Kingston valueram PC2700 using Hynix D43 chips, same chip used by PC4000 modules

[yeba]

I picked up a kingston valueram PC2700 512MB in CC using Hynix D43 chips. These chips are suppose to be the same chips that Buffalo PC4300 and Corsair XMS PC4000 use.

Right now, I can only run at DDR450 cas 2.5 stable @ 2.8V. My mobo px845pev pro , is limiting me to CAS 2.5.

Most of the PC4000 modules are rated at cas 3 ~ 2.6-2.7V.


Has anyone reached PC4000 and up with these modules at CAS 3.

I need to know so I can buy more memory with this chip and also replace my mobo with a 865pe that can do CAS 3.

Thanks.

[Merc Zephyr]

Hello Yeba

I recently purchased 2x512 of this same memory just to see if it could be a cheap way to approach pc4000 speeds. Tomorrow I will be purchasing a 2.6c or a 2.8c and hopefully in a couple days I can post some results here regarding this memory. My current setup only allows me to get the ram to 200mhz where the timings were 2.5-3-3-6 at auto dram voltage.

I don't think the hynix d43 on this ram is the exact same as ocz, corsair, etc. are using. The last two letters on the KVR are AT and on the corsair etc. it is BT and so there are two different revisions.

Cheers

[NiTrO bOiE]

Hey yeba, I have the same board/memory you have. I can't post pass DDR 400, no matter what timings or voltage though. I'm not sure which chips the memory is using either, I highly doubt they are the same ones in your ram.

[TheGr8s1]

I picked up a kingston valueram PC2700 512MB in CC using Hynix D43 chips.

What is CC? Got a link to the ram?

[metra]

Circuit City. Im pretty sure he's talking about the deal that included a rebate. In total it was something like 40-50 for a 512 stick.

[TheGr8s1]

That's a very nice price. I'm not much into rebates though.

[chung_chang]

I got in on that deal over the weekend. Memory is not that great. Best I could do was 240fsb 1:1 on my P4C800E with 2.4C at 3.0-4-3-7-8 at default vDimm. Any higher fsb at even the highest vDimm AND most relax timings would not run stable in memtest86. So, no PC4000 for me.

These are indeed very different from what OCZ, Corsair and Buffalo use. Those use BT and these use AT. BT is third gen while AT is second gen. They behave very differently.

However, Kingston do make PC3200 ValueRAM that supposedly use BT. I haven't been able to confirm it yet. Those cost under $100 for 512MB. Anyone care to try one of those out?

EDIT: some of you may question why I say these are not good since they do 240fsb 1:1 for under $60. Well, if you are a believer of lower latency=better performance, these won't even do CAS 2.0 at 200fsb. By the way, the Kingston PC3200 that's suppose to have the BT chips are the newer "KVR400X64C3A" model, NOT the "KVR400X64C3".

[yeba]

Thanks for the info.

It's a good deal though for a $47 512MB that can do at least DDR450.

As for timings, the timings are more critical for AMD chipsets. As for Intel platforms, FSB speed is more important than timings. DDR466 at cas 2.5 is still faster than DDR433 at cas2. Well , at least for my mobo 845pe that is.

Regarding the valueram PC3200. I looked at the newegg pic of the 'A' C3, it does not look like it's a hynix. THe pic might be outdated also. THe regular non'A' pic shows it's winbond BH-5 which is a very good performer.

Please let us know what people are getting on the valuerams.

[ps24eva]

you guys are getting the wrong chips.

you want 56822BT-D43


notice its BT, not AT

[chung_chang]

Originally posted by yeba


As for timings, the timings are more critical for AMD chipsets. As for Intel platforms, FSB speed is more important than timings. DDR466 at cas 2.5 is still faster than DDR433 at cas2. Well , at least for my mobo 845pe that is.


Haha, not sure if you frequent abxzone.com. There is a new wave of thinking going on there recently. A few ppl have proven that lower latency even at lower fsb is faster than higher latency at higher fsb. Sandra is one of the few benchmark program that says otherwise. But in 3dmark and some other real world apps, lower latency was shown to be the winner. These were tested on the I875/865 mobos. So, the idea of running highest fsb 1:1 no matter what latency is beginning to become undesirable. I'm sure this debate will be settled very soon as it's very easy to test and prove.

