@Gunnar

Quote:

Gunnar wrote: Correct ship AHI driver already pre-installed in ApolloOS and also available for download since many years. These AHI driver allows you to use high quality 16bit Audio. Regarding AHI. AHI by design works somewhat against the concept of Amiga designers. The Amiga computer design is based on the Amiga DMA hardware doing the work. What does DMA mean? DMA means "direct memory access" This means the Amiga chipset can by itself read or write from memory and the CPU not needs to push memory around to e.g. display gfx or make sprites move, or make a sound play. The Amiga Computer design is based on this all be done by the Amiga chipset with DMA. Where is the violation of AHI? The violation to the Amiga spirit of the AHI abstraction layer is that it requires the CPU to push all the memory around for Audio is played. Is this violation good or bad? If you speak to Amiga designers and engineers like Dave Haynie or if you speak to Amiga Modplayer developers then that you will hear is that using DMA on the Amiga is one of its key strength. And using the CPU for this is against the good concept of the Amiga.

The main point for custom fixed-function ASICs is to shift the bottleneck on the available memory bandwidth.

If a very powerful SIMD-enabled CPU (with fast thread switching) can saturate the available memory bandwidth, any talks about DSP are pointless. Switching between software threads on a single-thread CPU incurs context switch overheads.

68K Amiga has the context of weak CPUs doing 16-bit multimedia use cases.

For the AmigaOS with a fast 68K CPU, GMPlay's or Timidity's or Doom's software synthesizers can be brute forced.

Last edited by Hammer on 14-Mar-2024 at 03:50 AM.
Last edited by Hammer on 14-Mar-2024 at 02:16 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Hammer

Quote:

Hammer wrote: @cdimauro Quote: I don't read Wikipedia: I read Intel's and AMD's architecture manuals. Hence, I've already acquired the knowledge singe AGES. The RIGHT knowledge. And I don't need to read something that anyone can write and write it wrong. https://community.intel.com/t5/Intel-ISA-Extensions/unaligned-loads-avx-128-vs-256/m-p/949434 AVX-128 is a real terminology for direct SSE intrinsic to AVX-128 port.

Seriously? You're so desperate to report a link of any USER which has written a post on Intel's COMMUNITY forum and that has written "AVX-128"?

And just for this then the AVX-128 term magically existed, right? Even if there's NO trace, at all, on Intel's and AMD's architecture manuals...

That's a typical example of "information" which is completely build on/in internet. Which, unfortunately for you, still has ZERO value

Because internet != knowledge. Internet does NOT build the knowledge. Internet is NOT the literature of reference when talking about a topic.

At least, not for people which wish to acquire the knowledge studying the proper literature.
Quote:

https://www.intel.com/content/dam/develop/external/us/en/documents/11mc12-avoiding-2bavx-sse-2btransition-2bpenalties-2brh-2bfinal-809104.pdf Subject : Introduction to AVX-SSE transition penalties Intel has "128-bit Intel AVX" instruction terminology usage.

I've already reported before an example of 128-bit instruction. Which does NOT mean that "AVX-128" existed!

There's not a single trace of this or "AVX-256". Guess why: they do NOT exist! They ONLY EXIST ON YOUR MIND!
Quote:

Unlike you, I'm transparent with the microarchitecture's implementation.

Unlike, I've studied all Intel's and AMD's manual and I've enough knowledge to understand the difference between an ISA = Instruction Set Architecture and one of its possibile implementations AKA microarchitecture. And I do NOT mix them like a pizza Hawaii (SIC!!!).
Quote:

Front-end instruction sets are meanless when performance is attached to the debate.

In fact, performance = microarchitecture's domain.

NOT architectures's domain. An ISA can expose features / extensions for addressing performance problems, but it's entirely to the microarchitectures to exploit it.

