Printable View
Playing around with 8rdavcore, I noticed something I missed on this board, AGP latency. On my VIA boards, AGP latency is set to 248, this Infinity has it set to 32 according to 8rdavcore.
I haven't thoroughly tested it yet, but bumping the latency to 96 appears to have made my 3dmark runs much more reliable, even at high FSB.
If this is true, perhaps nVidia cards like owning the bus for longer periods of time than do the ATI cards.
You bet I'll be playing with this over the next few days :)
How big of s difference has it made? any screens? Very interesting. :)
i can almost predict another WPCREDIT trick? anyone?
LOL. I've been pulling out my hair trying to find out why nvidia owners can run 3dmark at higher FSB speeds then Radeon users. I downloaded the new 8rdavcore yesterday and noticed the exact same thing concerning the AGP latency. I increased mine to 64 and gave me a tremendous difference in running 3dmark at 244fsb. I haven't had a single kickback since yesterday. I wasn't going to say anything cause I haven't really got around to making my system stable at 244.
I thought about emailing OSkAR_WU over at the Amdmb forums to see if he could some how throw in the option of increasing the latencies in a beta bios to try. I will do just that and give full credit to sjohnson.
Great find.:toast:
Well its definitely helping me. After benching and gaming stable all day yesterday at 244fsb agp set to 64, I increased it to 96 and ran my first complete 3dmark bench at 252fsb.
http://service.futuremark.com/compare?2k1=7465141
Just ignore 9600XT, should be getting my replacement 9800 pro in this week. But I will say thats not a bad score for a no OC card.
Setting the latency to 248 not only gives me greater 3dmark stability but adds 40-50 points to my runs as well :D
My current high, Siluro Ti4200/64M: http://service.futuremark.com/compare?2k1=7464449
The nice things about that score is that I did it in one straight-through run instead of coaxing it out, it's my current highest, and *it's repeatable*, no kicking to the desktop anymore :D
RacerX - no credit needed, just get the word out so more folks can try this, we need more testing...
Very nice catch.
guys do the vcore mod as well even if you don't plan to run over 2.0v. I did one thinking for some reason this board is undervolting big time so in the bios when I select 1.9v=1.96v with mod.
Just something to confuse us more. I cannot get a true voltage reading of the vcore on this board cause I covered the mosfet with heatsinks. MBM still shows me a huge fluctuation pre-mod but system overall is way more stable. I do not need to raise the agp latency when running 245-250fsb in 3dmark.
More voltage helped me :D
Check the sig 242 FSB 1:1 publishable with a 9800 PRO :toast: . Not a bad jump from 224 before. "TOP :banana: "
After reading this, i downloaded 8rdaVCore, and im going to play with it today. Set AGP Controller Latency to 96, see if it wil help me at all.
I am running a Soltek 75FRN2RL, and currently hold the second highest score for a Palomino Athlon XP :P (Getting a 2500+ in a couple days) Hopefully this will help me take the highest score.
I'll post back with my findings
Hi
By setting the latency to 96, I am able to reach 19,798 3dmarks.
Still not breaking 20k, but more than 200 pts then pre 8rdavcore, and more than 400 pts since my NF7-S rev 2.0 days.
Eh? your at 245 x 10 with a 9800 PRO and you can't break 20K? :confused:
I got 19,870 with these settings:
Abit NF7-S V2.0 @224 MHz FSB x 10 || 2x256MB Muskin DDR3500 Level II 1:1 Dual 6/2/2/CAS2.0 || AXP2500+ "Barton" AQZFA - 2246 MHz @1.75v || Sapphire 9800 PRO @420.4/378.0
Whats your raddy at? You benchin on win95? :stick:
Great find guys:toast:
I'll give this a shot on the Infinity when I switch out the A64 test bed in the Overclocking Competition.:)
Randi:D
I got 19500 with a 9500np at 470/335
and 11x219 with a TbredB
lght1 should be getting higher.
[Edit]I haven't yet found the register or registers set by 8rdavcore.