This brings up another issue with this Kinston pc2700 ValueRAM. It does not like 5:4 or 3:2 dividers at moderate fsb(250+) so you are pretty much stuck with 1:1 which will only land you around 240fsb. Not good if you have a 2.4C/2.6C or even 2.8C.

Originally posted by yeba


Regarding the valueram PC3200. I looked at the newegg pic of the 'A' C3, it does not look like it's a hynix. THe pic might be outdated also. THe regular non'A' pic shows it's winbond BH-5 which is a very good performer.


Yes. Winbond BH-5 is very good. But if we can confirm that the other one is Hynix w/BT chips, for about the same price, the BT will absolutely kill BH-5. Can you say 260fsb+ 1:1 with CAS 2.5??!!!

[Merc Zephyr]

you guys are getting the wrong chips.

Actually, I was well aware of which chips I was getting and that they were somewhat inferior to the "BT" version. But if these can approach, if not quite reach DDR500 for less than half the cost of current pc4000 offerings than I think it is a good deal.

My new P4 2.6C is up and running and I should be able to provide my results with KVR Hynix "AT" D43 later today.

A few ppl have proven that lower latency even at lower fsb is faster than higher latency at higher fsb

This may very well be the case but if you are someone currently deciding which route to take for your cpu&mem purchase, the high fsb 1:1 route gives you more purchasing options as far as memory is concerned. Several vendors are offering pc4000 at this time. If you want 2-2-2-5 then your only real option is something with winbond bh-5 with the best choice here being Mushkin and this seems to be a great choice according to the majority of Mushkin customers. Kingston modules with bh-5 are a hit and miss (as far as finding them goes) and something like Buffalo pc3700 is impossible to find outside of Newegg which doesn't ship to Canada. Based on the posts of others, even the quality ram from Mushkin will only overclock so far at 2-2-2-5 timings which is typical of all bh-5 based on what I have seen unless you can give it crazy voltage like 3.2-3.3V.

Right now I think either route is a valid option but only when and if the latency improves on pc4000 will the choice be much easier.

Cheers

[chung_chang]

I know what you mean. Deciding CPU and memory on the Intel platform is more painful. You don't know what fsb your CPU will hit and then trying to match that to the best performing memory you can get within your budget.

But, for enthusiasts, this particular Kingston w/AT chips will disappoint. 240 fsb max at 1:1 with very lousy latencies will not be good enough for most 2.4/2.6/2.8C owners. Then, it won't let you use 5:4 or 3:2 dividers at 250fsb+. You are basically stuck! You'll get better performance with something that'll run 220fsb+ at tight timings and use 5:4 divider. So far, I've had Winbond BH-5, Winbond CH-5 and Geil all hitting over 230fsb at good timings.

If you just want lots of memory cheap, then these are good. You get what you pay for.

[okayfine]

do these come in regiestered ECC version?

[Merc Zephyr]

Well I promised ya some results so here they are

I was also able to get my Kingston ValueRam with Hynix "AT" D43 to DDR500 at 1:1, 2.85V, 3-4-4-8 but unfortunately not without some errors in Memtest. I have not had a lot of time to test thoroughly since I just got my new p4 2.6c a couple days ago (first processor I ever installed on my own, by the way) but here are some results:

on a p4p800d, MAM: auto, Turbo: standard

175mhz, 2-3-2-7, 2.85V, 4:5 with my old 1.6a
205mhz, 2.5-3-3-7, 2.55V, 1:1
240mhz, 3-4-4-8, 2.55V, 1:1
245mhz, 3-4-4-8, 2.75V, 1:1
250mhz, 3-4-4-8, 2.85V, 1:1 with errors in test #5

After using memtest more thoroughly, the best error-free (all memory, all tests, 3 full passes) result was:
244mhz, 3-4-4-8, 2.85V
The following url is hopefully to a screenshot I attached over at abxzone and it shows Sandra/CPU-Z/Aida32 for 250fsb.

