ASRock 775Dual-VSTA bios, mod, recap?

[trodas]

This is another 775 mobo I get for the sole purpose of fun and pushing the AGP limit(s) with some serious CPU's. The mobo does support not only AGP, but also PCIe and not only DDR(1) rams, but DDR2 as well:

 

http://hwbot.org/hardware/motherboard/775dual_vsta/

 

So, it is a remarkable piece of hardware Also after updating the bios to latest official P3.10 version ( http://www.asrock.com/mb/VIA/775Dual-VSTA/ ), there are plenty of bios options to play. More that I imagined:

 

 

And that is not all! There are existing modified bioses ( http://pctreiber.net/asrock-bios-downloads?did=103 ) that since v3.10a give the possibility to boost the maximum supported ram from 2G (2x1G, you cannot use DDR and DDR2 rams at once, lol) to 4G (2x2G sticks). Current unofficial bios version is 3.19a.

 

Of course the really available ram on 32bit systems are something around 3.3GBy (all the buffers and your GFX card memory must fit into 4G address space), but it is still a good leap forward, compared to being stuck with "just 2G of ram."

 

What attract me on the board is 4 phase regulator and good overclockability. When the poor Celeron D 336 on ASRock 775i65G R3.0 hit the wall at FSB 154 ( http://valid.canardpc.com/lregg7 ), the same CPU is not having problems at FSB 164 with this mobo ( http://valid.canardpc.com/rnd2tg ) and when pushed beyond stability, then 175MHz FSB is possible: http://valid.canardpc.com/fxw94m

 

Sure, sure, 3677MHz from 2.8GHz CPU is nothing spectacular, but w/o incrasing the Vcore it is not actually bad And it looks like that incrasing Vcore could be done by just replacing one SMD resistor with zero resistance to another SMD resistor: http://www.controsensi.it/Mods/4CoreDual-SATA2/Contents_Eng.htm

 

That is quite good and what it even better is, that near the CPU socket, there are place for 4 (!) stolen CAPS for Vcore regulator! That empty places could get very nice polymer (and small-enought to fit under the heatsink mount) caps. And on the bottom side, there is place just under the CPU for two serious tantal-polymer caps :nana:

 

So I see a big potencial in this mobo. Just if I understand the settings a bit...

 

 

PS. flashing bios is a "little" nightmare, the mobo refuse to boot from CD I create (all other computers do boot), USB flash... and when it finally boot from floppy, not even uniflash was able to flash the bios...! So beware, you really need to use the Windows flasher and NOT latest version but v4.xx (v4.48 from there works: https://www.wimsbios.com/amiflasher.jsp ) as mentioned in this guide (page 2): http://pctreiber.net/2006/howto-biosflash-asrock-motherboards.html

For future reference, there is the latest 3.19a bios WITH the flashing tool:

http://depositfiles.com/files/9ran5etgd

http://www.mediafire.com/?x4518tmwc312wsg

https://mega.co.nz/#!PYMX1CaB!p7lk1NkWWSCv86ykDNRYwvWM4mTDVWh0UYbt2nEnsdU

Edited September 9, 2015 by trodas
added bios flashing info - it was HELL!

[dx4picco]

lol crazy MB, great compatibility

[trodas]

...and yes, it can take on even the Core 2 Duo E8600 "Wolfsdale" CPU, as visible on this score:

http://hwbot.org/submission/2322337_matti_oc_3dmark2001_se_geforce_7800_gs_48003_marks

 

That CPU did not have unlocked multiplier like my Core 2 Duo Extreme X6800, but it feature 45nm process, less TDP (65W as oposed to 75W) and higher overclockability. On top of that, 6MB L2 cache really counts, compared to "just" 4MB on X6800 Overal it could reach 5GHz when in luck, average O/C is 4.6GHz:

http://hwbot.org/hardware/processor/core_2_e8600_3.33ghz/

(me getting bad ideas )

 

All-in-all, it is rather interesting board and we see how well it will overclock. So far I only have a little troubles with it, probably related to previous used bad behaviour: there is one pin on the CPU socket with added melted ball on it. Everything works, but it is just weird and hopefully other CPU's (beside the tested Celeron D) will work well too. And I cannot fully push down the 24pin ATX PSU connector into the ATX 20 pin. There is visually nothing that get in the way, but somehow at the end of the connector (neat the CPU) it just leave about 1,5mm from the fully seated connector. Could be because of using 24pin PSU connector and the need to cut a bit the plastic at the end, I check this later to see.