When I set 32, 96, 248, or any value using 8rdavcore, WPCREDIT shows offset OD as being set at F8. So I haven't found the register or registers being set by 8rdavcore.[/Edit]
[wrong]Here's a WPCREDIT for those who'd rather use it and WPCRSET.
The register in question is at offset 0D in device 0253, VGA Compatible. Not sure if the device number (0253) is the same across all Infinity boards or not. But, the VGA Compatible device is the same for any board I've ever come across, as is the offset 0D.
This SS shows my board with the AGP latency register set to 248. You'll need to think in hex to use it, but those used to H. Oda's tools already know that.[/wrong]
Cheers!
Nice job again sjohnson !
Why all theses tricks come after the enthousisame for NF7-S ?
Bad Dog 'ABIT' ! Bad Dog ! No More money ! No More Credits ! :mad: => Going to DFI soon
just ordered myself a DFI Ultra Infinity can't wait to get it seems like a very fun board to have. Nice job again sjohnson
great find sjohnson :toast: If I can get my Infinity to boot (outrageous memory setting :p: ) I give it a try.
Sigh - you can't always mix kids with your hobby. I should have double-checked my WPCREDIT finding. Sorry folks, that SS above does not show the right register - either that or 8rdavcore is setting something else.
With 8rdavcore showing 32, 96, 248, *anything*, WPCREDIT shows offset OD as being set at F8. So I haven't found the register or registers being set by 8rdavcore.
very very interesting... on my system, stock delay of 32 @ 250 x 9.5 = no go...
change to delay of 96, no go, crash first test...
change to 248, completes all tests like it was a walk in the park... I dont know what to say... this owns :toast:
wow, gotta try this!!!!!! :DQuote:
Originally posted by Snoodge
very very interesting... on my system, stock delay of 32 @ 250 x 9.5 = no go...
change to delay of 96, no go, crash first test...
change to 248, completes all tests like it was a walk in the park... I dont know what to say... this owns :toast:
What about Side Band Adressing?
should we try disable that? would it make it slooooow?
Heh - I tried disabling SBA as a test. My system crashed within seconds.
Generally, disabled SBA means less 3dmark anyway so no tears here.
Here's a bit more on what changing AGP latency is doing on my rig.
Using 8rdavcore to set the latency seems to have set a value or values in the CMOS register(s) that have increased 3d stability, even at a latency of 32, like after a reboot. Else, my system is "bedding in" and is simply more stable than before.
Setting a high latency hasn't eliminated kicks to the desktop. I *am* at my current CPU and memory limits so that might be the cause of kicks to desktop but overall, it just "feels" like it's better - but not *perfect* with higher latency set.
Hi,
if somebody wants to know which register is set, why not asking me? :D
well please let us know :) Great program by the way love it.Quote:
Originally posted by Hasw
Hi,
if somebody wants to know which register is set, why not asking me? :D
Ive been testing this board out a lot also and i have found a few interesting results. Im a linux user, and in linux every single game test and system test i have done is exactly on par if not marginally better than my nf7-s, nf7-m, and 8rda. Now i havent compared all of the boards in windows, but ill be doing that shortly
Well, it's bus 0 device 30 function 0, the PCI-AGP bridge. Register 0x0d (PCI latency timer). This is the latency for the PCI side (to internal PCI bus) of the AGP bridge.
Thanx Hasw i'll try it out later
I thought I looked at that device, but must have missed it. Thank you Hasw...Quote:
Originally posted by Hasw
Well, it's bus 0 device 30 function 0, the PCI-AGP bridge. Register 0x0d (PCI latency timer). This is the latency for the PCI side (to internal PCI bus) of the AGP bridge.
Great program, 8rdavcore, thanks much for that, as well.
