Dissapointing upgrade comment

[paparazi]

I just upgraded from a KT266a chipset motherboard, an MSI KT7 ultra to the new KT400 chipset on a gigabyte GA7-VAXP, the reason mostly was because of all the integrated extra's (onboard lan,USB2, Firewire and 8X AGP) and oh yes! maybe a little extra performance in my games.
512mb 2100DDR
2X Seagate Barracuda 7,200 drives
Yami writer
Pioneer slot DVD
56X CD rom
Leadtek winfast 250LE ViVo GF4 4200Ti

Used to get about 10,500 in Mad onions 3D benchmark 2001SE with the old motherboard and now with the new KT400 chipset Gigabyte just 9,800.
Now I know that it has been widely reported that the new KT400 doesn't perform much better than the KT333 but surely not worse than the older KT266a?
The same performance I could put up with but LESS! is a bit of a pisser.
Oh yes! tried flashing motherboard and graphics cards with latest.
Latest 4in 1's
latest DirectX 8.1b
Latest Detonator drivers
plus best performing bios settings
AND older drivers just to C

ANY COMMENTS or experience's please of the new KT400's.
I have to add that appart from this performance issue everything else seems brilliant.
Martin Replying helps further our knowledge, without comment leaves us wondering.

 

[legend2b4]

OS windows XP,9X ????

carn't be linux

 

[CitrixEngineer]

I went for the MSI KT4 Ultra FISR, and have to say that there is no noticeable usability speed increase over the KT3 setup I built for a relative. There is just an odd feeling of more power behind the scenes.

I compared this with my old system, which is running an XP1600+ on a KG7-RAID (AMD 760/KT133 chipset), and there's not a lot to choose between them in terms of useability, although the KG7 certainly feels slower.

Enough of the "touchy-feely" - I'll run Mad Onion on both and post the scores - when I get a moment.

I'm surprised that your benchmark figures are lower, even given the fact that MSI boards consitently outperform Gigabyte boards in most benchmarks, and the KT266a is a landmark chipset; Maybe it's to do with latency?

I assume you're running a Palomino with the older "Mustang" or A core (266Mhz) rather than the newer "Thoroughbred" B core (333Mhz).

My theory is *and it's only a theory - if anyone wants to rip it apart, then great!* that there is a bottleneck in terms of latency between the 166Mhz Memory, the 200Mhz FSB and the 133Mhz processor caused by synchronisation. It's possible that everything has to be stepped to a common denominator, or that additional wait states are introduced by this combination.

On to slightly simpler things - if you're using all the onboard stuff, it's probably stealing CPU cycles - especially sound, and probably LAN. Try disabling some of it.

The manufacturer's "Best Performing" settings aren't always the best - does your RAM allow CAS 2? Is there a "Normal, Fast, Turbo(AKA lockup) setting? Check everything that has wait states - it's possible that these can be lowered in many cases, but you may experience the odd lockup if you change too much. You may even get away with a Mhz or two increase in FSB speed, even on a "locked" processor...

Here are a few BIOS tweaks you could try, although RAM tweaks usually only account for a 5-10% gain in performance in any setup;

You should have a setting for RAM clock. If set to auto, this will run at the same speed as the CPU. You should be able to set this to 166.


***The following is taken from Ars Technica's BIOS tweaking guide (

http://www.arstechnica.com)***

Delay DRAM Read Latch
Options: Auto, No Delay, 0.5ns, 1.0ns, 1.5ns

Marginally different from the one that only has Enabled or Disabled, this tunes for high RAM loads. One single-sided DIMM isn't a high load. Two double-sided 256MB DIMMs are. For lower RAM loads, lower the latch and have it at no delay. With more DIMMs, you need more of a delay or the chipset may find itself unable to properly latch onto a bank. A higher latch reduces performance. If you have crashes after upgrading your RAM, this is the very first setting you should look at.

DRAM CAS Latency Time / DRAM Cycle Length
Options: 2, 2.5, 3

This is the famous CAS latency. It's part of the wait between the chipset requesting data and the DRAMs getting ready to send it. A shorter delay is better; 2 is less of a delay than 3. However, your RAM needs to be able to handle the increased rate and may not be able to do this if the FSB is overclocked or the RAM is of a lower specification. Increasing the CAS latency will therefore allow you to overclock the RAM further (if the chipset and CPU will let you). 2.5 is only available with DDR.

