Printable View
Hi All
ive done the tutorial on the Jap site and attempted to overclock my DC.
The good news, It works.
The bad news I cant actually tell the difference. kajamin, Im a bit confused aboutthe crystal, I have two types so i might swap them. I tried sega rally, and the frame rate was much better (I think).
I tried DCNeogeo (the rom emu, not the cd emu) and it was giveing a lower framerate.
Erm, confused! Kajamin are you about?
Cheers
Q
The mod is only going to speed up the CPU, but in reality that also going to speed everything up..
I'd understand if the 3D would stay the same, but the emu's should speed up. before I pop open DC to check let's start with...
What is the xtal speed your using? and what's confusing you about the xtal?
What I'm thinking is that if the CPU goes out of synch with the system bus, the memory will be effectively slowed down because the CPU will miss the timed calls to the memory that are needed. Even it it would speed up raw CPU calculations a bit, it would still slow down the effective speed.
The only good analogy i can think of is how if you overclock a computer CPU without regard for the multiplier and bus speed, the system could lose performance. From what I understand in some systems the system bus is not very important and keeping the CPU and ram in sync isnt either, but afaik, in the DC it is.
First off, the only chipset I am aware of that has a real problem with out-of-sync memory is the nForce 2 (and maybe 1). But that's a case of the memory being out of sync with the FSB, not the other way around Â*;). How can the "CPU go out of synch with the system bus" when your processor speed is FSB x multiplier?
Besides, we don't have enough info yet. I don't know how the buses in the DC derive their clockspeed, for instance. There's got to be a way to keep things in sync, even if it requires you to overclock other components.
[quote author=Alexvrb link=board=hardware;num=1094318646;start=0#3 date=09/04/04 at 22:10:02]First off, the only chipset I am aware of that has a real problem with out-of-sync memory is the nForce 2 (and maybe 1). But that's a case of the memory being out of sync with the FSB, not the other way around Â*;). How can the "CPU go out of synch with the system bus" when your processor speed is FSB x multiplier?
Besides, we don't have enough info yet. I don't know how the buses in the DC derive their clockspeed, for instance. There's got to be a way to keep things in sync, even if it requires you to overclock other components.[/quote]
There are many different things you can do to manually overclock a CPU idependantly of the FSB. Like I said this was mostly a problem back in the p2/p3 era.
Actually in the case of the sh4, it's Xtal input to PLL1 to divider to PLL2 which might be gated by a Clock pulse gen unit... to SDRAM
There actually hasn't been a "system bus" in ages, in the past the CPU/RAM/ROM clock was considered the system bus, since each component has it own bus today, and each can be clocked async.. well..
There's alot of myth out there... even people who know what they're talking about, fall into the trap sometimes
for example.. what's wrong with this paragraph? when I saw this.. I thought if he only knew
"Anything else you want to do is up to how you want to design it, and what the device's requirements may be. For example, if one wanted to build a PCI bus that actually operated to spec (33MHz CLK minimum), one would need to add an oscillator for the 33MHz clock, and do some weird FIFO buffering to talk to the DC side of things. Not very simple."
Back to the topic.. the cpu to ram can't be the issue, mainly due the fact that it's actually working, since the sh4 controls the ram.. But what Quzar said was almost the same thing I though of first.. just off to the side a bit, one of my guesses is that the interface to the PVR might actually be having problems, and it's causing the slow down.. or the DC's kernal is actually smart enough to add wait states on the ram.. gotta do more spec reading before this makes sense though..
its a 40 mhz crystal. the inputs are nc/c 5v GND and output
I plced the output on the mb once i lifted pin 11 and the nccc on the chip pin which I lifted. (Ive since removed that connection)
I have another 40mhz which doesnt require the ncc if thats more what I need?
[quote author=Kamjin link=board=hardware;num=1094318646;start=0#5 date=09/05/04 at 00:51:56]Actually in the case of the sh4, it's Xtal input to PLL1 to divider to PLL2 which might be gated by a Clock pulse gen unit... to SDRAM
There actually hasn't been a "system bus" in ages, in the past the CPU/RAM/ROM clock was considered the system bus, since each component has it own bus today, and each can be clocked async.. well..
