terriblefire wrote: ↑Thu Jan 13, 2022 1:21 pm
Benchmark wise, my DMA reads from my FastATA V are in the ballpark of my Blizzard PPC 060 on alpha, which was most definitely not the case on original firmware. However, screen redraws and boot speed are slow on alpha compared to original firmware. I'm curious what's fundamentally different about the mainboard access methods and if there is symmetry between chip read and write speeds.
Chipram access is measurably faster than the previous firmware. Hugely. By almost a factor of 2. Reads are slightly slower than writes. 7Mb/s writes is the limit for AGA chipset. This is what we get. Its not really possible to go any faster.
Use bustest (
https://aminet.net/package/util/moni/bustest) with chip/fast parameters and you will see.
I downloaded bustest and put it through its paces. First let me say that in all cases booting with no startup-script didn't significantly change the benchmarks of bustest. Here's where I'm at:
Bustest (OFW = Original FirmWare):
ACA1233@40MHz (68030): chip readm = 6.4 writem 7.0 <= AFAIK this is a bit of an apples vs oranges
Blizzard PPC/060 @ 50MHz: chip readm 3.5 writem 7.1
TF1260 OFW @ 50MHz: chip readm 3.5 writem 3.5 <= This firmware draws as fast as the other cards
TF1260 Alpha @ 50MHz: chip readm 4.4 writem 6.9 <= This firmware draws slow and boots slow (see below)
So the tests confirm what you're stating yet my actual performance is as such:
ACA1233 boot time: 27s
PPC/060 boot time: 47s (note this does a patch and reset before it even starts to boot)
TF1260 OFW boot time: 30s
TF1260 Alpha boot time: 57s <= Slower than I realized
A moderately long directory listing in a shell window takes about 50% longer on alpha than OFW.
I am lost as to what the root cause of this is. I can say that I distinctly remember beta having the same issue as alpha, though I did not re-flash it for this round of tests.
I can say that overall alpha does what I need much better than OFW despite this UI sluggishness in that things like playing audio on my megamix is FAR more stable on alpha.
And yes, I'm not changing anything - not screen modes, bpp, etc., etc. I definitely don't want to add variables.
I did take a video of my boot time on OFW + alpha flashing via Xilinx tool + boot on Alpha that clearly shows I'm not making things up or changing parameters somehow.
My view right now is this is merely additional data-points in getting feedback to the project, but if I can be of help as a testing edge-case or if you have any inkling of something else it could be please let me know.
Again, I appreciate all your feedback and the massive amount of work you have done.
Greg
P.S. If anyone else out there with deep Amiga hardware or software knowledge has any inkling of what could possibly cause such weird behavior, please let me know. Logic would dictate that the faster bustest speeds would dictate better, not worse, boot time. Note, I do also get faster disk benchmarks in sysinfo on alpha vs OFW. So bizarre...