Optimizing my 68060 machine

[litwarski]

Hello all,


I have recently been testing my A4000T, which has an A3660 CPU card and a ZZ9000 RTG card. The ZZ9000 gives me 256 MB Fast RAM, so in total I have 272 MB Fast RAM. The CPU is rev 5 and is running at 50 MHz.


The problem is that every 68060 demo I'm running seems slow. For example Starstruck should be super smooth with an 060 according to Youtube, but instead I'm getting a slideshow (not sure about the FPS, but I would have guessed I'm using an 040 on this machine). Also, ZDoom is unplayable because it's also running too slow.


SysInfo says that I have almost 40 Mips and 37578 Dhrystones. I suppose this is ok for an 060 CPU.


Is my CPU really just too slow for these demos, or have I not optimized my machine correctly? I have tried using CyberPatcher and the demo version of HSMathLibs, hoping they would make ZDoom run smoother, but no dice there. Another possibility I can think of would be to have a different 68060.library, but I am using the libraries I got with the CPU card on a floppy disk.


I have attached pictures from WhichAmiga and SysInfo.

[Vypr]

Does sysinfo show the FPU? Some reduced cost versions of the 68060 don't have the FPU enabled and a lot fo the demos require that to run properly. Also, depending on what libraries you use it can show the FPU as disabled.

[indigolemon]

Could this be due to the lack of local RAM onboard the 3660? I assume this means the system will be pushing/pulling all the data across the Zorro bus, so even with a fast CPU some things will suffer.

[Chucky]

this is why local RAM on the cpu card is important.

sysinfo is a crappy measurment of performance.
demos usually are very heavy on memoryperformance, so thats why demos run slow on a 3660. if you have 64MB hack on the motherboard it runs better (as motherboard ram is faster than z3 but nowehere close as good as local ram)

so in short. 3660 is a "okish" 060 card but compared to any other 060 card it is actually kinda worthless.

[litwarski]

@Vypr Yes, SysInfo shows that there is a 68882 FPU.

@indigolemon it is possible. However, I'm surprised how for example ZDoom is completely unplayable in full screen, whereas I've seen Youtube videos of super smooth Doom on 68060.

I did also notice that if I'm using the full 1920x1080 resolution in Workbench, it slows down the demos somewhat. So I switch to a smaller resolution first and then launch the demos, and it does point to the direction of the ZZ9000's RAM being a bit slow.

[litwarski]

@Chucky thanks for your response. It confirms that I'll need a better CPU card that has RAM on it, if I want to watch the demos at the intended speed.

[daxb]

Run bustest to see how fast/slow your RAM accesses are. http://aminet.net/package/util/moni/bustest

[trixster]

Motherboard ram is roughly 12MB/s read whereas ram on an accelerator card will be 3-4x faster. As others have said, this is why your performance is ‘poor’.

When running memory speed sensitive games such as Doom, even the difference in ram performance between different accelerators with local ram can be quite dramatic.

[grelbfarlk]

For those stuck with an A3640/60 you can always try Speedramsey to get a little more oomph out of your Motherboard RAM if you have 60ns RAM. Better still is using Speedgeek's 030 state machine but that requires reflashing some PALs on your A3640.

When running Doom you shouldn't really be using higher than 320x240 resolution, it should be quite smooth however.

[Kin Hell]

From an Engineers point of view at Hardware Level:

The Best Cards to have are the CSMKIII or the CSPPC.

They have 64Bit memory access across Dual Channels. (2 x 32Bit) - Is why you have to populate these cards with 2 pairs of Identical SIMM's to get the max performance.

Having said that, a Warp Engine 040 is blisteringly fast with 60ns Ram, though I have no experience of a Warp Engine card modded to use an 060. - The 040 had some slick architecture under the hood.

Trixster & DarrenHD are probably the best guys to advise on Memory Scores.

OS3.9 lets you assign Graphics to Fast Ram rather than Chip Ram. This will help speed up your Amiga slightly, but the Wait States on your A3660 are & always will be High without any Fast Ram on it.