DRAM Cycle Time Tras/Trc / Minimum RAS Pulse Width / Row Active Time
Options: 5/6, 6/8

Like the CAS latency, a lower setting is faster. Like the CAS latency, it's more stressful on the RAM when it's lower. Hannibal's excellent RAM guide will take you through this and all other RAM-related jingo. A low Row Active Time will force the data out of the RAM sooner, but it may not leave the row open long enough for transactions to complete, in which case you either need to get faster RAM or increase the setting up to 6/8.

DRAM Interleave Time
Options: Disabled, 0ms, 0.5ms

Specifies the time to wait between interleaved transactions. Disabled and 0ms are the same thing and are the fastest. You know the drill with these RAM settings. Give it that 0.5ms if you're using three or four sticks of high capacity RAM just to keep the chipset playing nicely with the high load.

DRAM RAS Precharge
Options: 2, 2.5, 3

The third part of the x-y-z notation used in SDRAM, the other two being CAS and RAS to CAS. Like its brethren, it's better lower but also more stressful lower. See the pattern? 2.5 is only available with DDR.

DRAM RAS to CAS delay/R2C Latency
Options: 2, 3

When RAS is asserted, there must be a small wait before the CAS can be pulled. This setting controls length of the wait. Like CAS latency, it's a delay before you get your data, so while your system is faster at a lower setting, it's also more stressful at that setting. Your RAM may handle it, or it may not.

Fast R-W Turnaround
Options: Enabled, Disabled

This is a setting related to the CPU's side of the chipset, but it still involves RAM. When the CPU switches from reading to writing, it has to delay. This shortens that delay. Enable it for best performance; disable if it causes problems. You know how it is with these things.

Force 4-Way Interleave
Options: Enable, Disable

Even if the chipset doesn't detect four banks, this will insist that the chipset uses four-way interleave. If the chipset gets it wrong, the user obviously knows best...right? You should really disable this. The chipset is likely to turn four-way interleave on if you specify it in the Bank Interleave setting.

MD Driving Strength/Memory Data Drive
Options: Low, High

With a high RAM load (as mentioned in Delay DRAM Read Latch), the signal strength may be insufficient. Change from Low to High to remedy this. Also, this increases stability when using overclocked RAM. This setting only affects stability and not performance.

Read Around Write
Options: Enable, Disable

The chipset can issue read and write commands out of order from the chipset's R-A-W buffer, resulting in higher performance if this feature is enabled. The buffer also has other benefits, so it's a good idea to enable this.

Read Wait State/Read Latency
Options: 0, 1

When something reads from RAM, the chipset services the request. However, it usually holds the data for one cycle before making it available. Disabling this delay (set to zero cycles) helps performance, but the data could arrive too early and the requesting device could not be ready for it, which would result in instability.

Refresh Interval/Refresh Mode Select
Options : 7.8 µsec, 15.6 µsec, 31.2 µsec, 64 µsec, 128 µsec

Normally SDRAM and DDR are refreshed every 64ms. However, refreshing every cell simultaneously will result in a power surge. That's not good. So the refreshes can be staggered from one row to the next. 128Mbit and smaller DRAMs like to have this at 15.6 microseconds. 256Mbit DRAMs have twice the rows, so half the interval, 7.8 microseconds, is appropriate.

The JEDEC standards do call for 64ms (not µsec!) but today's DRAM can handle more than that between refreshes, so for performance and power (mobile users listen well) reasons, you may want to increase this all the way to 128 microseconds to add a small delay on top of the 64ms already. It helps performance by keeping RAM available for longer. It helps power usage by not refreshing as often.

SDRAM Command Leadoff
Options: 3, 4

Yet another setting that's faster lower. Set it to 3 if you have badass ninja RAM. Set it to 4 if 3 doesn't work.

SDRAM Idle Limit/SDRAM Idle Timer
Options: Disabled, 0 Cycle, 8 Cycles, 12 Cycles, 16 Cycles, 24 Cycles, 32 Cycles, 48 Cycles
(Note: cycles may also be called "ticks&quot

This tells the SDRAM how long it should idle before recharging. Increased values allow the SDRAM to postpone a charge and reduce RAM latency. Tune this with the Refresh Interval to get the best settings. I recommend 12 cycles for machines with less than 512MB in total and 32 cycles for all others.

SDRAM Precharge Control/SDRAM Page Control
Options: Enable, Disable

This determines whether the chipset or the RAM controls refreshing. It's better to let the RAM do it itself (enabled), but doing so can cause stability issues with large amounts of RAM installed or with poor-quality RAM. In those cases, you should disable it for a small performance loss but better stability.

