ridiculous things with mvp3

Discuss software and how to tweak more performance out of your system.
User avatar
KachiWachi
K6'er Elite
Posts: 507
Joined: Wed Sep 21, 2005 10:53 am
Location: Pennsylvania, USA

Post by KachiWachi » Tue Sep 11, 2007 7:39 am

DPE only matters if you have a flat memory model OS (like W2K and XP).

How is your EWBEC set?

Thanks.
Moderator - Wim's BIOS

PC #1 - DFI 586IPVG, K6-2/+ 450 (Cyrix MII 433), 128 MB EDO. BIOS patched by Jan Steunebrink.
PC #2 - Amptron PM-7900 (M520), i200 non-MMX, 128 MB EDO
PC #3 - HP8766C, PIII-667, 768 MB SDRAM
PC #4 - ASUS P3V4X, PIII-733, 256 MB SDRAM
PC #5 - Gateway 700X, P4-2.0 GHz, 768 MB PC800 RDRAM
PC #6 - COMPAQ Evo N1020v laptop, P4-2.4 GHz, 1 GB PC2700 DDR
PC #7 - Dell Dimension 4600i, P4-2.8 GHz, 512 MB PC2700 DDR
PC #8 - Acer EeePC netbook, Atom N270 @ 1.60 GHz, 1 GB RAM
PC #9 - ??? ;)

Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim » Tue Sep 11, 2007 4:55 pm

I assume you are not talking to me, because what you said, (KachiWachi), is utterly meaningless to me. DPE??? Wha Dat?
EWBEC, a bit easier, something to do with settings that can be enabled through K6Speed, (though apparently they are set backward there), that nobody ever uses because there is no apparent gain from doing so. If you know something about these things; and can be persuaded to type more tha three sentances, Kindly explain them for those who don't.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card

socket7!
Newbie K6'er
Posts: 15
Joined: Sun Jun 10, 2007 4:18 pm

Post by socket7! » Tue Sep 11, 2007 5:35 pm

EWBE Control (EWBEC): I have both Global and Speculative enabled. According to "AMD K6 proc. BIOS Design" performance is best when:

GEWBED and SEWBED are set 1 (enabled) and Write Ordering set to None

KachiWachi: Does your comment means, that tests done in Win98 environment will be much different?

Jim raised a very good question: what is important for system performance? Highest pi results are not accompanied with highest everest score or highest CPUMark. At least in my case. It would be good to know, whether anyone can confirm that results on his system. I'm talking here about influence of write allocate and data prefetch on your benches.

I'm also wondering whether my mobo chipset configuration is common among mvp3 boards: I have VT82C598MVP but with VT82C686A? I would rather expect Vt82c586b southbridge.

and a small abstract from "AMD K6 Processor BIOS Design" about DPE (Data Prefetch Enable):

"DPE must be set to 1 to enable data prefetching (this is the
default setting following reset). If enabled, cache misses
initiated by a memory read within a 32-byte cache line are
conditionally followed by cache-line fetches of the other line
in the 64-byte sector"

I can't comment on that, because I'm not an engineer. :-(

And I also don't know what a "flat memory model OS" is. :-(

DonPedro
K6'er Elite
Posts: 575
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro » Tue Sep 11, 2007 6:41 pm

@socket7

how to measure system performance? yes indeed, this is a good question. if you want to know how good a system is at a specific task then you should look for a bench-programm that comes close to what you have in mind. if you want to get answers what is theoretically possible than go for the synthetic benches (like everest).

super-pi does a real task, its outcome has a practical meaning - numbers of digits of pi. there is nothing artificial about that. super-pi shows the calculation strength of a cpu (floating point unit), will surely rise with higher cpu-clock and has proven that it likes on-die cache and onboard cache (the more the better) and a fast memory access.

everest has its merits because you can quickly evaluate what some tweaking here and there yields in memory performance gain/loss.

the results of cpumark overemphasise the amount of on-die cache and onboard-cache. it is definitely not useful for answering any other question.

we have done a lot of system-benching in the past on the sis530 chipset. because you can not determine how fast a system is by using only one benchmark we have chosen to run the systems on a whole bench-suite. if you are interested in the impact of cpu-speed, size of on-die cache and impact of ram-speed just look here. although the numbers over there are only valid for sis530-chipset boards you will get an idea on the validity of a benchmark to measure system speed and how various benches react differently to some change in the system-setup.

User avatar
Stedman5040
Veteran K6'er
Posts: 271
Joined: Mon Sep 05, 2005 4:22 pm

Post by Stedman5040 » Wed Sep 12, 2007 2:55 am

I did some testing with and without "data prefetch" and with and without "write allocate" on my EP-MVP3G2. My operating system is Windows ME and the results are not the same as being reported with Windows XP.