http://www.abxzone.com/forums/attachment.php?s=&postid=535286

Cheers

[ps24eva]

Merc,

what fsb will yours do at

2.5-4-4-7

1:1

2.8v


sorry if i missed it somewhere

[Merc Zephyr]

Hey ps24eva

You didn't miss it anywhere. I didn't really test the max CAS2.5 setting yet because I first wanted to know the absolute max mhz that the ram would run at. I will see what I can do with those settings although I will have to chose either 2.75V or 2.85V on my p4p800. I take it that 2.5-4-4-7 are the most relaxed timings for your board? I'll try it at 2.75V and post back later.

Cheers

[TheGr8s1]

That's very impressive considering the they are "VALUE" ram. I might want to get me some of these.

[Chuck232]

Can anyone confirm that the KVR400X64C3A are indeed using the BT-D43 chips?

[Merc Zephyr]

ps24eva:

info you requested and a little comparison for ya

210mhz, 1:1, 2.5-4-4-7, 2.75V Sandra unbuff = 2484/2562
210mhz, 1:1, 3-3-3-7, 2.55V Sandra unbuff = 2672/2745

The RAS-to-CAS delay and RAS Precharge times have an important effect on bandwidth and these chips seem to love CAS3.

Cheers

[ps24eva]

interesting Merc


Chuck232,

Why would you even think that those are Hynix 56822BT-D43?
Someone had to have told you right?

I thought Kingston was keeping the Hynix label?

Or are they just keeping the Hynix label on the AT's, but putting their own for the BT's


i'm confused:rolleyes:

[Chuck232]

EDIT: some of you may question why I say these are not good since they do 240fsb 1:1 for under $60. Well, if you are a believer of lower latency=better performance, these won't even do CAS 2.0 at 200fsb. By the way, the Kingston PC3200 that's suppose to have the BT chips are the newer "KVR400X64C3A" model, NOT the "KVR400X64C3".

Direct from earlier this thread actually. ;)

[ps24eva]

Originally posted by chung_chang
By the way, the Kingston PC3200 that's suppose to have the BT chips are the newer "KVR400X64C3A" model, NOT the "KVR400X64C3".



chung_chang where did you get this info from?

[chung_chang]

Straight from Asus (http://usa.asus.com/products/mb/socket478/p4c800-e_d/P4C800_DDR400_QVL.pdf).

Unless someone knows that Kingston actually makes their own memory chips, I think Kingston laser-etched their own brand on the chips.

[ps24eva]

yes,


good detective work. I also like to read technical documents;)


I think, though, that 2700 ram will eventually come with original labelled 56822BT-D43, soon.

My guess is that Kingston only remarks the 3200 line.

[chung_chang]

I think the 2700 already come with original label. At least the ones I bought were.

As for 3200, I have my own theory about why Kingston use their own labelling. The official mobo/CPU specs jumped from 133Mhz (PC2100) to 200Mhz (PC3200). Being the latestest and greatest official fsb, Kingston expects volume sales with the Value PC3200 (for average Joes, not overclockers) thus they want to be sure ppl see their own brand on the memory chips....just my theory.

EDIT: Oh I see. You are saying Kingston will soon switch from AT to BT chips for the 2700 right? That might happen if they run out of AT chips. After all BT chips are supposed to replace AT since BT is the next generation of AT chips. But looking at the Hynix official specs (too lazy to link now), AT's are spec'd out at 2700 while BT's are spec'd out at 3200. Will Kingston simply drop PC2700 VR instead?

[Merc Zephyr]

Hi chung_chang

I think the 2700 already come with original label. At least the ones I bought were.
Sounds the same as the ones I bought. I think the only thing on the sticks themselves that identified them as Kingston was the sticker. The chips were clearly marked as hynix ...AT D43.

You are saying Kingston will soon switch from AT to BT chips for the 2700 right?
I was surprised to see either version of the hynix chips on the ValueRam modules considering the hefty duty imposed on hynix by the United States. I had to return my KVR due to financial constraints (I needed a new power supply). If the BT chips do show up on pc2700 sticks and the price is still reasonable then I may be interested in them again.

