend and scale)
move.l (4,a0,d0.w*8),d1 // Address Register Indirect with Index (disp, sign extend and scale)

x86_64 addressing modes
movl 4,%rax // Direct
movl (%rdx),%rax // Indirect
movl 4(%rdx),%rax // Base with Displacement
movl (%rdx,%rcx,8),%rax // Scaled index
movl 4(%rdx,%rcx,8),%rax // Scaled Index with Displacement

AArch64 has more addressing mode variations and more useful addressing modes than x86_64! AArch64 addressing modes more closely resemble 68k addressing modes in functionality! Do the addressing modes help AArch64 performance? I believe they do. They avoid several dependent instructions to calculate an address and reduce the number of instructions to some of the lowest I have seen, even a little lower than with 68k code. AArch64 still leaves significant room for improvement. The fixed length encoding means there are a limited number of bits for immediates and displacements. The number of bits varies from instruction to instruction even in a particular RISC ISA. Exceed the encoding field and multiple dependent instructions are needed or, in the case of branches, trampolines which are branches to branches. Variable length encodings can get rid of this problem by providing variable length immediates and displacements while providing superior code density. Of course a CISC instruction can do an operation at the same time as a memory access which reduces the number of instructions and gives better code density. Many GP registers and 3 op instructions are nice and help RISC ISAs gain back some of the performance lost but it is not enough to overcome their disadvantage when accessing memory.

Last edited by matthey on 24-Jun-2020 at 03:33 PM.

 Status: Offline
Profile     Report this post  
matthey 
Re: Amiga hardware poll
Posted on 24-Jun-2020 8:55:29
#87 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

Hypex wrote:
So, when all is said and done, MorphOS x64 may put pressure on OS4 and hardware. But, even now, MorphOS is closer to the OSX idea than OS4 is. It just retains the OS. What is presents to the user is an Amiga like OS in form and function. It has good ideas in there separating system components from the user.


MorphOS and AROS already have a head start on moving to x86_64. Does that add pressure for AmigaOS to follow or try something else?

x86_64
good performance, good performance/price

ARM
good performance/price, average performance, low power

POWER
good performance, poor performance/price

 Status: Offline
Profile     Report this post  
amigang 
Re: Amiga hardware poll
Posted on 24-Jun-2020 12:08:07
#88 ]
Super Member
Joined: 12-Jan-2005
Posts: 1501
From: Cheshire, England

I started to think maybe the Amiga needs to move away from hardware all together.

UAE has been ported to every single platform on the planet it would seem and it feel to me at least a massive under appreciated bit of software that emulate the Amiga nearly 100% accuracy. Windows version now even emulates nearly all the add on cards you can think of. Amikit in particular to me feel very close to AmigaOS4.

A Pi Zero that cost just £5 that can run most Amiga games fine. I personally think a little bit more focus should be making the emulator better and making maybe a custom AmigaOS & rom that a bit more aware its being emulated so better features like Alice / Rabbit hole (that part of Amikit, allows you to run OS native apps with in the emulator) are improved, maybe have a feature where WHD games / Programs are stored on the internet / cloud but look like there part of the OS would be a cool feature that I'm sure could be done. etc.

just a idea.

_________________
AmigaNG, YouTube, LeaveReality Studio

 Status: Offline
Profile     Report this post  
BigD 
Re: Amiga hardware poll
Posted on 24-Jun-2020 12:32:17
#89 ]
Elite Member
Joined: 11-Aug-2005
Posts: 5356
From: UK

@amigang

All the profit comes from the hardware. BeOS proved that there is no money in an alternative OS whatever hardware platform it is on. Windows and macOS no longer charge for updates so what is there left to sell other than hardware?