There's alot of myth out there... Â*even people who know what they're talking about, fall into the trap sometimes
for example.. what's wrong with this paragraph? when I saw this.. I thought if he only knew
"Anything else you want to do is up to how you want to design it, and what the device's requirements may be. For example, if one wanted to build a PCI bus that actually operated to spec (33MHz CLK minimum), one would need to add an oscillator for the 33MHz clock, and do some weird FIFO buffering to talk to the DC side of things. Not very simple."
Back to the topic.. the cpu to ram can't be the issue, mainly due the fact that it's actually working, since the sh4 controls the ram.. But what Quzar said was almost the same thing I though of first.. just off to the side a bit, one of my guesses is that the interface to the PVR might actually be having problems, and it's causing the slow down.. or the DC's kernal is actually smart enough to add wait states on the ram.. gotta do more spec reading before this makes sense though..
[/quote]
I was just thinking something like...ok the CPU checks the ram, and every I/O device it is attached to every... lets imagine 10 cycles. when everything is working at the clock speed that is specified, then the devices will be ready at that time. If its going slightly faster, then they wont be ready till the next 10 cycle iteration.
Just tell me if that idea is wrong, or just my original statement about system bus.
[quote author=quzar link=board=hardware;num=1094318646;start=0#7 date=09/05/04 at 02:35:56]
I was just thinking something like...ok the CPU checks the ram, and every I/O device it is attached to every... lets imagine 10 cycles. when everything is working at the clock speed that is specified, then the devices will be ready at that time. If its going slightly faster, then they wont be ready till the next 10 cycle iteration.
Just tell me if that idea is wrong, or just my original statement about system bus.
[/quote]
Neither is wrong.. just the sync problem was actually a 486-P1 era thing, and then it became a myth.. my mistake.. I forgot to mention that's what I was talking about..
actually the interface to the PVR, and ARM is what I was thinking of, with the exact same thought you were having.
There's two ways of interfacing them 1st where a device in the middle handles each device asynronously, and separatly, in this case a clock change won't make a difference, the other method is by holding off one device, until the signals align. This is where the problem might be happening.
PS: PCI devices can run from DC(0Hz)-33MHz
You've got it hooked up fine.. Now there's either some sort of alignment issue, or the DC was smart enough to play with clocks and wait states..Quote:
its a 40 mhz crystal. the inputs are nc/c 5v GND and output
I plced the output on the mb once i lifted pin 11 and the nccc on the chip pin which I lifted. (Ive since removed that connection)
I have another 40mhz which doesnt require the ncc if thats more what I need?
Or another thing, take a .1uF ceramic capacitor, and solder it to the xtal from gnd.
Also trying different xtals in the 34-40Mhz range.
Thanks,
looking at the webpage, (and some of my other DC motherboards). The actual chip number is different on the DC i modded. Its still labelled as the same component on the motherboard as my older DC (with the pipes). However this is a revision 1 mobo not a revision 0 which that is.
I however very much doubt that this is the issue. Kajaming does your DC have the pipes etc?
I tried disconnecting the crystal and the DC doesnt work. so that goes to show the crystal is connected and making a through signal.
I dont have the ceramic capacitor on it however. What differnce will that make? Ive also tried another different 40 mhz crystal(this one does not have a nc Â*on it (What is that btw?).
Is there someway that I might need to ensure that the pin is not 'arc'ing with the wire on the mobo and overriding the circuit?
Sorry for all the questions, im sooo close and I think this will be an interesting mod for all of the dc community, so Im trying to get it sorted.
Cheers
Q
[quote author=quzar link=board=hardware;num=1094318646;start=0#4 date=09/04/04 at 23:58:37]
There are many different things you can do to manually overclock a CPU idependantly of the FSB. Like I said this was mostly a problem back in the p2/p3 era.
[/quote]Indeed, but increasing the CPU speed independently of the FSB in a PC wouldn't cause the problem you were thinking of. Uh, not in recent times anyway, although even further back, just raising the multiplier shouldn't cause problems even with 486s (just given as an example of a way to increase clock without touching the bus... which is why I wouldn't directly compare OCing DC to PC). Anyway, if the issue is ever resolved maybe we can get a better guide (and in English too).
wonder ho hard it would be to make a dc into an atomisewave machine...