Cheers

[chung_chang]

Originally posted by Merc Zephyr
I was surprised to see either version of the hynix chips on the ValueRam modules considering the hefty duty imposed on hynix by the United States.


Yes. Interesting. I remember reading about that. Eventhough Kingston is headquartered in US, they sell products all over the world. Do you suppose they may be manufacturing their memory out of the US? If that's done, does the memory as a whole still qualify for the hefty US duty? This is way out of my league as I know nothing about Kingston's operations.

[Speed_Mechanic2]

The duty is only for Hynix selling it's own modules. It doesn't affect Kingston memory which is assembled outside of the country and then brought in as a Kingston module.

[Merc Zephyr]

There was a sticker on the modules I had that said they were "assembled" in the United States which means the chips must have been imported. I am not sure if these modules were assembled before or after the duty was imposed but the store where I bought them said they had just received them and that I had just missed out on the previous stock that was priced at $99 CDN for a 512mb stick. The price I paid for each of the new modules was $139 CDN. That is a 40% increase and it would be in line with the duty percentage. It was still an ok deal considering what the modules can do. I wonder if those $99 modules where the Winbond BH-6 that was the original reason I went to the store?

[ps24eva]

Originally posted by chung_chang
Oh I see. You are saying Kingston will soon switch from AT to BT chips for the 2700 right? That might happen if they run out of AT chips. After all BT chips are supposed to replace AT since BT is the next generation of AT chips. But looking at the Hynix official specs (too lazy to link now), AT's are spec'd out at 2700 while BT's are spec'd out at 3200. Will Kingston simply drop PC2700 VR instead?


yeah when BT shows up on 2700 ValueRam, oh man that would be sweet. And it should logically happen. Go to your stores now people and check!

I really hope they don't drop the 2700 line

[whatever2003]

I see nothing impressive with memory that overclocks high, nothing at all. What is important is what is the REAL bandwidth.
The fact is that with Corsair running at 2-2-2-6 on 400 in 5:4 you might get better bandwidth (check Memtest86, check Aida32) than with some DDR 500 running 1:1 at 3-4-4-8.
Retarded timings is the SPD trick they get this memory to overclock high, yet the "real world" performance will not be anything to brag about.

I think people are just getting awakening from these "overclockable" claims. Nevertheless this Kingston is certainly a great deal for the price.

[ps24eva]

nobody disagrees with you!


its just that these kingstons are such a huge value.


also IMO, 2.5-4-4-7 at 1:1 is a pretty good place to be as well

[newindy4130]

:cool: :D :cool: which one is better. 2.5-3-3-7 or 2-2-2-6?

[newindy4130]

:c
ool: can i run dual channels on my pc2700? check sig........thanks
my timing is now 2-2-2-6 it was 2.5-3-3-7

[Clevor]

Ya want overachieving PC2700? Then hope that SanMax starts selling in the U.S. I'm not talking high fsb at 1:1, but 5:4, 2-2-2-5 performance that is only 0.1-0.15 volts behind BH-5.

I bought two 256 MB sticks of SanMax PC2700 because they came with the vaunted, but rare, Winbond BH-6 chips. I noticed complex circuitry on the pcb similar to Corsair XMS series, so I bought two sticks since they were only $48 per stick U.S. (in Japan).

These are my results on an Epox 4PCA3+, but understand no board runs Winbond more brilliantly:

At 2.8 volts: DDR420, 2-2-2-5, DDR430, 2-3-2-5
At 2.9 volts: DDR430, 2-2-2-5
At 3.0 volts: DDR443, 2-2-2-5
At 3.2 volts: DDR478, 2-2-2-5
At 3.3 volts: DDR500, 2.5-4-4-8, but Memtest errors

The settings above (except the last one) are not only Memtest clean but 3D stable. My XMS3500 does about the same as the above with 0.1 less VDIMM (and will even do DDR500, 2-2-2-5 at 3.2 volts). Like their BH-5 cousin, relaxing timings do not help to get higher. I can't run 250, 1:1 at 3-4-4-8 or anything, even at high voltage. The sticks don't like anything above CAS 2.0, and interestingly, they did not like 2-2-2-6, which I though BH-6 liked. Who cares, with 2-2-2-5 performance like that.