In fact, a processor with a super cool vector extension can perform much worse than another one with just the class SIMD unit, and maybe even using the GP registers for that.
Quote:

Furthermore, certain AVX-128 small-datatypes don't scale with AVX-256-bit registers and they are not available. Read Page 3 from https://hpc.llnl.gov/sites/default/files/intelAVXintro.pdf

Guess what: there's no "AVX-128" neither "AVX-256" on this document. There's even no "AVX-" on that. Because... rolling drum... they do NOT exist!

And this shows that you don't even read what you've found desperately seeking on internet about something which could support the complete non-sense that you've invented and pretend to be real.

You're hopeless. And also ignorant.
Quote:

Newer AVX extensions have BF16 or FP16, VNNI (byte, word) smaller datatypes for 256-bit and 512 registers.

Where it was stated in Intel's and/or AMD's manuals that AVX should be fully symmetric?!? Even AVX (the first version) introduced instructions supporting only some datatypes.
Quote:

There are gotchas with AVX.

What a news...
Quote:

Quote: Is AMD's implementation splitting the AVX-512 instructions in two micro-ops working on 256 bits at the time? That's correct. For Zen 4, AVX-512 is used as an instruction issue slot conservation, 32-register programming model, and access to AVX-512's Ice Lake extensions. The four 256-bit SIMD units are not symmetric (two FMA3 complex, two FADD simple), hence they have gotchas.

Which explains why those processors don't reduce the clock frequency when AVX-512 instructions are (intensively) used...
Quote:

https://www.phoronix.com/review/intel-sapphirerapids-avx512/8 There's a larger AVX-512 uplift with Intel Sapphire Rapids when compared to AMD's Genoa's. AMD (96 cores with EPYC 9654) still beats Intel (60 cores with Xeon Platinum 8490H) due to a larger CPU core count due to perf/watt advantage. EPYC's Zen 4C has reached 128 cores.

So, AMD needs more cores to "beat" (!) Intel...
Quote:

Zen 5 has improved on Zen 4's AVX-512 cost-reduction design i.e. full-performance AVX-512. https://www.hwcooling.net/en/amd-confirms-zen-5-details-6-alus-full-performance-avx-512en/ GCC confirms that Zen 5 contains 512-bit SIMD units for the first time in AMD processor history.

So, finally AMD should reach Intel. It took some time...
Quote:

My main selection criteria for AM5 is a proper CPU microarchitecture upgrade roadmap.

Irrelevant.
Quote:

Hammer wrote: @Gunnar Do you support AVX-512 has 16 registers?

Why he should? AVX-512 had no 16 registers version: it has only 32 registers.

And his processors are NOT x64 compatible. They are (mostly) 68k compatibile.

So, it's totally irrelevant.

BTW, when do you plan to post images which are compatible with this very old forum's software? You have broken again the layout of the site! I don't need super zoomed images: to read things (at least not yet)!
Quote:

Hammer wrote: @Gunnar Quote: Gunnar wrote: @Hammer THIS FORUM IS AN AMIGA FORUM. GUNNAR: Hammer did you know that this is an Amiga Forum? Can you stop cross posting Intel stuff from Wikipedia to an Amigaforum? What is wrong with you? Are you on drugs? Reminder, WinUAE, Cloanto's Amiga Forever (with official Amiga Corporation's licensed AmigaOS 3.X, standalone bootable into Linux KX Light/UAE) and Amithlon run on X86 PCs. My licensed Amiga Forever 2016 is bundled with licensed AmigaOS 4.1 FE. My WinUAE's AmigaOS 4.1 FE installation is operational with 3D acceleration i.e. WinUAE's 3D API bridge or 3DFX Voodoo 3. I have a Ryzen 5 7600X PC dedicated to WinUAE and QEmu 8.2. Your ApolloOS (AROS-based) is not even Amiga Corporation licensed. I find it strange when Amiga Corporation's and Hyperion Entertainment's officially licensed product that runs on X86 PCs is ejected from this Amiga forum. LOL This is not your censored Apollo Forum. Look in the mirror. --- When the price and certain PowerPC model are right for AmigaOS 4.1 FE PowerPC hardware, there's a high probability I will purchase it. PowerPC e6500 is on my consideration list. I already reserved an AMD GCN GPU card (specifically, Sapphire's Radeon HD "7970 3G GDDR5 OC with Boost, 1197-03-40G) that is compatible with A-EON's Enhancer System 54. https://hdrlab.org.nz/projects/amiga-os-4-projects/radeonhd-driver/radeonhd-driver-hardware-compatibility

Which is totally irrelevant and a non-sense considered this part of discussion...

