Comments by "Valen Ron" (@valenrn8657) on "The Computer Chronicles - The New Amigas (1988)" video.

  1. Amiga wasn't ahead in packed pixels.
    2
  2. PC's MS-DOS sucked and it was dumped.
    2
  3. During the early 1980s, $240 CP/M-86 vs $40 PC-DOS... Freebie Linux destroyed commercial Unix. Gary Kildall's greed killed his CP/M-86 business model. Try again.
    2
  4. Commodore didn't fund a 16-bit ALU MOS 65xx CPU successor in a timely manner, hence why companies like Acorn created the RISC-based ARM CPU family. 🤣 Before the Amiga, Commodore's 16-bit solution was C900 with 16-bit Zilog Z8000 CPU and MOS Technology 8563 which is the same VDC chipset for C128's 80-column (640 × 200 pixels) mode. Zilog Z80 was a fork from Intel 8080 CPU. Intel 80286 evolved from 8086 which evolved from 8085 which then evolved from 8080.
    1
  5. Video Toaster's 24-bit color frame buffer doesn't benefit Amiga's general-purpose graphics.
    1
  6.  @madigorfkgoogle9349  By the end of 1994, Intel had reached 6 to 7 million Pentium ship goals. PC Master Race minority beats the entire Amiga install base within two years.
    1
  7.  @johannesdolch  Tramiel's Atari was the parallel failure as Commodore. Atari Falcon's 68030 has a 16-bit bus and that's Tramiel's cost-cutting mindset. LOL.
    1
  8. ​ @krokeman  For playing action 2D games, IBM PS/2 Model 55SX (386SX @ 16 Mhz and IBM VGA) will lose against the Amiga 500. PiStorm-equipped (ARM Cortex A53 @ 1.4 Ghz JIT'ing emulating a fast 68040 CPU) Amiga 500's custom chipset can drive Quake in HAM mode. https://www.youtube.com/watch?v=btoU_CQSg7A will beat K7 Athlon XP 2200+ (1800 Mhz) with IBM VGA playing Quake in a slide show. https://www.youtube.com/watch?v=octArwHpaiY& 320x200 - 8,6FPS IBM VGA is garbage even with a fast K7 Athlon XP 2200+ (1800 Mhz) CPU. PC clone with 386DX with 1989 era ET4000AX beats the Amiga 500. For playing Doom, PC clone with 386DX-40 with 1989 era ET4000AX is roughly equivalent to Amiga 1200 with 68030 @ 50Mhz.
    1
  9.  @madigorfkgoogle9349  Falcon's 16-bit bus is a fact and there's no need for Amiga fanboy. The Amiga 1200 had a 32-bit Fast RAM controller inside the Budgie chip and no memory chips. The same cost-cutting mindset. Commodore UK MD argued for uprated specs for baseline AGA Amigas with minimal cost difference and it was rejected by Commodore International CEO.
    1
  10.  @madigorfkgoogle9349  With Fast RAM, the Amiga 1200 can operate two 32-bit buses i.e. CPU's 32-bit Fast RAM and AGA's 32-bit Chip RAM. AGA chipset is DMA capable and can operate independently from the CPU.
    1
  11.  @johannesdolch  EGS was released in 1992 and it was Amiga's early RTG solution.
    1
  12.  @krokeman  VGA has hardware scrolling and it was simplistic.
    1
  13. Amiga's 1988-era Kickstart 1.3 / Workbench 1.3 (AmigaOS 1.3) already supports faster 68K CPUs such as 68020 and 68030.
    1
  14.  @thealaskan1635  Windows 95 has two button mouse GUI system which hence why Apple failed with their GUI legal battle against Microsoft.
    1
  15.  @thealaskan1635  The lead producer for Windows 1.x/2.x/3.x is Gabe Newell who later founded Valve Inc. 🤣😂 With Valve's SteamOS, Valve is promoting a Windows DirectX ecosystem clone called "Proton" over native Linux GNU programming API. 🤣😂
    1
  16.  @Blatstein  It depends on CP/M's wholesale asking price vs. Microsoft's DOS.
    1
  17.  @madigorfkgoogle9349  "Most customers were repelled by the significantly greater price IBM charged for CP/M-86 over PC DOS (US$240 and US$40, respectively)."
