SuperPI 32M for testing CPU and system speed and stability

[Clevor]

Those some are some golden sticks you got there...very very impressive, and just 3200 at that.

Yeah, there was talk of guys early on running 270-275 with VX3200, but nobody showed any 32M runs. I knew the sticks had potential, as they did 264 @ 3.4 volts on the 9NDA3+ with VDIMM mod (mainly they scaled).

Let me say though (as coming from anyone else it may sound like sour grapes), why does everybody seek 270, 2-2-2-? I mean my CPU currently maxes at 2910 or around 265x11. 270x10 is certainly slower in everything.

If all you can muster with VX4000 or whatever is 261 that's fine as most people would take 261x11 or 2871 mhz any day. At 2-2-2- that rig is plenty fast. What's more important is doing stuff at the lowest voltage possible, especially these days.

 

[Sucka]

Here's some 32m madness with the new CPU, what do you think of that G.Skill cruising along at 2.6v

With some tweaks and a little more testing for max clocks i think i could squeeze some more, we'll see. I get so bored running long PI sessions.

 

[Eldonko]

I have done some testings and preliminary analysis on TCCD and UTT, including SuperPI runs.

Testing UTT and TCCD memory modules in Winchester and DFI NF4 Ultra-D setup


I don't have the UTT any more, because I don't like using 3.3+ V on the memory modules with the DFI 5V Vdimm option, too much stress is put on the DFI motherboard voltage regulator. Since the DFI board and TCCD memory I am using can run 300 - 320 MHz 2.5-3/4-3/4-x 1T which is better or equal to UTT/BH-5 at around 250-260 MHz with tight timing. I prefer the flexibility of TCCD over a wider memory frequency range and timing.

I am also not trying to squeeze the last 0.01% of benchmarking results (don't have the time and the video card and cooling hardware to get to the very top).

That is some great analysis hitechjb1, good job

 

[El<(')>Maxi]

Hite it's very hard to read the graphs now. I know what you were trying to do, get around the silly image limitations here but in this case I think full size in a new window was better.

Another thing I'm curious about is Sucka's FX-57. That's a SanDiego core yet look at where it fits into the performance per Mhz. It's not consistent with other SanDiego (lower clocked) runs. I think that the formula is good but when encountering higher CPU clocks it needs some adjusting.

Super PI itself may be part of this. Look at the top scoring PI Japanese teams and their results. It takes a larger jump in system power beyond a certain point in Super PI to see a similar time gain. So the bench test itself might be skewing the CPU results and making high clocked CPU appear to perform worse per mhz.

Ideally I think you might want to consider adding a few more bench tests to broaden the results. Or you could have everyone run at the same MHZ and compare that way.

 

[hitechjb1]

SuperPI 32M uses much more memory than SuperPI 1M (268 MB vs 8.4 MB), so the run time can be affected to a much greater extent by memory performance and other system components and parameters such as memory frequency and timing, system bus, chipset and maybe even the hard drives if paging occurs. The intent of using SuperPI 32M (in this thread) has been for testing stability and at the same time for benchmarking the overall system performance (CPU, memory, chipset, (HD)).

For benchmarking the raw speed and performance per clock cycle of mainly the CPU, including its caches, SuperPI 1M - 4M should be used.

For those runs with the CPU above 3000+ MHz, it is understandable that if other system components and parameters are not "keeping pace" and "in balanced" with the faster CPU, the performance per clock of the CPU may be slewed lower.

SuperPI 32M is a relatively "quick" way to test speed and stability of CPU, memory and system. SuperPI 1M does not stress the CPU, memory and system enough. System that passes SuperPI 1M may have a hard time passing SuperPI 4M, 8M, 16M and 32M.