_________________
"Art challenges technology. Technology inspires the art."
John Lasseter, Co-Founder of Pixar Animation Studios

 Status: Offline
Profile     Report this post  
amigang 
Re: Amiga hardware poll
Posted on 24-Jun-2020 12:56:46
#90 ]
Super Member
Joined: 12-Jan-2005
Posts: 1501
From: Cheshire, England

@BigD

Quote:
Windows and macOS no longer charge for updates so what is there left to sell other than hardware?


exactly, Microsoft make it money, by services now, offering Office 365 monthly subscription, Cloud storage, they also get 5% cut on all items sold on its MS store, Apple i believe have 30% cut on all apps sold, they too now have subscription services.

Not that I would be thrilled or want to see this happen in the Amiga world, I personally hate how the market has moved to you no longer own the software you effectually rent it. But it how these company rake it in now a days.

There the Ant stream (trying to be the Netflix of Retro gaming) where you pay £10 a month to play a load of retro games over the net. they currently do a load of Amiga games on there, https://www.antstream.com/gameslist
maybe some kind of service like this could be done and expanded on just the Amiga platform.

And just look at Cloanto who have now effectively become Amiga, just by selling Amiga Emulator software!

_________________
AmigaNG, YouTube, LeaveReality Studio

 Status: Offline
Profile     Report this post  
BigD 
Re: Amiga hardware poll
Posted on 24-Jun-2020 14:01:25
#91 ]
Elite Member
Joined: 11-Aug-2005
Posts: 5356
From: UK

@amigang

Quote:
exactly, Microsoft make it money, by services now, offering Office 365 monthly subscription, Cloud storage, they also get 5% cut on all items sold on its MS store, Apple i believe have 30% cut on all apps sold, they too now have subscription services.


This model will lead to feature creep and bloatware. Companies no longer will have to justify the upgrade cycle they will just EXPECT you to pay up like car tax, TV license and home insurance.

_________________
"Art challenges technology. Technology inspires the art."
John Lasseter, Co-Founder of Pixar Animation Studios

 Status: Offline
Profile     Report this post  
Rose 
Re: Amiga hardware poll
Posted on 24-Jun-2020 14:14:31
#92 ]
Cult Member
Joined: 5-Nov-2009
Posts: 705
From: Unknown

@BigD

Quote:
All the profit comes from the hardware.


If there was money on OS4 hardware it would be manufactured bigger patches than 100 and you wouldn't need to wait for it for 6 months.

 Status: Offline
Profile     Report this post  
michalsc 
Re: Amiga hardware poll
Posted on 24-Jun-2020 15:12:53
#93 ]
AROS Core Developer
Joined: 14-Jun-2005
Posts: 265
From: Germany

@Raffaele

Quote:
Linux is good but mostly un-amigaish environment and it is not a solution.


Linux userspace is just a posix compilant. UNIX-ish and not amiga-like. But why not linux kernel alone? What is un-amigaish in the linux kernel?

Quote:
Our filesystems around BeOS/Haiku Kernel the same way that Apple did with Mach Kernel and MacOS environment, creating MacOS X.


The same can be done with linux kernel. Profit is, you have all the architectures you could imagine there, many of them very well supported.

Quote:
BeOS/Haiku is available both in PowerPC and IntelX86 versions.


Linux is available for PPC, PPC64, ARM, AArch64, x86, x86_64 to name just very few. It runs on old PPC Apple machines, it runs on Talos, on almost any ARM machine and on intel desktop PCs.

ALthough I agree in general with you here (new kernel, new OS inspired on AmigaOS but with modern features) I cannot really see advantage of BeOS/Haiku here.

 Status: Offline
Profile     Report this post  
bison 
Re: Amiga hardware poll
Posted on 24-Jun-2020 15:24:09
#94 ]
Super Member
Joined: 18-Dec-2007
Posts: 1637
From: N-Space

@matthey

The transition from the Pi3's Cortex-A53 to the Pi4's Cortex-A72 is bittersweet. The performance of the Cortex-A72 is a lot better, but it is also susceptible to out-of-order side-channel attacks, and there seems to be a new one of those every few months. By the time all these exploits are found and mitigated in software, I wonder what the actual performance differences will be.