    1
  18.  @madigorfkgoogle9349  CP/M was the original OS for Intel 8080 microprocessors. It was sold generally for around $400 a copy.
    1
  19. The PC has legacy business software support and Xenix X86 i.e. Microsoft's licensed AT&T Unix. 80286 has integrated MMU which is important for Xenix 286. Microsoft Xenix was the best-selling Unix since it run on commodity X86 PC hardware. IBM OS/2 was a distraction for Microsoft. Windows NT X86 and Linux X86 have displaced Xenix. Motorola wasn't serious about Unix until MMU integrated 68030. 68020 has an external MMU which is an additional component cost. By the time of 1993, Windows NT was available on a super fast Alpha CPU that crushed both 486 / Pentium 66 and 68040/released 1994 68060 which caused Intel to rapidly evolved Pentium CPU and CISC-RISC hybrid Pentium Pro (P6). The foundation for Amiga's defeat was IBM establishing VGA and 8514/A standards in 1987. ET4000AX was released in 1989 with fast VGA and 8514/A "compatible". VGA cloners improved and cost-reduced VGA / SVGA and 8514/A. Motorola's EoL 68K's R&D doomed Commodore, Atari, and any other vendors with heavy 68K-based computer business. Motorola wasted R&D time on RISC based 88000. Motorola 68882 wasn't a single-cycle floating point co-processor like AT&T's DSP3210. AmigaOS's multitasking lacks security and multiuser features since it's a toy OS when compared to Xenix.
    1
  20. Amiga Inc used the Atari loan for less than a month and Commodore paid the Atari loan.
    1
  21. ​ @madigorfkgoogle9349  Unix that runs on 68000/68010 has custom MMUs.
    1
  22.  @madigorfkgoogle9349  Where's 68882's 25 MFLOPS at 50 Mhz? 68882 @ 50Mhz doesn't demonstrate beating 68040's FPU at @ 25Mhz!
    1
  23.  @madigorfkgoogle9349  DSP3210 is a RISC core consisting of a 32-bit Arithmetic Logic Unit (ALU), a 32-bit Program Counter (PC), 22 32-bit general purpose registers (r0-r22), and a 32-bit barrel shifter. DSP3210's DAU consists of a 32-bit floating point multiplier, a 40-bit floating point adder, four 40-bit floating point accumulators (a0-a3) The DAU supports the following floating point formats: Single precision (32-bit) in both DSP32 and IEEE format. Extended single precision (40-bit) (uses 8 mantissa guard bits).
    1
  24.  @madigorfkgoogle9349  In the mid-to-late 1980s, Xenix was the most common Unix variant, measured according to the number of machines on which it was installed.[1][2] Microsoft chairman Bill Gates said at Unix Expo in 1996 that, for a long time, Microsoft had the highest-volume AT&T Unix license Intel 286 and 386 guaranteed standard MMU.
    1
  25.  @madigorfkgoogle9349  https://www.youtube.com/watch?v=2OoPYBlc-2A For Mandelbrot render program, AA3000+'s DSP3210 vs 68060 @ 50 Mhz DSP3210 is holding its own against 68060.
    1
  26.  @madigorfkgoogle9349  Note that DSP3210 supports single precision IEEE 32-bit floating point (IEEE FP32).
    1
  27.  @madigorfkgoogle9349  DSP3210 supports IEEE FP32 which is 32 bit floating point.
    1
  28.  @madigorfkgoogle9349  Fact: Memory protection requires PMMU.
    1
  29.  @madigorfkgoogle9349  https://www.youtube.com/watch?v=2OoPYBlc-2A For this Mandelbrot render program, DSP3210 is almost holding its own against 68060.
    1
  30.  @madigorfkgoogle9349 68882 is rubbish. I owned both 68030/68882 and 68060.
    1
  31.  @madigorfkgoogle9349  Don't confuse "unix without memory protection" vs Unix with memory protection.
    1
  32.  @madigorfkgoogle9349  >except that vast majority of workstations using UNIX ran on Motorola, not intel in early days, before switching to MIPS or SPARC and ALPHA. In the mid-to-late 1980s, Xenix was the most common Unix variant, measured according to the number of machines on which it was installed.[1][2] Microsoft chairman Bill Gates said at Unix Expo in 1996 that, for a long time, Microsoft had the highest-volume AT&T Unix license.