1. Using a K6-III+/450 @ 550MHz (5.5x100) with everything enabled and wpcredit tweaks enabled

Everest (2.20) MR/MW/ML 294/154/223

Superpi (1M)comes in at 5m10s

2. Using the same set up but at 600MHz (6x100)

Everest (2.20) MR/MW/ML 294/134/224

Superpi (1M) comes in at 4m59s

9.1% increase in cpu speed and 3.55% decrease in time for Superpi (1M)

3. Using the same set up but at 617MHz (5.5x112)

Everest (2.0) MR/MW/ML 332/173/194

Superpi (1M) comes in at 4m33s

Increase in cpu speed is 12.2% decrease in superpi (1M) is 11.94%. Clearly the extra memory bandwidth here is having much more of an influence than increased cpu speed for Superpi calculations. This is clearly shown with the result below.

4. Using the same set up at 575MHz (5x115)

Everest (2.0) MR/MW/ML 338/146/192

Superpi (1M) comes in at 4m54s

Increase in cpu speed is 4.5% and decrease in superpi is 5.2%

Disabling data prefetch on the cpu operating at 550MHz has the following effect.

Everest (2.20) MR/MW/ML 243/124/142

So memory read is down by 51
Memory write down by 30
Memory latency down by 80

Superpi (1M) 5m13s

Disabling write allocate as well as data prefetch at cpu speed of 550MHz gives

Everest (2.20) MR/MW/ML 241/78/142

Memory read down by 53 (18%)
Memory write down by 76 (49%)
Memory latency down by 80 (36%)

So what would you guess is the result of Superpi (1M) under these conditions for Windows ME :?:

Stedman

DonPedro
K6'er Elite
Posts: 575
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro » Wed Sep 12, 2007 4:56 am

if you ask that way then you probably experienced a new record.

User avatar
kalabok
Senior K6'er
Posts: 147
Joined: Mon Nov 06, 2006 11:40 am
Location: germany

Post by kalabok » Wed Sep 12, 2007 10:08 am

do i understand it right - 4:25 with no l3 at all? just disabling data prefetch and WA? ... :?

what the mvp3 pcr tells about offset 6c? what is that thing? socket7, which configuration u used before setting it to 08? some information would be helpful here.

User avatar
Stedman5040
Veteran K6'er
Posts: 271
Joined: Mon Sep 05, 2005 4:22 pm

Post by Stedman5040 » Wed Sep 12, 2007 3:57 pm

No records here with Windows ME. As expected the performance drops. Time for Superpi (1M) with K6-III+/450 @ 550 (100x5.5) is 5m25s, which is better than I would have expected with the lousy looking memory write.

Going back to the SIS530 chipset using the same cpu with 512mb memory, with WA on and Prefetch on and wpcredit tweaks applied we get

Everest (2.20) MR/MW/ML 304/123/223
Superpi (1M) 5m36s

So even when the EP-MVP3G2 is crippled it still manages to beat the SIS530 into touch given the same cpu speed and fsb.

So anyway why should XP give a performance boost to calculating pi when WA is off and when prefetch is off when Windows 98/ME doesn't?

Stedman :?

Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim » Wed Sep 12, 2007 6:11 pm

@ Socket7 : Thanks for the enlightenment. Unfortunately a lot of what KachiWachi says is Greek to me. Shame that, because he is quite knowledgable, but prone to be too technical for this old man. I suppose I could search online for "DPE" and things like that; but who has time?

@Stedman : Answer : I don't know. But this I do know, you will get about 10 extra points on memory read w/ XP all else being equal.

Also re the MVP3 results, the only comparable results in the ones you posted are those for 5.5x100 and 5.5x112 because we know that the mutiplier chosen has an effect of its own independant of FSB speed and other considerations; wherein for reasons unknown the 5.5 multiplier will give better results all else being equal. Comparing just those two we find that with a 12.2% increase in FSB speed you got a 13.553% decrease in SuperPi run time, (310 seconds over 273 seconds), which would indicate that the results are almost directly proportional to clock speed, with some small additional gains to be attributed to improved memory bandwidth and / or latency reduction.

NOTE : If you want to make comparisons using the 5x multiplier then you should be comparing 5x100 with 5x112 and 5x115. Similarly making comparisons w/ the 6x multiplier you might try 6x100, 6x103, and 6x97.5.

@Kalabok : Setting offset 6C to 08 Means using WpCredit to set the "DRAM Start Cycle" to "0=With Cache at 66 Mhz" as opposed to "1=After Cache at 100 Mhz" and also setting "Fast single-Cycle" to "1=On pipelined" as opposed to "0=Off" Which means changing the default settings of both. (in the case of a DFI K6BV3+/66 at least) Also thinking about it, the "Fast single-Cycle" bit may have a disproportionate effect when doing "SMALL" reads and writes.

@ All Concerned : Also having thought about all this some while I was coming home from work tonight, I concluded : 1) SuperPi is essentially a number crunching bench. 2) It is memory intensive only in the sense that it is continually doing SMALL writes and reads to and from memory as it stores and pulls from storage numerical values used in calculating the value of Pi. 3) Apparently where SMALL reads and writes are concerned, latency is more important than read speed or write speed. 4) How much of a real world value this benchmark has is an open question. It depends I suppose on what percentage of the reads and writes your machine does in normal operation, doing whatever you normally do with it, are SMALL ones.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card

Post Reply