[quote author=qatmix link=board=hardware;num=1094318646;start=0#9 date=09/05/04 at 14:06:36]Thanks,
looking at the webpage, (and some of my other DC motherboards). The actual chip number is different on the DC i modded. Its still labelled as the same component on the motherboard as my older DC (with the pipes). However this is a revision 1 mobo not a revision 0 which that is.
I however very much doubt that this is the issue. Kajaming does your DC have the pipes etc?
[/quote]
The motherboard I'm looking at is the heat pipe version, it's actually dead though.. but you're connected in the right place.
That actually shows you're in the right spot.Quote:
I tried disconnecting the crystal and the DC doesnt work. so that goes to show the crystal is connected and making a through signal.
Wires have an inductance (a resistance to a change of current), the longer and thiner the wire then the more inductance it will have, also the higher the frequency the higher the inductance, so what happens is that the wires that feed the ground, and power to the xtal begin to act like resistors (actualy they'll start oscillating as well) The cap will act like a little reservoir for the xtal, that way you're sure the the signal it's creating isn't flawed. In reality it might be a little to low of a signal so the cpu isn't catching some of the ticks.Quote:
I dont have the ceramic capacitor on it however. What differnce will that make? Ive also tried another different 40 mhz crystal(this one does not have a nc Â*on it (What is that btw?).
The NC/Case is simply a contact to the metal case of the xtal module, you'd normally connect it to the sheild ground so it won't cause radio interference.
Doubtfull, since when you remove the xtal it stops working it proves that there's no contact.Quote:
Is there someway that I might need to ensure that the pin is not 'arc'ing with the wire on the mobo and overriding the circuit?
When I have some time I'm going to open up a DC and start to look at it as well, I'd say to try and get a hold of other values between 33-40Mhz and see how they change the behavior. in the end it just might be that the 40Mhz is causing the CPU some sort of internal error that's slowing it down..Quote:
Sorry for all the questions, im sooo close and I think this will be an interesting mod for all of the dc community, so Im trying to get it sorted.
Q
Unfortunately the multiplier is at the max, so we're kinda stuck in that sense. Although early press releases from Sega had the CPU clocked at 210 or 225Mhz so they must have been overclocking it when they were developing it.Quote:
Uh, not in recent times anyway, although even further back, just raising the multiplier shouldn't cause problems even with 486s (just given as an example of a way to increase clock without touching the bus
[quote author=qatmix link=board=hardware;num=1094318646;start=0#9 date=09/05/04 at 14:06:36]Thanks,
looking at the webpage, (and some of my other DC motherboards). The actual chip number is different on the DC i modded. Its still labelled as the same component on the motherboard as my older DC (with the pipes). However this is a revision 1 mobo not a revision 0 which that is.[/quote]
There were two different versions of the SH chip used in the dreamcast. About a year ago i bought an eBay lot of 22 DCs and parts. From them I learned much about the different revisions. Basically other than the different place and date differences, the main difference is that older chips say something like sh4400 or something, then the newer ones say the same thing, but with an R at the end. There are also two versions of the PVR chip.
hi
quzar wrote:
The diference in the SH4:Quote:
There were two different versions of the SH chip used in the dreamcast. About a year ago i bought an eBay lot of 22 DCs and parts. From them I learned much about the different revisions. Basically other than the different place and date differences, the main difference is that older chips say something like sh4400 or something, then the newer ones say the same thing, but with an R at the end. There are also two versions of the PVR chip.
- The sh4xxx is the original Hitachi, sh7750
- The sh4xxxR is the Renesas revision of the sh7750 named sh7750R
They are the same but Renesas owns it now
this helps ;)
-Mekanaizer-
Are you POSITIVE about that? When I asked Renesas they said that they really didnt know about that since they were specialty modifications of a chip.
Quote:
I am sorry to say, but all I can say is that this was a custom chip
that was done for Sega.
Sorry,
lyle
-----Original Message-----
From: Quzar [mailto:*******@yahoo.com]
Sent: Monday, December 08, 2003 5:37 PM
To: TechSupport
Subject: Re: FW: SH-4 chip identification.
Thank you very much. If you could also maybe tell me
any information about the differences between the
BP200 HD6417091 and BP200 HD6417091R chips? or maybe
what the HD number means? I would apreciate it greatly
if you could. Thank you.