    1
  33. ​ @madigorfkgoogle9349  For Xenix, 286 doesn't require another 286 since the 286 has a built-in MMU. Two 68000 CPUs require two 16-bit front-side buses on the PCB. DSP3210 supports 32-bit integer and IEEE FP32. DSP3210 is designed as an RISC core. Modern GpGPUs and X86's 3DNow/SSE SIMD support IEEE FP32. Motorola 56000 is a 24-bit integer fixed-point DSP. 68882 is a joke for high-performance 3D gaming.
    1
  34.  @madigorfkgoogle9349  Run Quake benchmarks test between 68030/68882 @ 50 Mhz vs 68040 @ 40 Mhz. You will find that FPU-enabled Quake runs faster on a 68040 CPU.
    1
  35.  @madigorfkgoogle9349  The MC68882 was a single cycle FPU capable of calculation 1-67 bits in one cycle. For Lightwave 3D, are you claiming 68030/68882 @ 50 Mhz is faster than 68040 @ 25 Mhz? "1 IPC" is for most instructions for the given processor.
    1
  36.  @madigorfkgoogle9349  I doubt that MC68040 can reach 25 MFLOPS. Are you claiming 68040's FPU beats 68060's non-pipelined FPU? That's absurd. 68040 focuses on integer performance with the floating point coming a distant second and 68040 @ 25 Mhz FPU is still faster than 68882 @ 50Mhz.
    1
  37.  @madigorfkgoogle9349  Memory is managed by the second MC68000 that is serving as paging MMU 1. That's for virtual memory. 2. Two 68000 CPUs cause 32-bit bus PCB expenses.
    1
  38.  @madigorfkgoogle9349  PMMU solves two problems i.e. virtual memory and memory protection. Two 68000 configurations were replaced by external MMUs for 68010 and 68020.
    1
  39.  @madigorfkgoogle9349 Memory is managed by the second MC68000 that is serving as paging MMU, you still do not understand what you read. Apollo machines used a proprietary operating system known as Aegis and a dual 68000 CPU configuration. This is not AT&T's UNIX. The dual 68000 CPU's page fault workaround was eliminated by 68010.
    1
  40.  @madigorfkgoogle9349  this is nonsense.... I did have 68030/68882 and 68060 systems. You're shit.
    1
  41.  @madigorfkgoogle9349  After owning both FULL 32-bit 68030/68882 and 68060 systems, your argument is absurd. A 16-bit bus equipped 68030 is half-baked.
    1
  42.  @madigorfkgoogle9349  You really didn live the UNIX workstation era. You're the one confusing the requirements for virtual memory and memory protection.
    1
  43.  @madigorfkgoogle9349  Your DSP3210 argument is wrong.
    1
  44.  @madigorfkgoogle9349  You implied Apollo Computers' dual 68000 system and their proprietary operating system, Aegis as Unix which is NOT AT&T's licensed UNIX. The 1st Xenix is licensed UNIX from AT&T Corporation's UNIX version 7 source.
    1
  45.  @madigorfkgoogle9349  While the DSP3210 can work in single FP precision (24bit x 8bit). DSP3210 supports IEEE single precision 32-bit floating point and 32-bit fixed point integer. Commodore engineers selected DSP3210 to speed up 3D rendering since 68040 FPU and 68030/68882 FPU configurations weren't enough. DSP3210 is designed as "a RISC core" with most of its floating point instructions in the 0.5 IPC i.e. 66 Mhz yields 33 MFLOPS or 50 Mhz with 25 MFLOPS. Here is a text chart of some common FPU instructions and timings in cycles for the 68882, 68040 and 68060 in this order: FMove FPn,FPn 21 2 1 FMove.D ,FPn 40 3 1 FMove.D FPn, 44 3 1 FAdd FPn,FPn 21 3 3 FSub FPn,FPn 21 3 3 FMul FPn,FPn 76 5 3 FDiv FPn,FPn 108 38 37 FSqrt FPn,FPn 110 103 68 FAdd.D ,FPn 75 3 3 FSub.D ,FPn 75 3 3 FMul.D ,FPn 95 5 3 FDiv.D ,FPn 127 38 37 FSqrt.D ,FPn 129 103 68 The 68882 @ 50Mhz is a dog compared to the 68060 FPU @ 50 Mhz and 68040 FPU @ 40Mhz
    1
  46. Amiga's weakness is the 68K CPU's R&D road map i.e. Motorola terminated 68K as a high-performance CPU. The Amiga is a desktop computer not an embedded computer. 68040 and 68060 were late when compared to 486 and Pentium releases. The Amiga is leading PiStorm-Emu68 development to break free from Motorola/Freescale's influence. Amiga's future direction is with ARM CPUs.