[trixster]

Some 060 bustest figures

1) WarpEngine 96mhz (ram 48mhz) = 54.3MB/s readm, 34.3MB/s writem
2) WarpEngine 80mhz (ram 40mhz) = 45.4MB/s readm, 28.0MB/s writem
3) WarpEngine 50mhz (ram 25mhz) = 27.7MB/s read, 17.8MB/s write
4) CSMK2 66mhz (Ram 66mhz) = 43MB/s read, 30MB/s write
5) CSMK2 100mhz (ram 50mhz) = 37MB/s read, 23MB/s write
6) GVP Trexx-II 060 50MHz = 41.3MB/s readm, 22.3MB/s writem
7) Blizzard 1260 80mhz (ram 80mhz) = 57.6MB/s read, 38.5MB/s write
8) Matze’s yet to be released 68060-TK 100mhz (ram 100mhz) = 87.8MB/ readm, 79.8MB/s writem
9) DarrenHD reported his CSMk3 at 50mhz = 45MB/s read
10) CSPPC 50mhz (60ns ram 50mhz) = 54.4MB/s readm, 38.9MB/s writem
11) TF360 (CD32) 50mhz = 16.5MB/s read, 9.8MB/s write (beta firmware)
12) TF360 (CD32) 75mhz = 24.8MB/s read, 14.5MB/s write (beta firmware)
13) Warp1260 50mhz = 39.4MB/s readm, 30.3MB/s writem
14) Warp1260 80mhz = 58.7MB/s readm, 44.5MB/s writem
15) Warp1260 100mhz = 69.8MB/s readm, 52.0MB/s writem
16) TF1260 50mhz 7 Sep 21 beta fw = 42.6MB/s read, 41.4MB/s write
17) TF1260 80mhz 7 Sep 21 beta fw = 66.6MB/s read, 67.2MB/s write
18) WarpEngine 040 40mhz = 51.8MB/s readm, 28.5 writem

A3000 ZorRam 256MB Zorro3 ram = 6.8MB/s read, 4.0MB/s write
A4000 64MB mobo fast ram = 11.5MB/s read, 6.1MB/s write
A4000 A3640 25mhz 040 = 11.5MB/s readm, 6.1MB/s writem
A3640 with waitstate mod = 13.9MB/s readm, 8.1MB.s writem

[Chucky]

<reminds me to do that test on my TF4060>

[jotd]

same issue with 68060 Blizzard A1260 when game code is located in chip memory... slower than running on 68020...

[johnnybarker]

You may find a striped down workbench will leave more motherboard RAM free (it should be used up first) and so there should be more of the faster motherboard RAM available for running apps/games. When I had a3660 I had another SD card with a light os 3.1 installation when I needed more motherboard ram.

[matt3k]

One other suggestion to squeeze speed is to run a disk cache program like fcache.

As everyone has said, not having local memory to the 060 will greatly reduce performance.

One other longshot to help, track down an SSD. They are much fast than CF.

[Chucky]

What would a SSD help with MEMORYACCESS?

it is not a PC that swaps memory against a disk

[Higgy]

@trixster - I just went to use Bustest but there are a whole load of command options. Could you write down what you are typing in, then I can carry out a comparison test.
Thanks

[trixster]

Try bustest fast

[Higgy]

Cheers for that trixster, the readme looked more complex!

TF360 (CD32) 50mhz = 16.5 MB/s read, 9.8MB/s write
TF360 (CD32) 75mhz = 24.8 MB/s read, 14.5MB/s write

[grelbfarlk]

GVP Trexx-II 060/50MHz:
BusSpeedTest 0.19 (mlelstv) Buffer: 262144 Bytes, Alignment: 32768
========================================================================
memtype addr op cycle calib bandwidth
fast $08830000 readw 58.0 ns normal 34.5 * 10^6 byte/s
fast $08830000 readl 96.2 ns normal 41.6 * 10^6 byte/s
fast $08830000 readm 96.9 ns normal 41.3 * 10^6 byte/s
fast $08830000 writew 58.7 ns normal 34.1 * 10^6 byte/s
fast $08830000 writel 117.6 ns normal 34.0 * 10^6 byte/s
fast $08830000 writem 179.4 ns normal 22.3 * 10^6 byte/s