Welcome, Guest. Please login or register.
Amiga Kit Amiga Store Hollywood MAL AMIStore App Store A600 Memory

AuthorTopic: Gold 2.7 release for the Vampire accelerators  (Read 14511 times)

0 Members and 1 Guest are viewing this topic.

Offline guibrush

Gold 2.7 release for the Vampire accelerators
« on: March 02, 2018, 10:28:03 AM »
Hi

The Apollo team is happy to announce the immediate disponibility of the Gold 2.7 core for all Vampires 500, 500+ and 600. You can right now dowload the appropriated core on this link :

http://wiki.apollo-accelerators.com/doku.php/start#latest_core

The Gold 2.7 core has many inprovements, the most notable one being of course the hardware FPU. But this is not the only one new thing ! Here is the changelog :

[GOLD2.7] (02.03.2018)
* Added fully pipelined hard FPU with peak 78 MFlops
* Added native support for all 68882 FPU operations without need of extra libraries
* Added new high performance memory controller, reaching 600 MB/sec memory speed
* Added MapRom support
* Added Workbench hardware sprite support
* Added Scanlines optional mode for retro look of games on HDMI output
* Added RangerMem (A500 slowmem) for better compatibility with some old games/demos
* Added new boot pic
* Improved MPU (memory protection) with catching memory access error Enabled
* Improved AMMX 2 instruction set with acceleration features for video decoding, image decoding, 2D-games, 3D-games
* Improved WHDLOAD compatibility improved
* Improved Faster Boot time

Like always, we are waiting your feedback on IRC :
chat.freenode.net 6697 (SSL)/6667 (non SSL) #Apollo-Team

We wish you a good installation !
 

Offline midway

Re: Gold 2.7 release for the Vampire accelerators
« Reply #1 on: March 03, 2018, 02:09:38 AM »
Great! Will test tomorrow. So how much faster is this fpu than the standard 68060fpu, which was not pipelined?
 

Offline AJCopland

Re: Gold 2.7 release for the Vampire accelerators
« Reply #2 on: March 03, 2018, 01:37:21 PM »
Quote from: midway;836829
Great! Will test tomorrow. So how much faster is this fpu than the standard 68060fpu, which was not pipelined?


The only quote I can find is for 36 MFLOPS @ 66 Mhz from http://www.amigawiki.de/doku.php?id=de:parts:68060

Offline midway

Re: Gold 2.7 release for the Vampire accelerators
« Reply #3 on: March 03, 2018, 06:36:55 PM »
Yeah saw that on their site. I usually like waiting with updates just to be safe ;). Anyone tried it yet?
 

Offline PanterHZ

Re: Gold 2.7 release for the Vampire accelerators
« Reply #4 on: March 04, 2018, 01:58:15 AM »
Since an FPU is now included, what on earth will certain people have to whine about now :)
 

Offline slimf

Re: Gold 2.7 release for the Vampire accelerators
« Reply #5 on: March 04, 2018, 06:07:13 AM »
This is Amiga.org - they will find something.
 

Offline nyteschayde

  • VIP / Donor - Lifetime Member
  • Hero Member
  • *****
  • Join Date: Mar 2002
  • Posts: 638
  • Total likes: 2
    • http://www.nyteshade.com
Re: Gold 2.7 release for the Vampire accelerators
« Reply #6 on: March 04, 2018, 07:48:46 AM »
Woohoo! Seems more stable, installed well on first try from the Amiga side. And I love the new bootpic!!

Keep up the good work. I love my Vampire. Will be buying more once the V4s are out.
Senior MTS Software Engineer with PayPal
Amigas: A1200T 060/603e PPC • A1200T 060 • A4000D 040 • A3000 (x2) • A2000 Vamp/V2 • A1200 (x4) • A1000 (x3) • A600 Vamp/V1 • A500
 

Offline pVC

Re: Gold 2.7 release for the Vampire accelerators
« Reply #7 on: March 04, 2018, 08:29:04 AM »
Quote from: PanterHZ;836857
Since an FPU is now included, what on earth will certain people have to whine about now :)


Quote from: slimf;836867
This is Amiga.org - they will find something.


Not on this forum it seems. Other forums have already reported that the FPU implementation doesn't work properly.