    1
  47.  @madigorfkgoogle9349  The MC68882 was a single cycle FPU capable of calculation 1-67 bits in one cycle For context, MC68881 has a high-speed 67-bit barrel shifter. FADD and FMUL singles aren't good. 6888x is not RISC!
    1
  48.  @madigorfkgoogle9349  your argument about my argument is double wrong. You're wrong when I have instructions clock cycle document for 68882. 68882 is NOT a RISC design FPU!
    1
  49.  @madigorfkgoogle9349  Read Page 321 of 409 of MC68881/MC68882 User Manual from Freescale. RTFM This page shows clock cycles for 68882's floating point instructions. 68882's FP32 instruction clock cycle time examples FADD = H30, T38, Total 69 FMUL = H30, T58, Total 89 This is NOT a RISC-designed FPU! MC68882 is dog slow.
    1
  50.  @madigorfkgoogle9349  You do not understand basic principles and are very confused from things you have red. Fact: Linux was born on i386, enough said.
    1
  51.  @madigorfkgoogle9349  it is crippled by same way as Falcon I did state that both machines have it's strong cost-cutting measures.
    1
  52.  @madigorfkgoogle9349 You do not understand basic principles and are very confused from things you have red. Dual 68000's page fault method is just virtual memory. You're confused. Don't confuse memory protection requirements with virtual memory. There's a reason why external MC68851 and other custom MMUs exist in the first place! Intergrated MMU is guaranteed on every 286 and 386. MMU is a "nickel and dime" product segmentation on the 68K product stack.
    1
  53.  @madigorfkgoogle9349  It has pipelining and can execute multiple instruction concurrently. Meanless against RISC-designed FPU!
    1
  54.  @madigorfkgoogle9349  And no, actually Linux was born (or was carried) on MC68008 FALSE. When Linux announced to the world Linus Torvalds: Yes - it's free of any minix code, and it has a multi-threaded fs. It is NOT portable (uses 386 task switching etc), and it probably never will support anything other than AT-harddisks, as that's all I have.
    1
  55.  @madigorfkgoogle9349  You are the one who is here wrong, you always change topic, I NEVER stated the 68882 is a RISC design FPU. You argued, "The MC68882 was a single cycle FPU capable of calculation 1-67 bits in one cycle". I dig up the old 68881/6882 User Manual to debunk your argument.
    1
  56.  @madigorfkgoogle9349  You are the one who is here wrong, you always change topic, I NEVER stated the 68882 is a RISC design FPU. MFLOPS matters for basic comparison. You're wrong. Amiga's SysInfo's MFLOPS rating is better than your wrong argument. MacOS runs on the Amiga via Shapeshifter para-VM and MacOS's Byte Mark benchmark also runs.
    1
  57.  @madigorfkgoogle9349  There is a reason why MFOPLS is not used since 90s. FALSE. SGEMM benchmark still uses the FLOPS rating e.g. MFLOPS, GFLOPS, and TFLOPS. AIDA64 GGPU benchmark still uses the FLOPS rating. Or do you believe that A1200 when equipped with FastRAM has 2.3 power of stock A1200? You haven't factored that the shared memory bus will degrade the A1200's CPU when compared to PC style two discrete memory pools. SNES has at least two discrete memory pools i.e. WRAM and VRAM. 3DO has two discrete memory pools i.e. system RAM and VRAM. PS1 has two discrete memory pools i.e. system RAM and VRAM.
    1
  58.  @madigorfkgoogle9349  paging MMU is NOT just for virtual memory The dual 68000 setup is for virtual memory. Try again. I did state that PMMU is needed for virtual memory and memory protection.