These sticks are rated 2.5-3-3-6 @ 166 fsb, so they are real overachievers. They probably do 2-2-2-5 at 166 fsb at 2.5 volts.

Unfortunately, they are too much like XMS3500, which I already have. I was hoping for some kind of ram that would do DDR450-500 at 2-3-3-6, at reasonable voltage like 2.8-3.0, on a variety of motherboards. As I mentioned, my XMS3500 have done 250, 1:1, 2-2-2-5, but it needed 3.2 volts and most SD/CW boards do not like BH-5 at 5:4.

[Merc Zephyr]

overachieving PC2700
No argument here. It seems that nothing can achieve low latency like winbond bh-5/6 with lots of voltage driving it.

I was hoping for some kind of ram that would do DDR450-500 at 2-3-3-6, at reasonable voltage like 2.8-3.0
That looks to be right in the ballpark of ocz pc3700gold.

[Clevor]

That looks to be right in the ballpark of ocz pc3700gold.

I had terrible problems getting my XMS3500 to work on an IS7 at 5:4, so I ask OCZ about ram that could do CAS 2.0 in that range. They hedged and told me some of the PC3700 Gold might do it and some might not. They declined to guarantee me anything. The sticks are pretty expensive so I'm not inclined to give them a try. The boards that do work pretty well, if not brilliantly, with XMS3500 are the 4PCA3+, P4C800-E, and I suspect P4P800. I will be giving Abit another shot with an IC7-MaxIII and it's 3.2 VDIMM, but I'm skeptical.

BTW, do you know what chips are on OCZ PC3700 Gold? Samsung?

[whatever2003]

2.5-4-4-7 at 1:1 is a pretty good place to be as well
2.5-4-4-7 is no better than 3-4-4-8.
Half a cicle on the CAS get lost/absorbed, and TRAS makes little or no difference at all.

[Merc Zephyr]

BTW, do you know what chips are on OCZ PC3700 Gold? Samsung?
I think they are samsung chips sanded down a little for better heat dissipation.

[Clevor]

OCZ confirmed it is Samsung 4.3 ns chips.

BTW, after testing several 865PE and 875CW boards, I am convinced most of them have 5:4 async issues. It's not the ram. I just tested a Solteck SP-865SE-L and it too has memory issues at 5:4. At least this board Memtest is clean at 283, 5:4 with ADATA PC450, but not 3D stable until you drop way down. Same problem I had with the IS7. Or you see a lot of Memtest errors at 5:4. But ADATA PC4000 runs fine at 279, 1:1, 3-4-4-7-8 at - get this- 2.8 VDIMM! Also XMS3500 runs fine at 2-2-2-5 as long as I run 3:2. Heard the Gigabyte 8KNXP has the same problem.

The only boards I've seen that run 5:4 OK with BH-5 are the P4C800-E and 4PCA3+. The other board manufacturers should find out what's going right with those boards.

[Merc Zephyr]

The only boards I've seen that run 5:4 OK with BH-5 are the P4C800-E and 4PCA3+.
Interesting. Have you seen how BH-5 runs in the P4P800 at 5:4? I am seriously considering new RAM for my system and this was one of my considerations along with OCZ 3700gold or something with Hynix "BT" D43, both of the latter I would run 1:1. (Please no one bring up the low latency vs high bandwidth debate. I am well aware of it but my primary consideration is finding a memory solution that will be compatible with and maximize cpu speed in my system.)

BTW, after testing several 865PE and 875CW boards, I am convinced most of them have 5:4 async issues. It's not the ram.
Here's where I'm at in my system: after trying a new power supply, active northbridge cooling and going through the entire bios, I still wasn't convinced that my cpu or my mboard was holding me back so I tried my RAM individually in single channel:
- In dual channel, max fsb was 255mhz
- With stick that I new was better, max fsb was 269mhz - must be the reduction in stress on the northbridge, right?
- With other stick, max fsb was 255mhz
In my case, it is either the RAM or asynch (5:4 and 3:2) issues between the mboard and the RAM. Either way I am convinced that different RAM is what I need. It doesn't help that I don't know what chips are used on my OCZ although I assume either Samsung or CH-5.