 

Other than this I just could not wait to get good caps to the mobo, mod it (Vcore increase) and see, how far it will go with the overclock.

 

Also I try to persuade guy, who made a ASRock Conroe885 bioses with different memory dividers that the default ones (like 5:4 and 1:1 forced settings) to try something like that for the ASRock 775i65G ... and when possible for the VIA PT 880 Pro chipset, then for this mobo as well.

Dunno, how if this old chipset could handle for example 400MHz on the DDR2 rams (PC 8000 with 2:3 divider) ... but if it can, then I'm sure that some records will be broken :nana:

Edited July 27, 2015 by trodas

[ObscureParadox_3XS]

The E8600 won't really have very good overclockability on that board though simply because you will be FSB limited very quickly. The stock FSB on the CPU is 333MHz and you'll be lucky to get 350FSB realistically. So if you have an X6800 it's probably better for you to use that if you can.

 

The absolute best CPU I have seen for that board is the QX9650 which is a quad core 45nm CPU. That will give you the most performance out of your system by far. BIOS mod may be needed but not 100% sure on that, all I know is that there are plenty of people using that CPU on that motherboard.

 

There are a few DX10 agp cards that will benefit from having the extra cores when running vantage too

[Massman]

Love those Asrock boards ... I wrote an article on AGP overclocking with one of the variants like 7 years ago (OMG, can't believe I'm getting old) http://www.madshrimps.be/articles/article/670/Tweaking-the-fastest-AGP-Based-System-Part-1#axzz3f9tSCfo7

 

Did some theoretical stuff on memory bandwidth in that article too. Reading through it again, it appears memory bandwidth was the key. I can't remember messing around too much with dividers other than 1:1 because of instability and compatibility issues.

[trodas]

ObscureParadox -

The E8600 won't really have very good overclockability on that board though simply because you will be FSB limited very quickly.

 

We see. It is still not there, but Noctua NH-C14S with the supporting kit for 775 are ready. I like the 6MB L2 cache, so I hope for some reall power there But yes, x10 multi is low. E5700 with x15 multi is probably better choice, it will reach far higher clocks, hence at least for SuperPi 1M it will be faster. Likely.

We see. I will report.

 

So if you have an X6800 it's probably better for you to use that if you can.

 

Yep, have one Extreme X6800 ready to roll there

 

best CPU I have seen for that board is the QX9650 which is a quad core 45nm CPU

 

Updated to the moded bios that support even 4G ram sticks and this quad core, but that CPU is probably not going to help me to get higher scores in anything besides Cinebench or other multihreading benches.

 

I did not fancy 4 cores for benching also because it is power hungry and I did not have faith in the caps on CPU VRM there either.

 

There are a few DX10 agp cards that will benefit from having the extra cores when running vantage too

 

Now you are thinking big

 

 

 

Massman -

Love those Asrock boards ... I wrote an article on AGP overclocking with one of the variants like 7 years ago (OMG, can't believe I'm getting old)

 

I *LOVE* your article. You replaced the caps to quality ones, you mod the board bad (completely new Vdimm regulator - you are good!) and you did plenty of interesting testings to compare the speed.

Fantastic reading even 7 years after!

 

Reading through it again, it appears memory bandwidth was the key. I can't remember messing around too much with dividers other than 1:1 because of instability and compatibility issues.

 

Depending on CPU and ram used, the board set crazy FSB:mem dividers. Sometimes things are even running notably slower on higher ram clock...

I took from your article that setting Pipeline DRQCTL on does increase the memory bus bandwitch. Thanks!

 

 

So there is my contribution to the ASRock 775Dual-VSTA. First - ram voltages on the ASRock 775Dual-VSTA:

DDR1: 2.53 - 2.63 - 2.73

DDR2: 1.85 - 1.95 - 2.05

(low - auto/normal - high)

Not very optimistic, right? The top limit is very low. IIRC all fastest 2-2-2-5 DDR1 rams require at least 2.8V, if not more for mentioned timings. Also most DDR2 rams would like at least 2.1, better 2.2 or 2.3V...

