Page 3 of 4

Posted: Mon Jan 15, 2007 4:56 am
by Stedman5040
I'll post some results for Hotcputester tonight. My DFI K6XV3+/66 board is running fine with Hynix memory with 2-2-2-5 timings. From what I remember the results on Everest with wpcredit tweaks were a bit better than I got with the Epox EPMVP3G2 or GM boards.

Everest figures using 384MB RAM and a K6-3+/450 @ 550 (5.5x100)

MR 307
MW 157
ML 214

The only SS7 boards I have that give better Everest benches are based on the ALI V chipset such as the Asus P5A. The DFI is the top of the pile MVP3 board with regards to Everest benches. We will see how this relates to performance when I run the other benchmarks.

Stedman

Posted: Mon Jan 15, 2007 5:10 am
by DonPedro
Jim,

I really would like to display your test results. the reason why I did not so far is that either your setup did not "fit" with what I wanted to talk about or, and this is most important, it needs full sets of benchmark data (not only everest or sandra) to make a comparison sound and insightfull. also I am confused by the sheer number of superpuppies you keep on going. it is mandatory to know exactly what hw-setup was used to learn from the data gathered. so as an invitation and little incentive to get your data to the public I am going to send you a spredsheet template I am using for collecting the bench numbers.

for example I wanted to show the impact of onboard cache. this I did with ali-boards but since they only feature 512kb I started to collect some via boards. I bought a shuttle hot 597 with 512kb, a dfi k6xv3+b2 with 1mb and an epox mvp3g5 with 2mb cache. unfortunately the epox was dead on arrival (broken mosfet transistor) and the dfi, well, seems to have that much troubles that it does not make sense to use it in a serious investigation. this investigation would have been the proper place where I would have humbly "insisted" ;) that you join in. I am sorry that it found an early end. but why not start a journey into that field by yourself? I would like to participate! the data of the shuttle hot is already collected and if there is reason to also show data of ali-chipset boards I will surely prepare them.

Posted: Mon Jan 15, 2007 6:06 am
by Stedman5040
@Peter

I have one board with a 2mb onboard cache, but I have not been able to push it to the limit with wpcredit. At the moment it is sitting in mothballs and is the TMC TI5VGF. I can start to gather data on this board if you like after I have benched out the DFI.

By the way Peter, I took a close look at my DFI board the other day and noticed that the caps look as if they are starting to swell. Are the caps OK on your board? Could be one source of problems you are having.

Stedman

Posted: Mon Jan 15, 2007 6:06 am
by Stedman5040
@Peter

I have one board with a 2mb onboard cache, but I have not been able to push it to the limit with wpcredit. At the moment it is sitting in mothballs and is the TMC TI5VGF. I can start to gather data on this board if you like after I have benched out the DFI.

By the way Peter, I took a close look at my DFI board the other day and noticed that the caps look as if they are starting to swell. Are the caps OK on your board? Could be one source of problems you are having.

Stedman

Posted: Mon Jan 15, 2007 7:44 am
by DonPedro
stedman,

the caps are alright, I checked everything. I used the same other hardware (ram, cpu, hds, cables, power supply etc) I use with all oher boards (sis, via). seems to me that I just bought a "problematic" board ... :(

Posted: Mon Jan 15, 2007 2:40 pm
by Stedman5040
Ok,

Set up is as follows (at the moment)

K6-3+/450 @ 550 (5.5x100)
DFI K6XV3+/66 1MB cache on
Windows ME
Intel i740 graphics
512mb memory 2x256mb hynix pc133 cl3 set at 2-2-2-5
No memory interleaving
All other wpcredit tweaks on

everest 2.20
MR--------------------------------------307
MW-------------------------------------158
ML--------------------------------------214

Sandra2004 membench 174/174 rising to 186/190

Superpi 256k---------------------------62s
Superpi 1M----------------------------329s