So to sum up, I can try BH-5 and hope it works better in asynch or just steer clear of asynch altogether and look for the best 1:1 option.

Anyways, Cheers and thanks for the info.

[Clevor]

Interesting. Have you seen how BH-5 runs in the P4P800 at 5:4?

I've been planning on trying that when I can find a cheap, used 35Mxxx board, e.g., an early May P4P800. But I imagine it will run as well as on the P4C800-E: DDR440, 2-2-2-5 at 2.85 volts, DDR453, 2-3-2-5 at same volts. With a VDIMM mod, there's no doubt in my mind the P4C800-E can run almost as well as the 4PCA3+ which offers a max 3.32 actual.

For me, things are simple: I get best benches by running 5:4, 2-2-2-5 with XMS3500. If the board can't run 5:4 it will usually run ADATA PC4000 fine at up to 282, 3-4-4-5 at maybe 2.9 volts. I think I got pretty good sticks and they probably can do 300, 1:1 at less than the usual 3.2-3.3 volts, 'cept I don't have a CPU that can do that high and I use air.

If you are on a low budget, ADATA PC4000 should work on most any board at 1:1. At 265+, 1:1, 3-4-4-8 you should be at or over 3k unbuffered, which is pretty good.

[larva]

Originally posted by Clevor

If you are on a low budget, ADATA PC4000 should work on most any board at 1:1. At 265+, 1:1, 3-4-4-8 you should be at or over 3k unbuffered, which is pretty good.
Man, I would sure hope so...

unbuffered at 380MHz
http://www.theforumisdown.com/uploadfiles/0103/unbuf380.jpg

If you are really on a budget it is worth the trouble to find some BH5s that the Abit will abide by. The Kingston HyperX 3000 is a BH5 variant that is agreeable with the IS7 and IC7 in the 5:4 mode. Of course a volt-mod helps, but as you noted this is true of the Asus as well.

[Merc Zephyr]

If you are on a low budget, ADATA PC4000 should work on most any board at 1:1.
Thanks Clevor. I forgot to mention that I was also considering the ADATA. Could you recommend a vendor to buy it from? I know it's not easy to find in the U.S. let alone in Canada. The only ones I know of are Hardwaregods and Komusa and of those only the first ships to Canada.

Larva: I am definitely considering the khx3000 and 3200(with bh-5). They are nicely priced and surprisingly still available. BTW your 1.8a should appear in the dictionnary next to "overachiever" :eek: That's amazing.

Thanks guys. Cheers

[larva]

Yeah, HyperX3000 and an Abit is a recipe for (relatively) cheap speed. Abits are super picky on the ram but are capable of outperforming anything I've seen. Quality control and compatibility are not stellar, but performance is.

My 1.8a is one of the fabled Dell-pull SL6LA C1 stepping chips. It's stable to 3.42GHz on air with a mild fan at 1.68V. I have to jack my Vcore up to 1.71 nominal to maintain at leat 1.68 under load. This is due to my incredibly overloaded power supply, the next thing on the upgrade agenda.

[Clevor]

Larva are you running Win98SE? That gives maybe 150-200 higher Sandra memory marks. If not, those are very good scores for WinXP.

If you've seen my other posts on the various forums, I mentioned I've tested several Springdale and Canterwood boards where I was stuck at CAS 2.5 at pretty good DDR settings: DDR440+, 5:4, 2.5-3-3-5/6. My unbuffered memory scores were 2400-2700, which can be bested by an RDRAM board running 156/4x. I've seen the high side too, where I've run 250, 1:1, 2-2-2-5 on the 4PCA3+ at 3.2 volts, where I got close to 3700 unbuffered (up to 3900 in Win98SE). That's two 512 MB XMS3500 sticks.