 

Second - I got G.Skill DDR2 800 (F2-6400CL4D-2GBPK) rams. They have s/n 152692002354233, so according to infos there:

http://www.xtremesystems.org/forums/showthread.php?283666-Figuring-out-G-Skill-s-SNs

...they are made in 26 week 2015 (on the hologram logo it says 2015 Jul anyway), 9200 should mean "Powerchip/PSC (sometimes eTT/uTT)" chips used and the 33 on the end is undefined, most close is "xx30 = DDR2-800 5-5-5" ... and it makes sense, because the ram is rated DDR2 800 4-4-4-12 (1.9-2.0V).

 

With some finetuning with drive settings I managed them to work stable at 3-3-3-10 at 262MHz, where they provide a *BIT* better results that 197MHz DDR1 with 3-3-2 timings. That IMHO terribly suxx.

 

For example default memtest MB/sec are 1772 for the Quimonda DDR1 rams and I pushed them to 2370 MB/sec (5:6 FSB:ram divider):

 

CAS - 3

DRAM Bank Interleave - 4 Way

TRCD - 3

TRP - 2

TRAS - 8

TRFC - 9

TRRD - 2

TRTP - 2

TWTR - 1

TWR - 2

DRAM Bus Selection - Dual Channel

DRAM Command Rate - 1T

 

DDR2 G.Skills started at promising 2243 MB/sec on default settings, witch give a 5:8 FSB:ram divider (533MHz ram settings), but fall down to 1849 MB/sec on the 667MHz setting (1:2 FSB:ram divider) when running at 5-5-5-13 settings at 328MHz.

Suxxx!

 

So I tried and failed and tried till I get to 2439 MB/sec on these settings stable:

 

CAS - 3

DRAM Bank Interleave - 8 Way (Hoooray!)

TRCD - 3

TRP - 3

TRAS - 10

TRFC - 28

TRRD - 3

TRTP - 2

TWTR - 2

TWR - 3

DRAM Bus Selection - Dual Channel

DRAM Command Rate - 1T

 

It required to change these to manual settings:

DRAM Drv DQS

DRAM Drv CS

DRAM Drv DCLKA,B

DCLK Output Duty Ctrl

 

...and give all the drive setting 44. (FF crashed, 00 give errors too fast, tried 22, 33 and 44 then, 44 works)

 

Changed these:

CLKBUF DDRB group delay

CLKBUF FB_OUTB delay

...to 1x50ps delays (it should IMHO add delay for signals to the futher away dimm from CPU, but not sure about it

 

Also changed both last x4:

CLKBUF FB_OUTA Drv

CLKBUF FB_OUTA Drv

...to x3 drive. Not entierly sure what I was doing, but in the end it worked. There are absolutely minimal gains, tough. Like 48:703sec ( http://hwbot.org/submission/2915891_ ) vs 48:187sec ( http://hwbot.org/submission/2923748_ ) in SuperPi 1M test, 45:32:531sec ( http://hwbot.org/submission/2915892_ ) vs 44:54:203sec ( http://hwbot.org/submission/2924350_ ) in SuperPi 32M test and so on. Quite a disapointment. Some higher gains that expected happen in Cinebenchs, but that it is. Unless the Celeron D 336 crap CPU is the limiting factor there, the DDR2 performace is shockingly bad.

 

I mean... they need for 2439MB/s to run at 262MHz http://valid.canardpc.com/szgjvh but a DDR1 rams need only 197MHz to get 2370MB/s! http://valid.canardpc.com/rnd2tg

 

If I get any nice 2-2-2-5 DDR1 rams that can run at the top 2.73V voltage there, they are likely to outperform these DDR2 rams... is not that sad?

 

 

...

 

 

PS. later I found, that just changing these two special settings are need to get the stability for 32M SuperPi:

CLKBUF DDRB group delay - 2x50ps

CLKBUF FB_OUTB delay - 2x50ps

 

And nothign else in the advanced settings. Ugh!

Edited September 8, 2015 by trodas
added info about the wrong 1x50ps - need 2x50ps to be fine

[Rasparthe]

If you are looking for RAM that operates at 2-2-2-5 at those voltages look for Samsung TCCD sticks. They should be able to do those timings at that voltage. Those boards and their variants are absolutely horrible when it comes to RAM, no matter how much tweaking I've done with the settings in the BIOS it seems to make almost no difference and sometimes actually hurts stability.

 

Incredibly fun board though, if you really want to see what those boards can do you should follow the Team Cup stage that will feature a good many of those....