Video RAM Cacheable
Options: Enable, Disable

Disable this. You don't want to be wasting the L2 cache on fast video RAM when you have slow system RAM to deal with. The bandwidth of your system RAM is unlikely to be over 3GB/sec, yet video RAM can easily top 10GB/sec and a 4x AGP bus is 533MB/sec. The tiny amounts of L2 we get these days to go with our massive system RAM sizes relegate this setting to being Disabled.

***The above taken from Ars Technica's guide to BIOS tweaking***


CitrixEngineer@yahoo.co.uk

 

[paparazi]

Thanks Citrix Engineer,
XP pro and yes I have read and implemented in part bios tweaks as I had for the MSI.
I missed out all the interim explaination of experimentation I have carried out, I have left no stone unturned (I don't think) lol
I havn't clocked anything but then I hadn't clocked anything on the MSI setup.
XP2.1+ Palomino core.
I will have another tweak session but still feel there is no more to come short of clocking.
Martin
Replying helps further our knowledge, without comment leaves us wondering.

 

[GoLdFiNgeR]

i dont know how true this is, but i heard that with teh 400 chipset and slower ram, it doesnt really even give you the pc2100 ram performacne, in other words, it degrades the performance of your ram, again ive never tried it or seen it yet, so im not speeking of experience,

its just another thought

 

[dakota81]

A lot of the minor differences in speed depends on who makes the board too. And it's true that the KT400 chipset barely stands up to the KT333 chipset but i think that's mostly due to ddr 400 ram not being able to perform as well as ddr 333 ram.

I don't think there's really much difference in the workings of the 266A, 333, and 400 besides the integrated features (agp8x, usb2, etc.) and the support of higher speed memory.

When you get into speeds like the processor you've got, the memory becomes the big factor in system speed. So if you want to see improvement you'll have to switch up to ddr 333, CL2.

 

[cdogg]

I agree with Dakota. I suspect another setting in the BIOS is interfering, or perhaps the board itself has a few bugs yet to be hammered out in future BIOS revisions. Upgrade to DDR 333 CL2 when you get the chance to take full advantage of your new board.

If it helps, you can be proud that you now have V-link technology (533MB/s pipeline connecting the Northbridge to the Southbridge). It used to run at 266MB/s. Plus, you've added Firewire, AGP 8x, ATA133, and a nice little fan on the Northbridge, all of which weren't available on the MSI board. Yeah, they probably aren't much of a factor in overall performance, but bragging rights are always nice!!

Plus, you're not stuck at 133MHz for the FSB anymore. The MSI board didn't let you overclock, but that's history now!


~cdogg

"The secret to creativity is knowing how to hide your sources."
- A. Einstein

 

[CitrixEngineer]

None of this really answers why a KT266a board gives a lower benchmark than a KT400 board - this simply should not happen, no matter who the manufacturer is.

I suspect the combination of components or the build of the machine;

Did you do a complete rebuild of the machine (from scratch), or did you do like I do (not as I say ) and just shove the board into an existing Windows build? Makes a world of difference, even though it takes a long time.

Hope this helps

CitrixEngineer@yahoo.co.uk

 

[paparazi]

Thanks guys for the imput, you have all pointed out the obvious but the obvious I have covered several times over.
Yes CitricEngineer, I did a complete rebuild, all drives and roms on there own IDE channel with a clean install of XPpro plus SP1.
CitrixEngineer got the original point I was asking!
I know that there is little speed differance between the KT266a/KT333/KT400 but
IT SHOULDN"T BE SLOWER!
Anyone got some sensible FSB setting sugestions I can use with my CAS 2 2100DDR (stable ones)
Martin Replying helps further our knowledge, without comment leaves us wondering.

 

[paparazi]

I don't think I said a propper THANKS! Guys aspecially CitrixEngineer for that good bios information.
If anyone has a similar setup to mine I would be interested in having the benefit of your own experiences of the KT400 and any experiments you have done with LITE overclocking, graphics cards, CPU and memory.
I would be grateful to benefit from your successful tweaks. Martin Replying helps further our knowledge, without comment leaves us wondering.

 

[Win98User]

Hi paparazi,
I've been following this string with some interest - you have my sympathy for an upgrade that became a downgrade - been there, lol.

I would just like to add my 2 pennies worth.