If the BH-5 HyperX runs OK on the Abit boards, it's probably the PCB. The Corsair has the most complex PCB I've seen (resistors and stuff), and the SanMax is similar. Rumors are half the IC7-MaxIII boards run BH-5 (XMS3500) OK. We'll see, as I will be setting one up soon. But unbuffered Sandra scores still trail the 4PCA3+ by 100-150 points.

I believe you can get ADATA directly from the manufacturer. You may want to contact someone on their web site. Or sit tight, they say DDR500 with runs at CAS 2 is around the corner. I've found BH-5 is slightly better than BH-6 that is on good PCB. That tells me the long awaited BH-4 chips might only be slightly better than BH-5.

[larva]

Originally posted by Clevor
Larva are you running Win98SE? That gives maybe 150-200 higher Sandra memory marks. If not, those are very good scores for WinXP.


Lord no! Win2K SP3 at present. I rather despise XP's interface, although I ran it for about 6 months before reverting to Win2K.

I always enjoy your posts Clevor as they are technically sound and reflect the vast array of components you've tried. I would be surprised if you didn't find a combo that works for the Abits, they are capable of great things in the right circumstances. I do tend to classify them as junk compared to the Asus, but very fast junk

[Clevor]

I do tend to classify them as junk compared to the Asus, but very fast junk

Naw! I think highly of Abit boards, particularly the 845E/PE boards. Abit was the first guys to figure how to get on-board audio to work with the board overclocked to the hilt, even with AGP/PCI locked. Albatron and Jetway had problems on their 845PE boards (I have to run SB Live cards on them). You will also note there are no ATI 9700 Pro incompatibility problems with their boards, though they never did produce a Granite Bay product. They are simply behind the curve a bit on the SD/CW chipset.

I still have a BE-7, BH7, and IT7 Max2 Ver. 2.0. Some of the best overclocking boards ever. And their NF7-S Ver. 2.0 is currently the best AMD board on the market. I've gone 228 fsb, 2-2-2-5 on an unmodded board with XMS3500. If I seem partial to XMS3500, it's because they are such great dual purpose ram - with volts. They are left over from my 845PE wars. That's why I run two 512 MB sticks, otherwise I go with two 256 MB sticks when I buy new ram due to cost.

[Clevor]

I just tested an IC7-MaxIII last night. Serious, serious issues with ram dividers and memory. Just take a look at the Abit forum. CPU is not 3D stable beyond maybe 260 fsb, when it runs 277-287 on other boards. Memtest can be clean but 3DMark kicks out until you drop fsb to 260 or below. Abit really dropped the ball on this one. I had the foresight to purchase the board from Newegg so I can always return for a refund.

If you set VDIMM above 2.8 in 5:4, Memtest locks up in the #5 test. At least I can boot, others can't.

But yes, it does give good bandwidth near DDR400. I tested DDR380, 2-2-2-5 and got around 2800-2900 unbuffered. However, moot point as the board is not 3D stable much above that, which is what you need with the 5:4 and 1:1 dividers.

[larva]

I have no doubt that the Abit's are picky, and obviously just don't care for the memory types you have on hand. But there are memory types they will run on, and many people are in the high 290s with the Max3, and there are several people over 300fsb with the IS7. It's not that the boards don't work, but they are picky, picky, picky on the ram.

I think you can see why I run the IS7 though. Since I have an 18 multiplier I don't have to run crazy fsb to generate the clock speed. More fsb doesn't really hurt (until you can't run 1:1), but it doesn't really help that much either. It is the natural consequence of the cpu multipliers of the affordable c chips being too low in reality.

Hipro ran some test for me (well, all of us) with his ES 3.0, but unfortunately it will only go to 16x. I think 17x or 18x is really optimal, depending on how talented your ram is. As ram quality gets better 1:1 will be viable to higher fsb's, and the multiplier range of the current chips will be more useful.

Certainly Abits are not the paint-by-numbers affair your Epox is, but they are not faulty in their design or execution, either. The are unbeleviably highly tuned for performance, and the downside is they can appear unstable at high fsb as a result. But if you look at enough hot-rod setups you will see that Abits are at the core of a lot of them, and they are running high fsb's just like the Asus's or Epox's.