 

http://oc-esports.io/#!/round/team_cup_2015_sc2/2369/3dmark03_fastest_agp

[trodas]

Well, it is probably a time to get some good TCCD modules (sadly they are only 2x512MB, bot 2x1G) for comparing purposes. The OCZ (OCZ4002048ELDCPE-K) rams I have can do 2-3-2-5 at 200MHz with 2.50V, but still not 2-2-2. Another kit of 2x1G from OCZ I have (OCZP4001G) is also capable of 2-3-2-5 at up to 216MHz at 2.75V (some timings on AMD64 from it: http://postimg.org/image/6c06gn1t3/ ), but once again no 2-2-2. Only the Mushkin BH-6 Hi Performance ones I have are capable of 2-2-2 timings, however they also need 2.80V to be happy. I did not tried them yet in the ASRock 775Dual-VSTA to find out, what I can get from them at 2.73V in fear, that at normal settings (2.63V) they might refuse to post...

 

Another question is, if the DDR2 rams could get faster, when I get some good set of Crucial Ballistix 1066 16FD5. These should be capable of 3-2-2-8 1T timings near 300MHz, so at 262MHz it should not be a problem to reach this timings + fastest possible secondary timings. That way they can possibly beat the G.Skill DDR2 800 (F2-6400CL4D-2GBPK) rams and finally show some benefit for the whole DDR2 "thing."

 

Also is, IMHO, a good question, how other, that the lame old slow Celeron D 336, CPU's will react on combination of DDR1 / DDR2 rams. I did not have whole bunch of Socket 775 CPU's, but they are dirt cheap to get (some might be even donated for the test) and even ATM I got few, that are worth testing:

 

533MHz FSB Celeron D 336 (DDR1 5:6, DDR2 5:8 or 1:2)

800MHz FSB Pentium 4 650 ???

1066MHz Core 2 Duo Extreme X6800 ???

1333Mhz Core 2 Duo E8600 (in shipping) ???

 

To put it short - I have no idea, what dividers (FSB:ram) are possible with different CPU's and how it will affect the resulting speed. I have to agree, that except for the CPU clock, the memory bandwitch is everything there. Yet different CPU's might react differently and with different FSB's, they will get different memory dividers, hence the results might get quite interesting.

I did not yet even tried to set another ram options for the DDR1 rams, that the 200MHz DDR 400 setting.

 

As for the impact of settings and stability - well, almost all the settings cause some diferences. If the difference are not shown in the Memtest benchmark, it does not mean that it is not there and it will not reveal itself on another bench, like SuperPi.

 

Despite the mainboard have 4 stolen caps from CPU VRM and not a single bottom sided under CPU socket cap (where are place for two of them) and I have not do yet any mods to it, it run very stable and reliable for me. With one exception - freezes using the Radeon 9600 XT graphic card. But the card is slow and it is full of known bad caps, so before full caps replacement I won't comment much on stability. SuperPi 32M is stable and that it what counts ATM.

 

Team Cup stage - I see. But I did not have yet any good AGP card, so I could join the fun, not to mention E8600 did not yet made it as well, as I need more decent rams Also is quite possible, that CPU clock wins (even with 2MB cache ever 6MB on E8600) and a E5700 get to be better solution. After all, E8600 is 333x10, but E5700 is 200x15 - eg. the clock go up much faster when overclocking, so speed might be better with E5700, as it is likely to reach more MHz.

[lanbonden]

As a fellow AGP bencher, my tip is to just make it simple for yourself and do a Vmod for the memory, it should be well documenteted already for that motherboard so not even any reasearch needed.

 

Ive been benching like 10 agp cards so far on a Asrock AM2NF3-VSTA, the AM2 equivalent to the motherboard you got that supports ddr2, Phenom2 quads and AGP and my experiance so far is that memory dont really mater with the slower cards atleast.

With that said, my card was also limited to ~2.03V for DDR2 at ultra high settings so pretty much first thing was to Vmod that part. Other then that is the problem an extreamly weak powersupply for the CPU, Max Vcore is 1.55V but over 1.475V gives me random crashes, the motherboard also give me ~200Mhz lower OC then a decent board at same measured voltage with the same CPU.

[trodas]

Well, I would *love* to made a voltmod, but since sourcing components will take me easily half year or so (just check this Mouser order, made 15.6.2015: http://cz.mouser.com/OrderHistory/OrderDetail.aspx?qs=9C%2fvjDbyW3V1fmZ4G2SNFokJWM6STzNO and expecting to wait another 4 months to delivery even the basic parts for the recap...!), then Vdimm mod is out of the question ATM.