~****** ******
They're saying that the SH4 used in all DCs is a custom version of the SH4. Has extra instructions over the standard SH7750. All DCs have those instructions, and I'd say its safe to say they're essentially identical.
Suprised he didn't tell you.. since it's just the actual company name
HD = Hitachi Denshi
64 = should denote the sh
1 = This used to denote a FLASH or memory option on the older products
7091 = The actual model of the cpu (7750 on normal sh4's)
BP = is the Package (256pin BGA)
200 = Speed
There's an R+S suffix I'm not 100% sure on that but it either denotes a shrink of the dye, or a technology change.
[quote author=Kamjin link=board=hardware;num=1094318646;start=0#12 date=09/05/04 at 16:18:05]Unfortunately the multiplier is at the max, so we're kinda stuck in that sense. Although early press releases from Sega had the CPU clocked at 210 or 225Mhz so they must have been overclocking it when they were developing it.[/quote]
I know, I was just giving an example relating to 486s. That's why I noted that you shouldn't directly compare PC overclocking to DC. The examples and info based on typical x86 boxes often don't apply.
[quote author=Kamjin link=board=hardware;num=1094318646;start=15#17 date=09/06/04 at 00:25:02]There's an R+S suffix I'm not 100% sure on that but it either denotes a shrink of the dye, or a technology change.[/quote]
I had the feeling that was the case. But I still wasn't sure. This could still be important in that it may tell us if certain DC processors give off less heat or such.
Just a quicky, I chould solder teh capacitor on teh gnd pin on the crytstal or just along the wire? Yup?
Also Icant find any othwer crystals apart from 40, 50 and 60 mhz. Anyone know where I can get other speeds?
[quote author=quzar link=board=hardware;num=1094318646;start=15#19 date=09/06/04 at 02:37:54]I had the feeling that was the case. But I still wasn't sure. This could still be important in that it may tell us if certain DC processors give off less heat or such.[/quote]More importantly, if any of them were produced in a smaller process, they may well have more room for overclocking. ;D
[quote author=Alexvrb link=board=hardware;num=1094318646;start=15#21 date=09/06/04 at 15:11:43]
More importantly, if any of them were produced in a smaller process, they may well have more room for overclocking. Â*;D[/quote]
yea that too, or maybe just run more efficiently aka faster =P. wishful thinking though
well atm I think this mod doesnt work as ive followed the instructions and im not having any luck.. although ive yet to add teh capacitor.
Kajamin, what do you mean by the alignmet might be wrong?
cheers
Q
[quote author=qatmix link=board=hardware;num=1094318646;start=15#20 date=09/06/04 at 04:29:58]Just a quicky, I chould solder teh capacitor on teh gnd pin on the crytstal or just along the wire? Yup?
[/quote]
Yup, just keep it close the the crystal.
It's kind of a rough guess, most probably it's wrong.. but you never know..Quote:
Kajamin, what do you mean by the alignmet might be wrong?
It's kinda like the planets.. not much control over it.
The idea is that none of the devices are actually compatible to each other, so there has to be an intermeadiate circuit to "translate" the signals from one device to another. For instance.. in order for the CPU to takeover the vram. it has to wait until it's not being used, it's not being refreshed etc. Now imagine that this time only happens every 3 ticks, when the cpu was at 33Mhz the circuit was designed to syncronise the 3 ticks to the cpu.. but since the cpu is running faster it's ready every 2 ticks so it misses 3 times before it finally get access on the 6th tick.. so effictively we've reduced it's access to the ram by 1/2 although we've sped it up by 20%
Like I said.. rough guess.. today most of this stuff is handled by an intelligent device that actually takes, and puts data regardless of each devices speeds etc..
SUCCES!
I think, Ive played Shenmue 2 and ran around the golden Qtr with NO slowdown at all!!! I have added a capacitor to the circuit between the %v and Gnd pin and it seems to have made a bit of a difference. I tried an emu but that didnt seem any differnt. But Shenmue 2 always slows down in the golden qtr, so it seems somethign is working (although it was late so Im going to have to double check later today)..
Still fingers crossed. Is it possible any of the coder types can knock up a basic cpu speed test, just a program which carries out a single operation multiple timeswithin a frame. Then as we know a dc is at 200mhz it can say e.g 10000 = 200mhz, tehn iff the dc calculates 15000 then it will know that the base spees should be 300 mhz for example?