    1
  59.  @madigorfkgoogle9349  Reminder, 286 was released in 1982. In late 1984, IBM XENIX 1.0 for the PC/AT was released. IBM's $5,000 AT 286 made for the cheapest Unix workstation around, which became an immediate hit in university computing labs worldwide. Furthermore, because XENIX allowed inexpensive dumb terminals to connect to a central machine, it was perfect for any setting where users needed to share the same centralized resource. XENIX became popular with retailers, fast-food outlets, and for scheduling systems used in hotels and restaurants. Internally, Microsoft used XENIX for handling email, and it was used by everyone in the company, right up to Bill Gates. By the late 1980s, Microsoft had become the biggest Unix company, with XENIX having the largest number of installations of any Unix variant.
    1
  60.  @madigorfkgoogle9349  you see, and that is again a proof you dont understand what you read. That's a misapplication of CISC vs RISC. A RISC FPU wins math extensive FADD and FMUL floating point basic operations at near 1 IPC. Classic CISC like 68030 is mostly microcode with instructions being executed in many clock cycles e.g. run Doom with 68030 @ 50 Mhz vs 3DO"s ARM60 @ 12.5 Mhz. Doom on 3DO is no hardware accelerated. 68882's FMUL and FADD are in many clock cycles. Don't apply +1 IPC CISC capable 68060 / Pentium to classic CISC like 68030. On Lightwave rendering, MIPS R4400 murders MC68882 and 68040 FPUs
    1
  61.  @madigorfkgoogle9349  argued to you that MC68882 is a different animal to DSP3210 since DSP has just single precision, while MC68882 supports double and extended precision, so the DSP3210 while fast IEEE FP32 is in common with today's gaming GpGPUs which executes fast IEEE FP32 with very slow IEEE FP64 compatibility mode. AMD's 3DNow 64bit SIMD aimed at two IEEE FP32 math. For the FP64 use case, you need NVIDIA H100 GpGPU which includes fast IEEE FP64. 128-core AMD Epyc CPU for extended double precision FP80 use case. IBM POWER 9 for quad precision FP128 use case. NVIDIA's AD102 has fast IEEE FP32 with slow IEEE FP64. I don't need to be an Amiga fan to know modern-day gaming PC GPGPU is mostly IEEE FP32 monsters.
    1
  62.  @madigorfkgoogle9349  But you simply wish the Amiga or even potential Amiga hardware to win Note that the Macintosh Quadra 840av used AT&T DSP3210 as a fast co-processor for 68040. After the 68040/DSP3210 combo, Apple switched to PowerPC.
    1
  63.  @madigorfkgoogle9349  so you grab any straw without understanding the principles and the big picture. With ARM Cortex A72, Emu68 can translate 68K instructions into ARM instructions in a near 1:1 manner e.g. MD5 workload. CISC has an advantage in doing math on memory location which implies loading and executing math operations when CISC CPU can execute 1 IPC. The CISC advantage falls apart when the classic CISC CPU is heavily micro-coded like 68030 or 80386. You grab any straw without understanding the principles and the big picture.
    1
  64.  @madigorfkgoogle9349  MC68882 is compliant of IEEE FP32 MC68882's FP32 is dog SLOW. Your "you just need to stop be stubborn and start to listen/read what I write." can get the GTFO.
    1
  65.  @madigorfkgoogle9349  What you don't get is guaranteed MMU availability integrated inside a CPU to lower PCB system design costs. Intel integrated MMU in 1982 and available for ALL 286-based systems and the same for 386's 1985 release. A dual 68000 configuration needs two 16-bit front-side buses. Motorola will "nickel and dime" MMU as a separate feature in their product stack. The dual 68000 configuration was rendered obsolete since it was not cost-effective and Unix evolved with memory protection.
    1
  66.  @madigorfkgoogle9349  Apple had a 68040/DSP32 combo before PowerPC switch.
    1
  67.  @madigorfkgoogle9349  68040's FPU beats 68882 , let alone 68060's FPU. $20 to $30 DSP32 is a poor man's 68060 FP32.
    1
  68.  @madigorfkgoogle9349  For displaying the CAD's viewport on a modern professional Quadro-based TU102: AIDA GpGPU benchmark from Quadro RTX 6000. FP32: 16,999 GFLOPS FP64: 529.5 GFLOPS CPU's FPU handles 3D coordinates while GPU handles view port rendering which is mostly FP32. AMD 3DNow, Intel SSE1, and PowerPC Altivec focused on IEEE FP32.