Last edited by bison on 24-Jun-2020 at 07:31 PM.

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

 Status: Offline
Profile     Report this post  
bison 
Re: Amiga hardware poll
Posted on 24-Jun-2020 15:29:53
#95 ]
Super Member
Joined: 18-Dec-2007
Posts: 1637
From: N-Space

@amigang

Quote:
I started to think maybe the Amiga needs to move away from hardware all together.

UAE has been ported to every single platform on the planet it would seem and it feel to me at least a massive under appreciated bit of software that emulate the Amiga nearly 100% accuracy. Windows version now even emulates nearly all the add on cards you can think of. Amikit in particular to me feel very close to AmigaOS4.

A Pi Zero that cost just £5 that can run most Amiga games fine. I personally think a little bit more focus should be making the emulator better and making maybe a custom AmigaOS & rom that a bit more aware its being emulated so better features like Alice / Rabbit hole (that part of Amikit, allows you to run OS native apps with in the emulator) are improved, maybe have a feature where WHD games / Programs are stored on the internet / cloud but look like there part of the OS would be a cool feature that I'm sure could be done. etc.

Good points, all of them!

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

 Status: Offline
Profile     Report this post  
bison 
Re: Amiga hardware poll
Posted on 24-Jun-2020 15:41:18
#96 ]
Super Member
Joined: 18-Dec-2007
Posts: 1637
From: N-Space

@michalsc

I think the same. People generally don't know what they're talking about when they diss Linux. There seems to be this widespread conflation between Linux and bash, for example. They don't like bash, so they say "Linux sucks." For the record, I don't like bash much either, but at least I'm not confusing it with Linux.

A Linux-based system with a new shell and an Amiga-like Wayland-based UI (compositor and toolkit) could be really nice.

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

 Status: Offline
Profile     Report this post  
BigD 
Re: Amiga hardware poll
Posted on 24-Jun-2020 15:52:27
#97 ]
Elite Member
Joined: 11-Aug-2005
Posts: 5356
From: UK

@Hypex

Quote:
So you will get a PS5 instead?


That's the plan once the economy sorts itself out! At least it will be future proof playing current gen games without the jumbo jet engine style cooling fans, while also allowing the viewing of 4k Blu-Rays

_________________
"Art challenges technology. Technology inspires the art."
John Lasseter, Co-Founder of Pixar Animation Studios

 Status: Offline
Profile     Report this post  
BigD 
Re: Amiga hardware poll
Posted on 24-Jun-2020 15:55:27
#98 ]
Elite Member
Joined: 11-Aug-2005
Posts: 5356
From: UK

@Rose

AmigaKit is primarily a hardware supplier despite the AmiStore App Store! A-EON's reason for being is hardware. They use hardware sales to partially fund software development. This is a fact. I didn't say that there was mega-bucks to be made.

_________________
"Art challenges technology. Technology inspires the art."
John Lasseter, Co-Founder of Pixar Animation Studios

 Status: Offline
Profile     Report this post  
Hypex 
Re: Amiga hardware poll
Posted on 24-Jun-2020 17:11:21
#99 ]
Elite Member
Joined: 6-May-2007
Posts: 9974
From: Greensborough, Australia

@matthey

Quote:
MorphOS and AROS already have a head start on moving to x86_64. Does that add pressure for AmigaOS to follow or try something else?


It does somewhat. MorphOS is closer to OS4, being for the same purpose, and currently being on PPC.

AROS, despite being designed to be close to OS3, does lack the transparent OS3 emulation that OS4 has due to its nature. But work has been put into this lately to improve it. I haven't tried Icaros lately so don't know how it well works out of the box now.