Cheers
q
q if I throw you some money would you mind modifying one of mine on a weekend?
This sounds like an amazing thing for the dreamcast. Its really good somebody could work on making it faster cause i did find some games lag on dreamcast no matter if you buy for you the dreamcast has a slow down.
Now if this is correct that you have seen a difference for Shemue 2 which i believe would be a hard game to improve cause the graphics are so much more better then any game i have played for the DC (graphics i mean) and I believe this really would make a difference for other commercial games.
As for emulation i dont know what you guys can do to improve that besides the skills to get increased then we will see emulation and so on fullspeed and i think that might happen in a few years or so we will never know.
Eric
Again, I will double check it tonight. I can hardly believe it myself. Let me however note, it will not mean all games will run faster and smoother. Some games may slow down because of GPU overload. This Mod will nod fix that. Although, Im tempted to try overclockign that.
Kron, Ill PM you later for a chat.
Well here is a different view as i dont know what things can make the dreamcast faster cause upgrading one thing may down grade another. I still think its neat for speed ups for different games that need it but I wouldnt like to keep Overclocking and then put it back to normal just to play those other games again. Unless you have other dreamcasts thats pretty neat but for people who want this done i wouldn't recommend there was reasons why Sega kept it overclocked at the speed its at cause in someways it would downgrade some games with speed decreases and a possibility of other downgrades.
Eric
Eric... you don't quite understand. As far as commercial games go, it shouldn't cause anything to slow down any more than normal. So you don't need to reverse it unless some compatibility issue develops. Even then, you could just install a switch, so you don't need to crack it open. Also, if he successfully overclocks the PVR as well, it should remove slowdown from any commercial releases, not just ones that are CPU bound.
Yup, I can verify Shenmue 2 suffers from no slowdown :) Also I tried DC quake and it didnt drop a frame. Im going to do some fps tests in teh console to see if I can gauge any speed up tests.
Also I tried the grinch and it sometimes skips up to 60 fps and he runs in a crazy speed up manor!! Â*However it still runs at the std speed most of the time.
Im waiting for some devver to offer to help me with a speed test program. (Anyone)?
BTW alexvb - Speeding up teh PVR wmay help, but again slowdown may be because its drawign twice as much as the PVR can handle, And I doubt I can overclock it that much :) although you are right, it will hopefully speed up more titles (Ive just not had time to test that many atm, work and personal stuff is more important).
Next up is PVR and Im going to chuck in a higher crystal on teh CPU as im not gettign any overheating yet (Im runnign at 240 MHZ).
that aint to bad what kind of things need 40 mhz to run better and again any good for emulation i forget?
Mame DC locks up. Neomame seems no quicker! maybe its locked with the PVR side of things.
[quote author=qatmix link=board=hardware;num=1094318646;start=30#33 date=09/10/04 at 04:18:31]BTW alexvb - Speeding up teh PVR wmay help, but again slowdown may be because its drawign twice as much as the PVR can handle, And I doubt I can overclock it that much :)[/quote]Well again, I was referring to commercial releases. I don't know why homebrew is reacting like it is. But I doubt there are any games that dish out "twice as much" as the PVR can handle. Think about it, would you release a game that was riddled with constant and excessive slowdown caused by trying to render TWICE as many polygons/effects as your graphics engine can handle? I find it more likely that slowdown that isn't solved by overclocking the SH4, could be resolved by overclocking the PVR a bit as well. Unless they just really botched it ;)
In general acessing the PVR as a "dumb" frambuffer is to put it lightly.. dirt slow to begin with, when I create a test bit of code on the PC.. well my ancient p166 laptop and use a Vesa 16bpp flat mode.. I even make the code nice.. and slow.. and portable.. it's actually able to run circles around the dreamcast in speed.
So I guess the effect is dirt slow, to dog slow..
qatmix, give some of the SNES emuls a try, they might actually gain a bit of speed.
Try neogeo cd with metal slug 2 tell me what happens.
what do you type in teh console in quake to do a FPS check with the demos?
Ive tried Dreamsnes V 0.98 and its much smother (about another 25%) although I dont know how to get fps displayed. I had it running side by side with a std dc and the extra smoothnes was apparent.
does that mean more playable?