@Hammer

Quote:

Reminder, WinUAE, Cloanto's Amiga Foreve

Do you think that the Amiga users in this forum
like your irritating off-topic x86 and non logical copy-pasted posts from wikipedia?

Or is it more likely that people think that you have a mental health issue?

@Hammer

Quote:

The main point for custom fixed-function ASICs is to shift the bottleneck on the available memory bandwidth. If a very powerful SIMD-enabled CPU (with fast thread switching) can saturate the available memory bandwidth, any talks about DSP are pointless. Switching between software threads on a single-thread CPU incurs context switch overheads. 68K Amiga has the context of weak CPUs doing 16-bit multimedia use cases. For the AmigaOS with a fast 68K CPU, GMPlay's or Timidity's or Doom's software synthesizers can be brute forced.

Why do your post always sound like you googled them?


- Amiga Audio hardware provides "free" mixing of samples.
- A CPU can also mix audio channels.
While simply adding up samples of same sample rate is not difficult for a CPU.
Transcoding the sample rate is a lot more expensive for the CPU.

The AMIGA Audio hardware does support mixing of samples of different periods !
The Mixing Features of e.g. Super-AGA they give the Amiga for free high quality mixing -
this equal like having a free dedicated 68030 CPU just for mixing.

This makes the whole system faster and more responsive

@Gunnar

Quote:

Or is it more likely that people think that you have a mental health issue?

Why this obsession with "mental issues"?

There is now a mighty long list of people who you at some point have claimed or suggested have "mental issues", and the list keeps growing. At this point, it's almost like an honours list.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@kolla

How would you explain his behavior?


Hammer has spammed hundred of times Amiga discussion here with off-topic Intel/AMD posts.

I dont know if Hammer simply not understand what people talk about ...
and that his post are totally off-topic and irrelevant.
Or if this is an Intel posting obsessive-compulsive disorder?

Last edited by Gunnar on 14-Mar-2024 at 12:13 PM.

@Ferry

Quote:

Because almost ALL software available for Amiga is made for 68k.

And the 4350 titles available at OS4depot.net are... what? Note that unlike Aminet, OS4depot refrains from storing media files such as music modules or pictures, so most of the items there are binaries. And while of course not all of them are actual programs or utilities (some are libraries etc.), the number of OS4-native applications goes well over a thousand. No dispute that classic 68K Amigas have waaaaay more; there's absolutely no contest! But the idea that AmigaOS4 has no native applications, and that PPC users have to rely on emulating 68K software, is not based on reality. It would actually take me some time to remember which 68K programs I still use - there won't be many.

Quote:

why would anyone buy TODAY a VERY expensive PPC machine to run emulated 68k programs?

Exactly, why would they? Because they don't: as I've said above, OS4 has enough native software to get you going. 68K emulation under OS4 is nice when one gets the nostalgia bug from time to time, but it's more like a bonus.

_________________
The Rear Window blog

AmigaOne X5000/020 @ 2GHz / 4GB RAM / Radeon RX 560 / ESI Juli@ / AmigaOS 4.1 Final Edition
SAM440ep-flex @ 667MHz / 1GB RAM / Radeon 9250 / AmigaOS 4.1 Final Edition

@Gunnar

Quote:

How would you explain his behavior?

Or how do I explain your behavior?

Quote:

Hammer has spammed hundred of times Amiga discussion here with off-topic Intel/AMD posts.

Here you go with your "hundred" again... prost!