At the top you mention "onboard lan,USB2, Firewire and 8X AGP". Presumably, if we are talking a like for like comparison (ish), these items would have been serviced by add-ons or not at all.

Experience has shown me that when you have intergrated rather than add-on, you take a performance hit.

What do you think?

Kind Regards,

John.

 

[ceh4702]

I have heard of some motherboards that slow down the processor when the temps get too high. This is something that could be happening. Sometimes there are high/low temps in the bios and I usually turn that off after I think it will run ok. The RAM may be getting too hot or is just not up to running at that speed, or the motherboard memory bus or chipset may not be good enough to run at that speed.

I read somewhere once that Asus tends to use just slightly more voltage than is necessary for RAM due to possible fluctuation from the power supply. I wonder if you can adjust the voltage up say one tenth of a volt and what effect that would have. If you do not like my post feel free to point out your opinion or my errors.

 

[paparazi]

Temps are well sorted (never goes over 52C) idles around 47C.
Although I have all these extra's at the moment I only use the onboard Lan over and above what I was using before.
I have just picked up 512mb of PC 3200DDR to try, it's not a top brand or anything and I think it's CAS2.5 so we will see what if any differance it makes swopping it with the 2100DDR.
I will keep you posted.
I agree with the integrated versa's seperates sinario but don't think the integrated Lan is gonna slow it down that much.
Not using an 8X AGP card
Not using USB2 devices
Onboard sound disabled
Not using firewire so also disabled
And before anyone comments as to why I upgraded I will say these extra's are there if i need them.
Like I say the machine overall does actually feel quicker, it's just that 3Dbenchmark that should be the same or a couple of hundred points higher? according to the reviews I have read, certainly not less.
Keep em coming mates.
Martin Replying helps further our knowledge, without comment leaves us wondering.

 

[CitrixEngineer]

OK, I finally got my 2 systems benchmarked, and the results are very similar, although both MUCH lower figures. I'm happy with the usability speed of both, for what I do, however;

System 1:

ABit KG7-RAID (Via KT 133 + AMD 761)
Athlon XP1600+ max temp 34C
512Mb PC 2100
Maxtor 7200 RPM ATA 133 80Gb
Leadtek A170 GeForce 4 (64Mb)
Creative Live! Platinum
Plextor 24/10/40
Toshiba SW2002 DVD RAM
AOpen H600A with 4 case fans
Hiperpower 350W switching PSU

3DMark - 4635


System 2;

MSI KT4 Ultra FISR (Via KT400)
Athlon XP 2000+ max temp 32C
512Mb PC 2700
Maxtor 5400 RPM ATA 100 40Gb
MSI GeForce 440MX SE (64Mb)
Onboard sound - CAudio (I know! - I've got another SB Live! for future tests)
Creative DXR3 + 12x DVD
Sony 4x4x24
Thermaltake Xaser 5000A with 5 case fans
Hiperpower 350W switching PSU

3D Mark - 3984


Who would have thought it? I'm going to see what difference the SB Live! makes - but I would not have expected a lower score at all. Especially since system 2 is definitely a shade faster in use, and startup. Could be that the disk in system 2 is holding it back - that's getting an upgrade in the new year.

Not conclusive, but hopefully interesting. Just goes to show that there's not much to choose between socket A boards in general.



CitrixEngineer@yahoo.co.uk

 

[paparazi]

CitrixEngineer,
Always interesting reading your comments.
Fellow Brit?
Anyway who would have believed it! your two systems, I would have guessed at about 1,000 points minimum differance between the two if not more so what a surprise.
You maybe interested to know mine has now a default benchmark of around 10,600 compared to the 9,800 figure I was getting.
The story goes: I tried two sticks of 256mb PC3200DDR and they would only run at normal settings, it wouldn't even post with high performance settings, 3Dbenchmark was slightly lower than the 2100DDR at around 9,600 !! so that came out straight away!
I just happened to try the benchmark with 1 stick of quality winbound 2100DDR (my other stick is a no name)
and holy Jesus, mother of god, 10,598! so I managed to talk my friend at the shop into trading the cheepo for another Winbound stick to match the first, now I have a decent pair (said the actress to the bishop)
Just goes to show how critical memory can be! OR more to the point, how picky these motherboards are with memory.
I was tempted to go 2700 but I think 10,600 Default on my 4200Ti with 2100 is pretty dam good and working in a PC repair shop does mean I can trade up later.
Martin
Replying helps further our knowledge, without comment leaves us wondering.