Interleave tweaks & VIA

[Guest]

Since reading of the proggie from H.Ota at viahardware.com I've been trying to gather info on 4-way bank interleaving to improve the crappy VIA memory bandwidth.
I'm playing with a MSI Pr02-A which, in current beta-bios, has some support for 4-way if memory is set to SPD in the bios. ABIT and a few others impliment this better natively, but even this crude feature greatly improves performance with only one stick of generic sdram.
The H.Ota program allows many more variables and other features as well (such as settings for agp dividers not built-in on the MSI). When I receive the cas2 256 and 2 128's from Crucial I just ordered, I'll be able to compare the difference between a single and 2-stick approach.
It would appear that the memory bandwidth can be increased by as much as 40-70% with this low-overhead tweak, regardless of clocking.
Do any of you gurus have experience comparing 2 smaller to 1 larger sticks using this program for settings (or with more robust bios', for that matter)? There seems to be a lot of controversy on this topic (TS at Crucial think you need 4 banks of dimm filled to use it, which I know is wrong). Also, since miserable memory bandwidth is one of the few areas where VIA lags Intel, why hasn't Tom focused on there bios/software tweaks (you're losing that 'cutting edge', Tom!)?

 

[Guest]

Very interesting- any results yet logger?


Screw the risks!
-Street-AMD 1450Mhz-Abit KT7A-Raid-256CL2-Fop38

 

[pvsurfer]

Logsmith~ This is not to say you are wrong, but what's the background for your "crappy VIA memory bandwidth" staement?
Just curious...

 

[Guest]

Hey, I'm a VIA fan from way back.
But--have you looked at the comparative results in a Sandra Memory Benchmark results view lately? It's off-putting to see the Intel #'s.

 

[Guest]

Still waiting a Crucial rma on a mistakenly sent cas3 128.
For now, I can say that 1-256 gets better BW than 1-128 (both cas2). The 256 without 4-way (from MSI bios SPD) was 390/425 but with it went to 475/605 (Sandra results).
Very nice!!
Still waiting that stick and time to play with Wpcredit.

 

[pvsurfer]

Can you give me a reference link that corroborates what you are saying re: comparative results in a Sandra Memory Benchmark?

 

[phsstpok]

Skinner provided this link in an earlier post.

<A HREF="http://www.viahardware.com/wpcreditkt133.shtm" target="_new">http://www.viahardware.com/wpcreditkt133.shtm</A>.

The article has the benchmark results you are looking for.

 

[pvsurfer]

Other than presenting a memory interleaving tweak for KT133 (enabling interleaving will improve memory bandwidth regarless of chipset), I don't see anything in that reference to support logsmith's "crappy VIA memory bandwidth" comment (i.e., where is it shown that VIA's memory bandwidth is poor compared to others)?

 

[phsstpok]

What are you whining about?

You asked, "Can you give me a reference link that corroborates what you are saying re: comparative results in a Sandra Memory Benchmark?"

I gave you a link comparing memory benchmarks with and and without the interleave tweak.

 

[IntelConvert]

phsstpok: I don't see that pvsurfer is whining at all. He's just saying that while your link does show the advantage of enabling memory interleaving, that very same advantage would show up by using memory interleaving with any chipset.

He (and I) would like to understand the rationale for logsmith's "crappy VIA memory bandwidth" statement. Without any reference links showing that VIA has poorer memory bandwidth than other chipsets, amounts to bashing VIA without any evidence... Don't you see where we are coming from?

 

[phsstpok]

I, personally, was responding PVSurfer's comments not Logsmith's comments.

As for your comment about interleaving improving performance on any chipset. This is true but the link is talking about the Via KT chipset (and perhaps the KX) chipset. As you know, not all boards with these chipsets have the interleaving option hence the need for the tweak.

This is why I posted my response, to give evidence. That is what PV Surfer requested in his post, "Can you give me a reference link that <b>corroborates</b> what you are saying re: <b>comparative results in a Sandra Memory Benchmark?</b>". I interepreted this to mean that PVSurfer wanted evidence that the tweak, in fact, works. I showed him it does then he bitched and moaned. Well, too bad.

 

[Tempus]

Do you know by chance if the A7V133 has 4-way bank interleaving as an option in the BIOS?

- I don't write Tom's Hardware Guide, I just preach it"

 

[IntelConvert]

Tempus: Yes, it is enabled in the A7V133 whenever the "Optimal" setting is enabled in the BIOS.

 

[IntelConvert]

Hey man, I'm positive you misunderstood pvsurfer as he is a cool dude who has been very helpful to me and others in this forum. I have never known him to bitch and moan (as you said). I'm pretty sure he was asking for a link that supported logsmith's comment that VIA memory bandwidth is crappy (which both he and I took to mean worse than other chipsets)...

 

[Guest]

Yeow; all that!
Again, I always choose VIA, so it's not 'bashing'. As for substantiating, I didn't think that was needed in this company.
As I posted, just take another look at a Sandra Memory Benchmark screen, just below the current, the one that says: Intel i850 P4 1.5GHz.? See the bar sliding off the chart?
Really, I wasn't trying to stir anything up, I just thought comparisons were implicit. There is no better 'bang-for-buck' than VIA.

 

[pvsurfer]

First of all - truce!
Secondly, RDRAM kicks butt when it comes to bandwidth (especially with the i850 chipset)! That's its one and only advantage over all other memory types. RDRAM falls down when it comes to latency (which is quite high) and for many uses lower latency is more important than higher bandwidth.

 

[Guest]

Thank you.
Of course, both would be nice!
So anyway, did anyone have anything to say regards Wpcredit used in 4-way and/or agp dividers?

 

[hammerhead]

Enjoyed reading this thread and just wanted to point out that Sandra, while an excellent program, is somewhat suspect when it comes to memory tests.

It's useful to gauge performance tweaks in your own system, but not against other chipsets.

A couple of years ago I bought a motherboard which scored very badly in the memory benchmark. 'Real world' tests were fine and reviews on the net pointed out that although it scored poorly in Sandra, it was as fast as any comparable board.

Just checked and it was a Via chipset, VT82C693A

 

[IntelConvert]

Interesting post. I have read a good many posts in Tom's (and other) forums alledging that VIA memory bandwidth and VIA IDE (I/O) performance is not nearly as good as that of other chipsets (most noteably, the i815). So it would sure be nice to have benchmarks that provided reliable memory and IDE performance comparisons across platforms (chipsets).

 

[hammerhead]

May be true that Via is worse, not as bad as Sandra suggested in my case though.

Identical systems, one with Gigabyte BX board, one with Via mentioned above. Sandra showed the Via to have less than half the memory bandwidth of the BX.

Tests like 3dMark showed not much difference (BX was a little faster).

I went to all that trouble because I thought the Via board must be duff!

 

[ejsmith2]

Is there that much difference in running pc-133 memeory on VIA/Intel motherboards? The memory bandwidth for RDRAM is incredible, but does it give you any performance boost when you are using a program on the system?

I use an a7v133, and I've got a friend who has an i815 setup. I can't tell any noticable difference on the IDE performance, except when I hook my drive up to the ATA-100. It loads things ever so slightly faster; just enough to be noticed, but not revered.
Either way, Unreal Tournament, Dragon Naturally Speaking, and Nero all work just as quickly.

You can also look at the horsepower specifications for a tractor, car, and Cessna-150, but can you really compare them?