    1
  69.  @madigorfkgoogle9349  Note that 68882 is running ROMed microcode FPU instructions on a simplistic FPU core, hence why it takes many clock cycles to complete FADD or FMUL instructions. The RISC idealogy is to remove the microcode and implement the instruction directly on the hardware, hence the near 1 IPC goals. 68040 FPU implements more FPU hardware with 68882's instructions, hence there's less clock cycle to complete the same instruction with the 68040 FPU. It's even better on 68060 and AC68080.
    1
  70.  @madigorfkgoogle9349  The instruction set is nearly meaningless when microarchitecture is considered. 68060 FPU implements 68882 instructions towards the hardware side and less on microcode. As the transistor budget increased for the same money, Motorola was able to reduce microcode steps with more FPU hardware, hence the 68060's FPU executes the same 68882 instructions with fewer clock cycles. Apple 68040/DSP32 combo mirrors Apple's use case push with Altivec's fast IEEE FP32. Commodore just copied Apple's 68K/DSP32 combo.
    1
  71.  @madigorfkgoogle9349  Apple > Atari You need to grow up.
    1
  72.  @madigorfkgoogle9349  DSP3210 at 50 Mhz and 68060 at 50 Mhz was benchmarked running the same FPU extensive app. I can see the reason for Apple's 68040/DSP3210 bundle. You're wrong.
    1
  73.  @madigorfkgoogle9349  umping to another topic that is waaaay off topic that we discussed previousl You can't handle the truth.
    1
  74.  @madigorfkgoogle9349  gain this is UNUSABLE for calculations in professional area. NVIDIA Quadro card's fast FP32 is fine for professional use cases e.g. CAD 3D accelerator. You didn't attend GDC and NV lectures on fast FP32 emulating FP64.
    1
  75.  @madigorfkgoogle9349  Im right and you have no clue... Amiga gamer Apple bundled 68040/DSP3210, idiot.
    1
  76.  @madigorfkgoogle9349  I used my A3000/030 with 68882 @ 25 Mhz for raytracing and XCAD 3000, idiot. The 32-bit bus is needed for sustained FP32 and FP64 (two clock cycles for Fast RAM to CPU transfer). Peniutm has a 64-bit bus for sustained FP64. Raytracing is a search engine with geometry test problems. This is not factoring in color pixel shading.
    1
  77.  @madigorfkgoogle9349  The Amiga can run MacOS and EmuTOS without CPU emulation.
    1
  78.  @madigorfkgoogle9349  A3000 had the Video Toaster and Lightwave 3D. Lightwave 3D is a bottleless pit with FPU usage. I move into Windows Lightwave 3D and Cinema 4D for Pentium FPU power. 68060 FPU is not enough.
    1
  79.  @madigorfkgoogle9349  non of those benchmarks are cross platform FALSE. SGEMM benchmark is cross-platform i.e. it runs on various CPU and GpGPUs.
    1
  80.  @madigorfkgoogle9349  Some new off topic derailment?) It's a response to your Atari Falcon's use case. Look in the mirror.
    1
  81.  @madigorfkgoogle9349  I knew you will find some new topic to derail this thread. So, it's okay for you to state your Atari Falcon's use case about the audio process and "professional" apps while restricting the Amiga's side. You're a f_cking hypocrite.
    1
  82. ​ @madigorfkgoogle9349  In professional CAD/CAE the double precision is a requirement. Tell that to a very fast FP32-based Quadro RTX card that reused gaming GpGPU core. Modern AutoCAD's presentation layer is 3D accelerated by the same Quadro RTX card or Studio driver GeForce RTX 3090/4090 cards. X86-64 CPU's FP64/FP80 capable AVX2/AVX-512/X87 FPU handles the rest. RTX 3090/4090 supports workstation ECC GDDR6X and they are enabled in the control panel. Apple bundled 68040/DSP321 that includes improved FP64 (via 68040 FPU) and FP32 (via 68040 FPU and DSP3210) over the 68030/68882. The same approach is applied to the Apple-driven PowerPC Altivec and PowerPC FPU use case. For FP32, Altivec is 4X faster than the normal PowerPC FPU. This is the standard approach that is applied to most of today's professional workstations. NVIDIA H100 and AMD Instinct accelerators are the fast FP64 math accelerators besides being fast at Tensor AI workloads. The last gaming GPU product with reasonably fast FP64 relative to FP32's performance is Radeon HD 7970/R9-280X's Hawaii GCN and RX Vega 56/64/Vega II. After this generation, AMD split the fast FP64 GPUs from gaming/Radeon Pro NAVI GPUs. Commodore copied Apple's Quadra AV.