Well with the following specs, I took the number one score on the ORB for Palominos (Aircooled)
Soltek 75-FRN2-RL w/ vcore mod
Athlon XP 2000+ (Pally) @ 1.93 1.95v
2x 256 Twinmos PC3200 w/CH-5 @ 214 fsb
Powercolor Radeon 9800 Pro @ 450/400 no mods
I went from 17307 to almost 17600 by simply bumping the pci latency to 64. Havent played with the AGP *YET*
One things that seems odd, is that with my CH-5 I am able to run 2-2-2-11 at stock voltages, at 214 fsb, something that I thought wasnt possible with my mem... maybe there is more to this latency thing than we thought.
Will post back with more results as soon as I get them
theres been a few sticks of ch-5 coming out of the woodwork lately that will run 2-2-2Quote:
Originally posted by SWGalvatron
Well with the following specs, I took the number one score on the ORB for Palominos (Aircooled)
Soltek 75-FRN2-RL w/ vcore mod
Athlon XP 2000+ (Pally) @ 1.93 1.95v
2x 256 Twinmos PC3200 w/CH-5 @ 214 fsb
Powercolor Radeon 9800 Pro @ 450/400 no mods
I went from 17307 to almost 17600 by simply bumping the pci latency to 64. Havent played with the AGP *YET*
One things that seems odd, is that with my CH-5 I am able to run 2-2-2-11 at stock voltages, at 214 fsb, something that I thought wasnt possible with my mem... maybe there is more to this latency thing than we thought.
Will post back with more results as soon as I get them
damn if someone would have only discovered this 2 weeks ago when i bought another nf7 to replace a §§§§ty one i had before...
the reason i chose another nf7-s was since the infinity was slower in 3d
That's the weird thing, these havent JUST come out of the woodwork. I bought these when CH-5 just became available. :PQuote:
theres been a few sticks of ch-5 coming out of the woodwork lately that will run 2-2-2
I just may be able to break 18k with this setup if I keep messin with this latency thing. This definately would have been a HUGE bonus had it been discovered a couple months ago. Hell, considering the FSB people are achieving now, it might have helped keep the Athlon XPs in the top of the game.
Oh well, buying a 2500+, then moving on to S939 A64 FX-53, its so hard waiting for the Socket 939... grrrr
Why would a higher latency give you better performance?
I can see how it might be more stable, but performance gains?
Not that I dont beleive ya all. but why?
The AGP latency is different than memory latency. The higher the number, the longer the AGP port is able to control the PCI bus before it has to give it up to other processes. The higher the latency, the more priority the AGP has over other devices on the PCI bus.
Incidently, raising the AGP Latency has done nothing to my stability.
I can Prime and Memtest86 for hours without an error at 248FSB, but I can't run 3dMark at anything above 237
My board had a different problem - it was kicking to the desktop even at stock, when new. The problem was slowly getting better, probably due to break-in and the system tuning I was doing.
The 1-21 BIOS works identically for me as manually setting the AGP latency in 8rdavcore. In fact, for 3dmark tuning, you should find an AGP latency that gives the best score for your system. The stock setting (0xFF) on the 1-21 BIOS might be overkill.
You can also play with AGP and the other latency values to get the overall best performance for your system. On the ECS K7S5A board I used to run, I had to play around with the latency settings in order to find the values that let my rig run without sound drop-outs.
The new Infinity BIOS sets the AGP latency to 255 for those of you interested.
I've tried both the new bios and the 8rdavcore . . . no change, no differenceQuote:
Originally posted by sjohnson
My board had a different problem - it was kicking to the desktop even at stock, when new. The problem was slowly getting better, probably due to break-in and the system tuning I was doing.
The 1-21 BIOS works identically for me as manually setting the AGP latency in 8rdavcore. In fact, for 3dmark tuning, you should find an AGP latency that gives the best score for your system. The stock setting (0xFF) on the 1-21 BIOS might be overkill.
You can also play with AGP and the other latency values to get the overall best performance for your system. On the ECS K7S5A board I used to run, I had to play around with the latency settings in order to find the values that let my rig run without sound drop-outs.
what would the best value be for a NF7? 96? 128?
Test it yourself, who knows