For SuperPI 32M, in additional to CPU frequency, memory frequency, memory timings affect the total time as there is heavy memory access throughout the computation. I think system paging is also involved (to some extent depending on total memory size of the system), so the chipset and the system bus are involved. Unlike the small FFT in Prime95 (8K - 32K), only the CPU and the caches are stressed. The large FFT (1024-4096K and blend) in Prime95 can test both CPU, memory and system (paging) also, and certain Prime95 setup can stress the CPU, memory and system more with extensive testing in time.

SuperPI 32M runs longer (20 - 30+ min. depending on CPU processing power), requires more memory (~ 268 MB compared to 8.4 MB for SuperPI 1M) and system resources, and tests the CPU, memory, system and their stability more than that offered by SuperPI 1M which completes in a much shorter period of time (about 20 - 30 sec).

 

[Sucka]

Another thing I'm curious about is Sucka's FX-57. That's a SanDiego core yet look at where it fits into the performance per Mhz. It's not consistent with other SanDiego (lower clocked) runs. I think that the formula is good but when encountering higher CPU clocks it needs some adjusting.

So what now my score sucks

 

[hawtrawkr]

So what now my score sucks

yes thats exactly what hes saying. now kindly package it up and send it to me for me to confirm its suckiness!

 

[El<(')>Maxi]

This should be a Sticky, replace the "whats your voltage?" poll as that's kinda useless if you ask me, especially with all the different cores now.

 

[Sucka]

That or Sticky it over at the Benching forums. Nice to keep track of 32m runs not only for stability testing, but to use as a benchmark.

 

[El<(')>Maxi]

I don't care where it is but it should be stuck.

 

[Sucka]

With over 4,100 hits on this thread, you would think more people would have submitted.

Here's my 4th submission. Air cooled X2. Funny thing is i hate this benchmark, takes way to damn long

 

[Sucka]

Someone stick this!

We appreciate your hard work on this

 

[Eldonko]

Another vote for sticky and bump!

 

[Eldonko]

Update

Got an update, hitechjb1.

CPU: San Diego 3700+ CABGE 0515
memory: OCZ pc3200 Gold (BH) 2 x 512 MB
motherboard: DFI LanParty UT NForce4 Ultra-D BIOS 0618-3
cooling: H20
OS: Windows XP Professional SP2

- CPU: 2889 MHz = 263 MHz x 11, 1.7 V
- memory: 263 MHz, 2-2-2-5 1T, 3.6 V
- SuperPI 32M completed in 25 min 28.922 sec

 

[SavageBasher]

 

[ropey]

Considering the price of my 2 year old machine.

See below sig in red.

 

[d]g[ts]

CPU: San Diego 3700+
memory: OCZ pc4000 vx 2 x 512 MB
motherboard: DFI LanParty UT NForce4 Ultra-D BIOS 0623-2
cooling: vapo LS
OS: Windows 2000 SP4

- CPU: 3217 MHz = 292 MHz x 11, 1.7 V
- memory: 230 MHz, 2-2-2-5 1T, 3.2 V
- SuperPI 32M completed in 23 min 01.250sec

memory isnt playing well with 183 divider, so ran at 166 divider. left the voltage way too high for a measly 230. the overall time is nice tho.

And why the heck aint this stuck anywhere yet?????

 

[El<(')>Maxi]

Ohhh, hey d]g[ts that's looking very good! Looking forward to seeing some 3D stuff

 

[d]g[ts]

Ohhh, hey d]g[ts that's looking very good! Looking forward to seeing some 3D stuff

me too, but for now i just got the trusty old voodoo3 2000 PCI in there. till new vids can arrive.

 

[El<(')>Maxi]

CPU: FX55 San Diego 0518TPMW
Memory: Mushkin LII 2 x 512 MB BH-5
Motherboard: DFI LanParty UT NForce4 Ultra-D BIOS 4.14.3
Cooling: Baker SS loaded @ -58°
OS: Windows Server 2003

- CPU: 3.42 GHz - 262 x 13, 1.76V
- Memory: 262 MHz, 1.5-2-2-3 1T, 3.65 V
- SuperPI 32M completed in 22 min 5 Sec