 

Simply cannot be done w/o the much need components. That it is.

 

AMD did not offer comparable speed to Core 2 Duo CPU's, so I took the Intel conterpart there. I doubt that rams did not affect the score, even slight secondary timings change 3DMark01 results for me. So to get very best of the mobo, I need rams that are capable of tight timings (when I'm limited by the FSB reachable).

I also noticed that many took the wrong approach and tried the higher ram clocks, but they SUXX, because even that they are higher, the chipset turn into 2T mode (not changable) and results suxx:

 

DDR2 G.Skills give 2243 MB/sec on default (5:8, 262MHz), but fall down to 1849 MB/sec on the 667MHz setting (1:2, 328MHz).

 

That is where most people fail and their results can be easily beated. Very easily I already tried some preliminary tests and they are "blown away"

The weak Vcore regulator can be greatly improved by using quality caps and adding the missing caps. There is no question about it. And same there, ASRock 775Dual-VSTA does not let me to change Vcore. I have to mod it. Fortunetely, there is a VERY simple offet mod there: http://forums.vr-zone.com/hardware-depot/192571-volt-mod-done-right-asrock-775dual-vsta-l6714.html

Just replace the "0" resistor with higher one. How much higher Vcore depend on this formula:

Current CPU Vcore x (1 + 3 000/resistor value in ohms) = resulting CPU core.

 

So a 43k resistor gives +6.9% Vcore, 36k resistor gives +8.3% Vcore and 20k resistor could give +15% Vcore, but not stocked. 22k (RT0603BRD0722KL) is available and give serious +13.6% boost of Vcore. But waiting for such components is tiresome...

 

W/O quality components is the board as good, as dead. With them, I believe, it will shine

 

BTW, your board Vcore regulator has 4 phases and looks relatively good: http://forum.hwbot.org/showthread.php?p=398973 So it must be the caps that bring it down with over 1.475Vcore...

 

 

PS. preparing ASRock 775Dual-VSTA to voltage measurments in order to determine best caps replacements:

See the stolen white caps near the CPU? That is the key... :-) Also from bottom, there is place for two tantal-polymer 470uF 2.5V supercaps. Expect results that blow everyone away (hopefully, unless I'm mistaken badly)

 

On the top of that, the board accept Intel Q6600 with the latest 3rd party bios. That will kick all of the scores, that are ready for more threads (wPrime 32M, wPrime 1024M, PCMark Vantage, 3DMark Vantage - Performance, Cinebench 2003, Cinebench 11.5, Cinebench 15, HWbot prime and possibly PCMark 2005, 3DMark 05 and 3DMark 06) and in that case, any better timing on the rams WILL play a huge role, as the memory bandwitch will become the limiting factor there.

 

Just look what Massman says above:

it appears memory bandwidth was the key

Now I respect him greatly for his work and if HE determined, that the memory bandwith is the KEY - who I'm to doubt that? If he is selling these ASRock boards or memories, then I will be dubious (sales pitch), but as far, as I know, he have no compeling reason to distort his findings in any way, shape or form and if I have two memory-bandwitch hungry cores in my E8600 that arrived, then you can bet your arse that increasing timing by any primary setting cause great slow-down in the tests that required memory.

And most of them does.

 

Now imagine how 4 hungry cores in Q6600 want! You bet it - memory bandwitch or death! So reaching 3-2-2-8 timings with DDR2, or 2-2-2-5 with DDR1 (witchever will be faster) is, IMHO, imperative to get best result. But it will be the test results that will either prove or disprove what I claim

 

And a little hint you see above, that there will be hardly any CPU-memory related score that remain untouched by me... Sure, some reached 4.8GHz ( http://hwbot.org/submission/2349103_p.a.s.s.a.t_cpu_frequency_pentium_e5800_%283.2ghz%29_4802.19_mhz ), but not with seriously powerfull CPU E5800 is good (x16 multi) for overclock, but IMHO lack the cache to compete in CPU benches... but I might be proven wrong soon

However using 187.6MHz (8:5) DDR1 rams at 3-3-3-7 timings determined that there will be not much powerfull results anyway.

 

 

PS: interesting thread about zombie mod of ASRock 775Dual-VSTA: http://kingpincooling.com/forum/showthread.php?t=2545

Edited August 18, 2015 by trodas