For example, launch SYS:Tools/Calculator and try to calculate 5 * 1.3, and the result will be 6.499999999069.

More about issues in the FPU implementation over here:
https://blog.alb42.de/2018/03/03/vampire-2-7-fpu-part-2/
Daily MorphOS user and Amiga active.
 

Offline roomeo

Re: Gold 2.7 release for the Vampire accelerators
« Reply #8 on: March 04, 2018, 11:23:37 AM »
Quote from: PanterHZ;836857
Since an FPU is now included, what on earth will certain people have to whine about now :)


Quit whining youself.. head over to Apollo core forum for the hype. :rofl:
a500 - Vampire500+
a600 - aca620/indiECS/Subway - Soon to be sold. :(
a4000 - 16mb Fast / Apollo4060@66mhz (ram not working) / X-Surf + RapidRoad / FastlaneZ3 64mb + Acard ARS-2000FU -> CfIDE / Delfina Lite / PicassoIV ... Who needs PCI?!
 

Offline psxphill

Re: Gold 2.7 release for the Vampire accelerators
« Reply #9 on: March 04, 2018, 11:24:11 AM »
Quote from: pVC;836870
Not on this forum it seems. Other forums have already reported that the FPU implementation doesn't work properly.

For example, launch SYS:Tools/Calculator and try to calculate 5 * 1.3, and the result will be 6.499999999069.

More about issues in the FPU implementation over here:
https://blog.alb42.de/2018/03/03/vampire-2-7-fpu-part-2/

Assuming this is purely due to the cutdown 64 bit precision (rather than the full 80 bit precision of the 68881/68882/040fpu/060fpu) because the full 80 bit implementation takes too much FPGA space. Then you either have to wait for new larger Vampire FPGAs or for Gunnar to have an epiphany on how to make it fit.

It's possible it's a different issue, but supposedly you should only expect scene demos to look ok and not anything that expects the correct result.

This was already discussed on amiga.org before the release, nobody wanted to accept that people might expect correct results and that it would be a problem.

According to https://blog.alb42.de/2018/03/02/vampire-v2-7-with-fpu/ Apollo won't be compatible with any software that uses Round() until you can get a version that is 100% accurate.

"The Round() on 68k always uses a helper for now, because FPC's Round() is defined to return an int64 value (signed 64bit integer). This cannot be done by the hardware on 68k, so the code currently calls the round helper, which is a pretty big function, and depends on the IEEE bit-correct result of all operations, as it does some integer bit tricks on floats. (There could be some special cases added, when the expected integer result is less or equal to a 32bit signed int, but that's beside the point now.)"
« Last Edit: March 04, 2018, 11:46:38 AM by psxphill »
 

guest11527

  • Guest
Re: Gold 2.7 release for the Vampire accelerators
« Reply #10 on: March 04, 2018, 12:04:47 PM »
Quote from: psxphill;836874
Assuming this is purely due to the cutdown 64 bit precision (rather than the full 80 bit precision of the 68881/68882/040fpu/060fpu) because the full 80 bit implementation takes too much FPGA space.
The calculator uses DoubBas and is hence purely 64bit driven, even if there is a 80 bit FPU in the system. Unfortunately, the CPU-only implementation is not too good either, and it is not a good reference how IEEE math should be done.

Proper floating point math requires three additional excess bits, a round bit, a guard bit and a sticky bit. The latter gets 1 once "1"s are shifted out of the right of the mantissa (i.e. on a precision loss). Then, rouding according to the round bit, with "tie-resolution" in case we have the bit combination "100" at the end generates proper results. (Proper in the sense of: We do not have additional loss beyond the loss unavoidable due to the limited precision).

The MathIEEE libs do not get this done right, except if there is a FPU. Mathffp and MathTrans are jokes as far as the math model is concerned. Inproper rounding, no overflow handling, no NANs, no INFs.

One way or another: I already commented that you need some excess precision for double precision output of transcental math functions, i.e. 64 bit intermediate results are not sufficient for 64 bit output. In that sense, a IEEE to ASCII conversion is also a "transcental function". The calculator uses here the SAS/C built-in function, which is "ok", but does not deliver the best possible precision. Such a conversion cannot be done through the math libraries, unfortunately (at least not without loosing precision) and requires for 80 bit output even 96 bit precision.