Hotcputester

Score---------------------------------1047
metamark/MHz ratio----------------1.90
integer instructions------------------175
FPU instructions---------------------68
MMX instructions--------------------369
3D now instructions-----------------336
Mem fillrate--------------------------188
mem fillrate c++--------------------186
Memrep movsd----------------------188
Mem FPU 64 bit----------------------186
Mem assembly 32 bit---------------180
Mem fillrate assembly 64 bit-------185
mem fillrate MMX 32 bit------------186
Mem fillrate MMX128 bit------------187
Insertion sort------------------------7k
Shell sort----------------------------271
Merge sort---------------------------492
Heap sort----------------------------291
Quick sort---------------------------1045
Large object sort quick method---235
FFT standard method---------------486
FFT optimised method--------------585

These results at this speed of 550MHz put the board very close to the figures achieved with the EP-MVP3G2. One thing that was noticeable with this board is the sifnificant rise in Sandra memory bench figures with consecutive testing. I have not seen this on the EPOX board. Jim has shown this on earlier notes in this thread.

Stedman

Posted: Mon Jan 15, 2007 3:44 pm
by DonPedro
stedman,

I would be very interested to see a comparison of different via-chipset boards with different amount of onboard cache enabled. I hope the template I sent you will ease filing the bench-numbers but also will help display the results here for public analysis.

Posted: Mon Jan 15, 2007 5:46 pm
by Stedman5040
Well as far as the Via boards go I have as follows

DFI K6XV3+/66
Aopen AX59PRO
EP-MVP3G2
EP-MVP3GM
TMC TIVG+
TMC TIVGF
FIC PA2013
CT-5AGM2

It could take some time to compile results from this lot

All of the above have 1M cache apart from the TMC TI5VGF which has 2M cache.

Unfortunately I do not have a board with only 512k cache apart from the ALI boards I have and the SIS530

Stedman


Stedman

Posted: Mon Jan 15, 2007 7:06 pm
by Jim
@ Stedman : Thanks for the confirmation of the rising Sandra results. Have you tried running Everest after repetitive Sandra runs to see what you get then? I get a significant gain. Note I also am running 512 Meg of SDRAM, (Hynix Low Power); but w/ 4 bank interleave enabled on SP3.

One thing I notice is that you are getting somewhat better Everest Memory Read than I do @ 5.5x100 w/ 512 Meg SDRAM; - 307 vs 303. Memory Write is equal 158 for both of us. I, on the other hand, am getting significantly better Sandra results @ 5.5x100 w/ 512 Meg SDRAM; - 210 - 208 vs your 174 - 174. Would appreciate it if you could post a screen shot of your WPCredit settings when you have time; and I will do the same as soon as SP3 is ready for that sort of thing again.

Could you post a pic of your questionable caps? If you check my SPZ post you will find that I bought a second FIC KA-6100, (that was sold as new, - and looks new too, - except), that has worse looking caps than the one I bought used. The thread has closeups of the caps on both boards.

Re which MVP3 board gives the highest benchmarks, suggest you get your hands on an FIC. The FIC VA_503+, (AT board), blows away any other MVP3 that I have had. Though you only get 1x AGP w/ it.

@ Peter ; I did post a lot of runs of OS Mark that were run on SP3 along w/ my various Sandra and Everest results.

My machines are not that complicated. SP1, forget. It sits collecting dust, (Loaded w/ NT4). SP2 is my ASUS P5A-B machine, w/ 768 Meg of Hynix PC133 CL2 SDRAM. SP3 is my MVP3 DFI K6BV3+/66 machine, w/ 512 Meg of Hynix PC133 CL3 / PC100 CL2 Low Power SDRAM. The old SP3 had a 1Meg Cache board, as rebuilt it has a 2 Meg Cache board, (Both DFI K6BV3+/66 Rev B2); but I have not run any benching on the 2 Meg board yet. Still getting the machine loaded and setup and trouble shooting it as I go.

Exactly what problems did you have w/ your DFI anyway? And what OS were you running on it and what bios? Maybe I can help.

Posted: Tue Jan 16, 2007 8:00 am
by lazy_kalabok
Stedman5040 wrote:
Unfortunately I do not have a board with only 512k cache apart from the ALI boards I have and the SIS530
since i have got the ep-mcp3c2 with 512k, i can help out here. ive allready done some benchmarks, no prob to post them if they are needed.