I don't find Hammer's postings off-topic... on edge of relevance perhaps, but not really off-topic, no more off-topic than so much else going on here.

Quote:

I dont know if Hammer simply not understand what people talk about ... and that his post are totally off-topic and irrelevant. Or if this is an Intel posting obsessive-compulsive disorder?

Well, what can you do about it? Not much. So why obsess over it?

Is SAGA done yet?

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@Karlos

And also why I said it would be another tracker as well. OctaMED could have been capable. But as the "Octa" gives a clue, the extra channels are built off soft mixing and the improved mixing routines, which when played on a common 16 bit sound card tended to have only one stereo channel. To tell the truth, I can hardly hear the difference between Paula and the real thing. But, I haven't compared module mixing to Paula playing, side by side.

@Hammer

Quote:

FYI, X86-64v4's AVX-512's EVEX encoding has 32 vector registers i.e. zmm0 to zmm31 for AVX-512, ymm0 to ymm31 for AVX-256 and xmm0 to xmm31 for AVX-128.

It's met its match then. Things can only get better, as the song goes, but not for PPC. Well now the internal RISC core of x64 has come out of the shell. Perfect match for emulating PPC register to register!

Quote:

AVX supports both integer and floating-point formats.

If only the Tabor SPE supported such a dynamic arrangement.

Quote:

32 zmm registers x 512 bit = 16,384 bits.

Game over.

@Trixie

I used aminet as main source for my distribution. Most of the stuff there is for 68k, most of it uploaded until about 1998. Only a minority also exists on NG or even more rare is exclusively NG. I cannot tell exact numbers but it is easily to see. Browse it, take a normal category like utils and look what is 68k or NG (AmigaOS/MorphOS/AROS)

@Gunnar

Quote:

These AHI driver allows you to use high quality 16bit Audio.

And all the channels as well?

Quote:

AHI by design works somewhat against the concept of Amiga designers. The Amiga computer design is based on the Amiga DMA hardware doing the work.

How would that go against the AHI design? Obviously it's designed as an abstract API for retargetable audio. And letting the driver to program the audio hardware to do any DMA under the hood.

Quote:

The violation to the Amiga spirit of the AHI abstraction layer is that it requires the CPU to push all the memory around for Audio is played.

It has to be pushed to chip RAM by the CPU so what's the issue?

Quote:

And using the CPU for this is against the good concept of the Amiga.

Using CPU for what? Audio mixing? Well that's not he fault of AHI. AHI is just an upper level API for playing sounds. While AHI can offer soft mixing with various features it's not a requirement. It depends on the setup and audio hardware.

It's a common fallacy to think AHI means soft mixing. This is a misconception. Likely based on most supported sound cards, such as 16 bit ones, only supporting one stereo channel.

AHI supports DMA out of the box. But it needs a driver that supports it as well as the hardware. I'll give an example, the most well known, the Paula driver. This can be set to DMA. When I was helping James Conwell (RIP) port his multi module player in Total Chaos to AHI, I did it the "proper way" by making sure that DMA was used in Paula driver. As well as using the module playing mode of AHI.

Unfortunately, some modern module AHI players take it into their own hands to do all mixing themselves, so remove the possibility for AHI to use a DMA driver. But, they usually target 16 bit hardware. And also unfortunate, is that the AHI module mode locks out the audio, even though the same thing happened on Paula anyway.

Last edited by Hypex on 14-Mar-2024 at 03:55 PM.

@Gunnar

Quote:

- A CPU can also mix audio channels. While simply adding up samples of same sample rate is not difficult for a CPU. Transcoding the sample rate is a lot more expensive for the CPU.

At risk of adding up amplitude of different wave forms at its peak, causing clipping.

More typical be:

wr = (w0 + w1 + w2 + w4) / 4;

But that can cause, dampening effect.

A better approach:

wr = (w0 * weight0 + w1 * weight1 + w2 * weight2 + w3 * weight3 ) / tot_weight;

Weight number try to compensate for lack sound on channels.
balance is calculated depending on channels use. I’m sure there are more ways to mix channels / voices.