Another thing to consider, when shifting CPU, is how ASM friendly it is. Now, these days, doing ASM isn't as common place as it once was. But you don't want to punish the AmigaOS4 hackers either. Not exactly those hacking the OS, but those who liked to get down and dirty with some ASM. 68K is the bomb and will remain so for a long time I think. PPC tends to have some middle ground I think; it's readable, writeable, and sensible enough that you can understand it. It's also been around long enough that we know it well enough by now. X86+, still looks like cryptic nonsense to me, but it won't be left in the past, and I wouldn't want to touch it except from a ten foot long C/++ compiler. ARM is still strange to me, but looks readable enough, so my vote would be for ARM looking the most friendly. If Motorola didn't dump 88K for PPC, we might have had a better choice now, but they stuffed it up.

 Status: Offline
Profile     Report this post  
matthey 
Re: Amiga hardware poll
Posted on 24-Jun-2020 21:13:11
#100 ]
Cult Member
Joined: 14-Mar-2007
Posts: 756
From: Kansas

Quote:

bison wrote:
The transition from the Pi3's Cortex-A53 to the Pi4's Cortex-A72 is bittersweet. The performance of the Cortex-A72 is a lot better, but it is also susceptible to out-of-order side-channel attacks, and there seems to be new one of those every few months. By the time all these exploits are found and mitigated in software, I wonder what the actual performance differences will be.


Most OoO core designs were cheating instead of paying the full tax. Unused data in caches and memory while speculating needs to be destroyed or isolated caches and memory used. This significantly increases the cost of deeply speculating OoO execution. Changing between user, supervisor and hypervisor states and cache prefetch and flush type instructions become more expensive, especially if they are in user space like x86_64. More speculation results in more complexity which has more potential for security problems. OoO execution requires retaining a lot of internal state data to backtrack in the case of interrupts and it has to do it correctly, which was not the case for Meltdown vulnerable Intel CPUs. Spectre was often mitigated by greatly reducing speculation in older CPUs which had a significant impact on performance. Many newer x86_64 CPUs allow microcode/firmware updates but this feature adds cost and has the potential to be a security vulnerability itself.

RISC processors need OoO for performance. RISC ISAs typically require more instructions to be executed and they are mostly dependent instructions to load large immediates and calculate addressing modes. Dependent instructions means that parallel superscalar execution isn't possible so other instructions need to be placed (scheduled) in between. A separate load/store unit means loads usually have a load/use penalty before another unit like an integer ALU unit can use the result. This requires scheduling instructions between the load and op instruction. Now we need to unroll loops if possible and most of the time there is only one load/store unit and single ported data cache so we can't have 2 consecutive load instructions. Superscalar scheduling becomes difficult and we use many extra instructions and registers trying to avoid bubbles. OoO has to do more work and speculation trying to eliminate the bubbles.

I believe there is a demand for simpler cores with limited speculation which are more secure and predictable (low latency and jitter) yet offer best in class performance. Best in class code density is also a good selling point to allow smaller and cheaper embedded devices. The in order 68060 was outperforming most cores of its day in PPA (power, performance and area) including several OoO cores. The Apollo Core also shows good 68k performance. A chip like Apple's A12 SoC probably cost more to develop and produce than the GDP of several small countries but a simple SoC on older die sizes can likely be developed and produced for less than 10 million U.S. dollars.

I like the Semiconductor Engineering description of PPA which follows.

Quote:

PPA (Power, Performance, Area) - Fundamental tradeoffs made in semiconductor design for power, performance and area.

DESCRIPTION

PPA stands for power, performance and area, and historically these have been the three variables used in deciding how to optimize semiconductor designs.

Until 65nm, cost, which is a function of area, and performance, were the most-cited criteria for a successful design. But as battery-driven mobile devices have replaced PCs as the volume platform for SoCs, power has moved up to the top of the list, followed by cost and performance. Performance is still critical in certain markets, such as servers, but even there more accountability for server power and cooling has made it imperative to reduce power.

With the Internet of Things market, companies are beginning to think of a fourth element to go with PPA, namely security.


Yep, that security thing is becoming more important.

Last edited by matthey on 24-Jun-2020 at 09:22 PM.

 Status: Offline
Profile     Report this post  
Goto page ( Previous Page 1 | 2 | 3 | 4 | 5 | 6 | 7 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