    1
  83.  @madigorfkgoogle9349  NVIDIA used TFLOPS via SGEMM on the Tesla K10 You don't know shit.
    1
  84.  @madigorfkgoogle9349  You talked about Falcon's DTP
    1
  85.  @madigorfkgoogle9349  So you found out that 68040 that has a built in derivative of 68882 FPU can beat the 68882, or the 68060 that has more the decade newer architecture can beat 68882, Jesus.. I'm correct on 68882's floating point IPC is dog slow and you're wrong.
    1
  86.  @madigorfkgoogle9349  again this is UNUSABLE for calculations in professional area. My NVIDIA's Quadro card that reused gaming GPU with mostly FP32 math acceleration debunks your sorry ass. Apple is correct on the 68040/DSP3210 bundle concept i.e. FP32 is many times faster than FP64.
    1
  87.  @madigorfkgoogle9349  You don't know shit.
    1
  88.  @madigorfkgoogle9349  Your attempts to make 68882 to be 1 IPC is misinformation.
    1
  89.  @madigorfkgoogle9349  SGEMM is running on OpenCL platform Not true for Intel X86. Intel Math Kernel Library (Intel MKL) 2019 has its own Just-In-Time (JIT) code generation for the SGEMM and DGEMM functions on Intel Advanced Vector Extensions 2 (Intel AVX2) and Intel Advanced Vector Extensions 512 (Intel AVX-512) architectures. Sony used GFLOPS FP32 for marketing PS3. MS used TFLOPS for the Xbox Series X's marketing. NVIDIA still uses SGEMM TFLOPS marketing for its GPUs. SGEMM for NVIDIA's GPUs are optimized for CUDA. AIDA64 FLOPS benchmark measures the MAD (Multiply-Addition) performance of the GPU. Puget Systems has the Linpack FLOPS benchmark. The FLOPS rating is still used. You're wrong. You're out of touch. Try again.
    1
  90.  @madigorfkgoogle9349  When compared to INT32, INT24 is a not common datatype for fixed point 3D. Fixed point 24-bit for 3D is the stupidest thing and it only comes with Atari.
    1
  91.  @madigorfkgoogle9349  another stupid comment form a Amiga fanboy, You can't handle the truth when other platforms beyond Atari or Commodore implemented fixed point INT32 3D e.g. Playstation 1.
    1
  92.  @madigorfkgoogle9349  except I never give away such a information You tried with "The MC68882 was a single cycle FPU capable of calculation 1-67 bits in one cycle." You're wrong with the actual floating point math operation for MC68882.
    1
  93.  @madigorfkgoogle9349  Do you think the DSP3210 is 1 IPC? RISC's 1 IPC is a goal. Closer to 1 IPC or greater is better.
    1
  94.  @madigorfkgoogle9349  AT&T marketed DSP3210 as being a RISC-designed core i.e. hardware instruction implementation (less microcode), fixed instruction length and 'etc'. DSP3210 has a dual pipeline with IEU and FPU (known as DPU). DSP3210 @ 50 Mhz on 68020/68030's 25 Mhz 32-bit bus wouldn't be able to execute 1 IPC. My point: DSP3210's FP32 operations will saturate the 25 Mhz 32-bit bus. DSP3210 has a combined 8 KB L1 cache like 68040's 4KB inst + 4 KB data L1 cache or 8 KB combined L1 cache. Quadra 840AV has 68040 @ 40 Mhz with a front-side bus matching it. Apple connected a custom VLSI chip to DSP3210 @ 60 Mhz The DSP3210 is a class above MC56000 DSP, 68030 @ 50 Mhz, 68882 @ 50 Mhz (just for FP32), and 68040 FPU at 40 Mhz (just for FP32). The RISC threat is real against heavily microcode CPUs like 68020/68030 or 80386DX. DSP3210 is officially supported in SunOS 4.1(Sun-4 Series), MacOS (Macintosh II Series), MS-DOS (PC6300 and compatibles), and UNIX Berkeley 4.2 (Sun-3 Series).