Posted: Wed Jan 17, 2007 2:15 pm
by DonPedro
@jim,

thank you for offering your help re the dfi-board once again. I already detailed my probs in the "epox mvp3g5 board blown .." thread, you already gave useful advice there. still, the ram issues were not solved completely, 4-way interleave did not work and wpcredit tweaks did not give the same boosts as others get with their via-chip boards. so I gave up, a mountain of other work in the sis/ali arena keeps me busy :)

Posted: Wed Jan 17, 2007 8:26 pm
by Jim
Peter, I have never gotten four bank interleave to work on my DFIs except when I use my two sticks of Hynix PC133 CL3 PC100 CL2 LOW POWER SDRAM. Anything else I put in there doesn't support 4 bank interleave. I think it is because the default settings are for normal power SDRAM, and using LOW POWER at normal power settings is enabling the 4 bank interleave to work. It could be that most of the problems you have encountered are a result of inferior grade SDRAM. I don't know what you are using; but I do recall that you said that you could buy RAM on e-bay for a lot less than what the Hynix was selling for. You ought to break down and spend a few bucks to get some high quality RAM of some sort. -- Crucial, Mushkin, Hynix, whatever. Truth is some boards are less demanding than others regards RAM quality; but ANY board can be pushed further w/ good quality RAM than it can be w/ junk; so its a good investment.

Posted: Sat Jan 27, 2007 6:04 pm
by Jim
Something I have found out since bringing SP3 back up, (at least on DFI K6BV3+/66 boards), is that when using WPCredit to enable the "DRAM Read Pipeline Cycle" setting on a 2Meg Cache board, You must first enable (Offset 50 - 01 bit 3) "SRAM Banks = 2 Banks" or you will get a system crash. So if anyone out there has been getting system crashes trying to enable the "DRAM Read Pipeline Cycle" on a 2Meg board, try enabling "SRAM Banks = 2 Banks" first.

Posted: Mon Feb 19, 2007 7:22 am
by Jim
@ Peter: Just as I am an "ATIdiot", I suspect that you are an "NVidiot". That would explain why you think the P5A is such an easy board to work with, (Try an ATI AGP card in it and see how easy it is!), and why you have had such problems w/ your DFI, (Which likes ATI cards the same way your P5A likes NVidia).

Part of the problem is you seldom post anything about your machines, so we don't know what is in them.

Posted: Thu Jul 19, 2007 4:39 pm
by KGB
Heh, ATi bashing again Jim? ;) Sorry to be out of the loop for a while. Seems like I have a lot of catching up to do. Since this sort of thread is what I was planning anyway, here are some WPCSet Data factoids, and some benchmarks on my K6'er. I realize I have been "pimping" this system for such a long time now, but I do believe I have come to terms with the hardware.

Windows 98se (with SP21a)
K6-2 AFX 500 @550
GA-5AXv5.2 @110FSB
Micron 256Mbit tech 512 MB RAM
ATi Rage 128GL 32MB AGP
Intel 8255x-based Ethernet Adapter
Creative SB Live! Value
3Com Windows Modem TI
Seagate ST36811A (6GB @ UDMA2)
NEC 24X CDROM
IOMega 100Zip IDE

WPCredit Set markers which are set in Registry
Reg:07 Data:24|Reg:40 Data:5F|Reg:41 Data:46|Reg:42 Data:00

R:43 D:76|R:48 D:1F|R:49 D:CF|R:4A D:22|R:53 D:00|R:5E D:18

R:72 D:84


K6Speed .88 (at startup with Command Line, Write Allocation: Enable Range 512MB, NO 15-16MB Hole, Fast GEWBED)

RAM = Cas2-tRCD2-tRP2-tRAS4-tRC7

Averaged Everest Home v2.20.405 and SiSandra 2002 SP1 6.8.97
Read: 382-382-382-382-382
Write: 152-152-152-152-152
Laten: 199.7-199.6-199.5-199.4-199.7-199.6-199.5-199.4-199.5-199.5-199.5-199.4

Memory Bandwith: (211/196)-(195/194)-(198/199)-(200/198)-(200/198)

I think I've really got over this hurdle. It would not have mattered if I got a + CPU, as I had not figured out the CPCredit software, as proficiently as I do now. If you guys have any questions, just ask. I'll be on the lookout for new benchmarks as well, CPU mark etc.