W0 is wave form amplitude at function of time with period T.

Last edited by NutsAboutAmiga on 14-Mar-2024 at 09:18 PM.
Last edited by NutsAboutAmiga on 14-Mar-2024 at 09:17 PM.
Last edited by NutsAboutAmiga on 14-Mar-2024 at 09:16 PM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@Hypex

Some audio drivers do not support many channels, just 1, or only a few.
if that’s case, you better off, trying to mix the channels yourself, at least then you don’t lock other programs out.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@cdimauro

Quote:

Seriously? You're so desperate to report a link of any USER which has written a post on Intel's COMMUNITY forum and that has written "AVX-128"?

For vendor-neutral, I removed Intel from "128-bit Intel AVX".

There is no substantial difference between 128-bit AVX and AVX-128.

I was also annoyed with Intel's so-called 128-bit SSE with Pentium III when the fuking hardware implementation is 64 bits wide.

Pentium IV's so-called 128-bit SSE2 was a 64-bit hardware implementation. The same for K7 Athlon XP.

To remain competitive, both AMD and Intel needed higher clock speeds.

AMD K8's 128-bit SSE had 128-bit FADD and 64-bit MUL implementation. A no brainer to why K8 AThlon FX has better IPC when compared to Pentium IV EE.

Intel's 1st trueful 128-bit SSE hardware is with Core 2.

AMD's 1st trueful 128-bit SSE hardware is with K10 Athlons.

I give credit to PowerPC camp's 128-bit Altivec with truthful 128-bit SIMD hardware implementation.

Quote:

And just for this then the AVX-128 term magically existed, right? Even if there's NO trace, at all, on Intel's and AMD's architecture manuals...

AMD was minimizing Zen 1's four 128-bit SIMD units for its AVX2 implementation.

AMD's less than transparent bullshit AVX2 support. Zen 1's 128-bit AVX2 units are from Pipedriver's 128-bit AVX2. Hint: Bulldozer's FMA4 instructions still work when the feature check is ignored.

I didn't buy AMD's Zen 1 since I had Intel Core i7-7820X Skylake X during this era.

AMD's Zen 2 had a 256-bit hardware implementation.

Quote:

That's a typical example of "information" which is completely build on/in internet. Which, unfortunately for you, still has ZERO value

A full Gracemont SKU like the Intel Core i3 N205 SKU has AVX2 with two 128-bit vector FP/integer pipelines with 3rd vector 128-bit pipeline being integer only.

128b ALU/128b FADD/128b FMA/128b FMUL/AES/FDIV/SHA pipe1.
128b ALU/128b FADD/128b FMA/128b FMUL/AES pipe2.
128b ALU pipe3.

Fact: AVX's support does NOT guarantee 256-bit hardware implementation. Read the fine print i.e. the gotchas.

Fact: AVX-512 and AVX10.x has feature modes e.g. AVX-128, AVX-256 and AVX-512.


Your less transparent 256-bit AVX argument has ZERO value when attached to a performance debate!

Last edited by Hammer on 15-Mar-2024 at 04:57 AM.
Last edited by Hammer on 15-Mar-2024 at 04:39 AM.
Last edited by Hammer on 15-Mar-2024 at 04:36 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@cdimauro

Quote:

There's not a single trace of this or "AVX-256". Guess why: they do NOT exist! They ONLY EXIST ON YOUR MIND!

Wrong.

https://i.ibb.co/wR75gzd/AVX512-optional-under-AVX10.png
This is from the Intel's AVX10 document.

Fact: Intel released AVX-512/AVX10.x feature matrix. AVX-512 is not guaranteed on Intel platforms.

My AVX-128 (128-bit AVX) and AVX-256 (256-bit AVX) usage refers to the hardware implementation.

It's foolish to trust the "256-bit AVX support" feature tickbox marketing.


Quote:

Why he should? AVX-512 had no 16 registers version: it has only 32 registers.