Again, please note that the Os is not exactly a good reference here. IEEE math libraries are right with the FPU installed. With the CPU installed, the quality is pretty poor. There are no "completely bogus" results as you could get with the pre-V45 releases (pow() returned pure nonsense in some cases), but rounding is bad and precision is bad.
 

Offline kolla

Re: Gold 2.7 release for the Vampire accelerators
« Reply #11 on: March 04, 2018, 05:34:29 PM »
Quote from: PanterHZ;836857
Since an FPU is now included, what on earth will certain people have to whine about now :)


It's not an FPU, it's a Quake enabler.
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC
---
A3000/060CSPPC+CVPPC/128MB + 256MB BigRAM/Deneb USB
A4000/CS060/Mediator4000Di/Voodoo5/128MB
A1200/Blz1260/IndyAGA/192MB
A1200/Blz1260/64MB
A1200/Blz1230III/32MB
A1200/ACA1221
A600/V600v2/Subway USB
A600/Apollo630/32MB
CD32/SX32/32MB/Plipbox
CD32/TF328
A500/V500v2
A500/MTec520
CDTV
MiSTer, MiST, FleaFPGAs and original Minimig
Peg1, SAM460 and Mac minis with MorphOS
 

Offline psxphill

Re: Gold 2.7 release for the Vampire accelerators
« Reply #12 on: March 04, 2018, 10:17:14 PM »
Quote from: Thomas Richter;836876
With the CPU installed, the quality is pretty poor. There are no "completely bogus" results as you could get with the pre-V45 releases (pow() returned pure nonsense in some cases), but rounding is bad and precision is bad.

Is that something you'd be willing to improve?

I'm unclear about the rest of the post what the cause of the FPU errors are that the guy is reporting. Is the calculator always broken with an FPU (my amiga is currently boxed away)
 

Offline kolla

Re: Gold 2.7 release for the Vampire accelerators
« Reply #13 on: March 05, 2018, 01:38:03 AM »
Quote from: psxphill;836896
Is the calculator always broken with an FPU (my amiga is currently boxed away)

No, calculator works fine with real FPU.

Quote
People shall use the FPU for what it is intended to on the Vampire, that is to run most of the Amiga Demos and Games requiring an FPU.

As I commented, it's not really an FPU as much as a Quake enabler - any hope for a proper FPU on V2, is long gone. In many cases, FEMU works better (more accurate) than the FPU of core 2.7, but sadly it has its issues and development has stopped. So, another FPU emulator is needed for V2.
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC
---
A3000/060CSPPC+CVPPC/128MB + 256MB BigRAM/Deneb USB
A4000/CS060/Mediator4000Di/Voodoo5/128MB
A1200/Blz1260/IndyAGA/192MB
A1200/Blz1260/64MB
A1200/Blz1230III/32MB
A1200/ACA1221
A600/V600v2/Subway USB
A600/Apollo630/32MB
CD32/SX32/32MB/Plipbox
CD32/TF328
A500/V500v2
A500/MTec520
CDTV
MiSTer, MiST, FleaFPGAs and original Minimig
Peg1, SAM460 and Mac minis with MorphOS
 

Offline kolla

Re: Gold 2.7 release for the Vampire accelerators
« Reply #14 on: March 05, 2018, 01:51:01 AM »
BTW - I have not found a way to permanently disable the "FPU", apart from going back to core 2.5 - the VControl can only turn it off till next reboot (or to be more... hohum... accurate, next crash).
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC
---
A3000/060CSPPC+CVPPC/128MB + 256MB BigRAM/Deneb USB
A4000/CS060/Mediator4000Di/Voodoo5/128MB
A1200/Blz1260/IndyAGA/192MB
A1200/Blz1260/64MB
A1200/Blz1230III/32MB
A1200/ACA1221
A600/V600v2/Subway USB
A600/Apollo630/32MB
CD32/SX32/32MB/Plipbox
CD32/TF328
A500/V500v2
A500/MTec520
CDTV
MiSTer, MiST, FleaFPGAs and original Minimig
Peg1, SAM460 and Mac minis with MorphOS