    1
  95.  @madigorfkgoogle9349  For Doom, ARM60 @ 12.5 Mhz (3DO, unacccelerated Doom) rivals 68030 @ 50 Mhz. RISC is a real threat for heavy microcode CPUs like 68030 and 80386. DSP56000's 24-bit IEU is missing 8 bits to complete its FP32 ability. Note why modern CISC CPUs have RISCy cores.
    1
  96.  @madigorfkgoogle9349  comparing a gaming integer math to scientific floating point operations DSP56000 has process 24-bit integers. Look in the mirror.
    1
  97. Atari ST's graphics have 16 colors with a 512-color palette. No hardware scroll. PC EGA has 16 colors with a 64 color palette. hardware scroll is available. 640 × 350 × 16 colors. PC had VGA in 1987. VGA clone ET3000 was released in late 1987. VGA has a hardware scroll.
    1
  98.  @madigorfkgoogle9349 The Atari Mega ST is useless for gaming since it's not a large system seller.
    1
  99.  @madigorfkgoogle9349  The entire Atari ST install base is just about 2.1 million. Mega ST's custom blitter coprocessor wasn't included in all models.
    1
  100.  @madigorfkgoogle9349  https://www.youtube.com/watch?v=5o9yOBBWPgM PC 286 @ 16 Mhz with clone VGA matched the Amiga. PC's Prehistorik 2 has a parallax background.
    1
  101.  @madigorfkgoogle9349  One problem, baseline Falcon 030 doesn't have an FPU.
    1
  102.  @madigorfkgoogle9349  A 286 16 Mhz PC with a clone VGA's Prehistorik 2 platform game's performance rivals the Amiga's efforts. Prehistorik 2 has a parallax background. 286-16 + VGA PC's Body Blows rivals A500 version. 286-16 + VGA PC's Gods rivals A500 version. 286-16 + VGA PC's Pinball Fantasies exceeds A500 version with 256 color presentation. Search for "PC 286-16 VGA SB retro gaming machine".
    1
  103.  @madigorfkgoogle9349  https://www.youtube.com/watch?v=9vdsRD3oI3c 8086 PC with VGA delivers console-like 2D games using "mode X" (320x240 60 Hz).
    1
  104.  @madigorfkgoogle9349  From "PC 286-16 VGA SB retro gaming machine" PC game examples that rival or exceed Amiga 500's efforts Body Blows Gods Golden Axe Pinball Fantasies Prehistorik 2 Prehistorik 2 has parallax backgrounds.
    1
  105.  @madigorfkgoogle9349  On Youtube, Little Engine MS-DOS VGA. 8086 PC with clone VGA demonstrated console-like 2D games using "mode X" (320x240 60 Hz).
    1
  106.  @madigorfkgoogle9349  Again, the Atari ST has no hardware scroll. A dead end. https://www.youtube.com/watch?v=bmHsEHy3XwM Smooth Scrolling on an 8086 with a VGA upgrade with Double Dragon 3.
    1
  107. 68K's failure is Motorola's. Hector Ruiz is part of the problem.
    1
  108.  @madigorfkgoogle9349  For the desktop, 68K has a failed road map.
    1
  109. Both 68000 and 80286 have 16-bit ALU and 24-bit physical memory address features. 68000 has 32-bit instruction set and 32-bit registers which are double-pumped over 16-bit ALU hardware. 68000 views the entire 24-bit memory address as a linear memory model. 68000 userland program can easily transition into the full 32-bit hardware 68020 CPU. 1988-era AmigaOS 1.3 has support for 68020 and 68030. 68020 has a 32-bit linear memory model. 80286 has 16-bit instructions with a 16-bit 64KB segmented memory model. 80386 has a 32-bit linear memory model and 32-bit registers that require the full 32-bit Windows NT. 80286 has integrated MMU advantage which is missing on 68000 and 68020. 68000 is closer to 80386SX.
    1