Do you support the claim that AVX-512 has 16 registers? This is a minor mistake.

Quote:

And his processors are NOT x64 compatible. They are (mostly) 68k compatible. !

And?

Last edited by Hammer on 15-Mar-2024 at 06:20 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Gunnar

Quote:

Gunnar wrote: @Hammer Why do your post always sound like you googled them?

What's the matter?

PiStorm32-Emu68-PRi 4B's hyper-fast 68040 brute forcing GMPlay's or Timidity's or Doom's software synthesizers with A1200's 14-bit 56 kHz mode Paula chip is not a theory.

My Midi file collection survived from my 1992 era A3000 to my modern PCs. The same Midi file collection returns to my A1200's solid-state storage.

I have used HORNY midi sequencer with GMPlay software synthesizer and RTG workbench on PiStorm32-Emu68-PRi 4B.

For the point of reference, I do have a Yamaha ymf724 DSP XG midi PCI card and a Yamaha S-YXG50 software synthesizer during the Pentium II era. My current Yamaha XG piano is USB-connected with the PC and I also use S-YXG50 VSTi plug-ins. I'm also aware of AMD's True Audio Next (TAN) on game consoles and Steam Deck.

Quote:

- Amiga Audio hardware provides "free" mixing of samples.

I'm aware of this.

Quote:

- A CPU can also mix audio channels.

I'm aware of this i.e. refer to GMPlay and Timidity use case examples.

Quote:

While simply adding up samples of same sample rate is not difficult for a CPU. Transcoding the sample rate is a lot more expensive for the CPU.

I used GMPlay on my A3000 (030 @ 25 Mhz) and I know this CPU's limitation.

Simple adding up multiple samples can result in audio clipping. An additional process is applied to minimize this problem.

Quote:

The AMIGA Audio hardware does support mixing of samples of different periods !

I'm aware of this.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Gunnar

Quote:

Do you think that the Amiga users in this forum like your irritating off-topic x86 and non logical copy-pasted posts from wikipedia?

This topic is about PPC vs the world i.e. alternative CPU instruction sets.

I'm on topic and you're not on topic.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Hammer

Quote:

This topic is about PPC vs the world i.e. alternative CPU instruction sets. I'm on topic and you're not on topic.

So your goal is walls of text wikipedia posts was "bash" PPC for using 128Bit Altivec?

But is your argument maybe flawed?

Background:
Vector instructions help to improve performance for a some use cases.
Most important here are the "features" of the instruction - what operation can it do?
And also very important are what limits does it have - e.g. can they work only on certain alignments?

2nd comes how many instructions can I execute per cycle / per second
And how does this work in correlation with my Data Cache troughput and memory transfer rate.


While advertising with e.g 1024 bit vector registers looks very nice on paper - and can easily fool children - But if the CPU loads only 64bit from Data-Cache per cycle - or only can push 8 byte per cycle to memory. Then in most real world use cases this is a totally unimportant number-
And is only nice on paper!

@Hammer


Maybe it helps to look at Sprites and Vector instruction to visualize this better.


If your hardware supports 8 sprites each 16 pixel width.
And you upgrade this that your sprites support 8 sprite of 128 pixel width.
Then YES this help a few games.

Some games can use a 128 pixel monster sprite.

But many games use sprites for small items like bullets.
So having more sprites instead wider ones - will help more games.

The same is true for Vector code.
There are some algorithms benefit from wider vectors.

But many algorithm do not.
And for them having more small instruction would be MUCH better.
Or making your instructions "stronger" like adding more colors to your sprite..


You wonder why INTEL upgrades this like crazy?

Why do INTEL and IBM follow here different strategies?

Maybe this is because they have different customers?
The customers buying IBM super computers - are educated computer experts.
They understand the math and logic of vector units.


How about INTEL customers?
Many INTEL customer look for paper numbers so that they can boost in online forums:
"my 512bit ding-dong is longer than your 128bit ding-dong"