• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Beat the gimp: K10.5 SuperPi 32m challenge!

DrMrLordX

DrMrLordX

Lifer
All you Bloomfield/Lynnfield guys will kill me at this, so I already know you'll win. Can't stop you from humiliating me though . . .

Anyway, got my replacement board, so I turned in the best SuperPi32M time I could whip up on a moment's notice using the Sempr0n 140:

pi32m1.png


Didn't get cpu-z up for the shot (oops) but it was around 3.8 ghz, 2.75 ghz NB, uh, DDR3-1640 or so. This time was better than my x4-635 by about a minute.

Anyway, I'd like to see if any of you Deneb/Thuban owners can beat that on air. Or hell anyone else with a K10/K10.5 for that matter. This might even beat some Penryns and (locked) Clarkdales, who knows. Maybe it won't.

The 1M time for the chip sucks (high 19s is the best I can get) so let's get that out of the way. There's a method to my madness, sort of.
 
I'll chime in later with some unlocked to dualcore Sempron 140 results.

Is that 140 of yours unlocked or you simply running it as a single core?
 
Here you go. I almost thought I corrupted my OS while tightening subtimings. This is why I don't like to attempt pi contest once a 24/7 system is up and running! :hmm:

1090T @4.08 GHz (17x240) / 2.88 GHz NB

 
Q6600 @ 2.4GHz (stock everything) in Win7 64bit I get 21m 29.500s

Comparison to QX6700 @ 4GHz (vapoLS) going back to 2006 (wow has it been that long already!?) I got 14m 24.625s for 32m in sPi.
 
@Idontcare: Not sure what you're comparing them to (to each other?) but 14mins+ for a 4.0 GHz QX6700 seems slow. ^_^
 
Nice time on the Thuban. I have seen some 4 ghz Denebs up in the 18 minute range so I was curious about what was going on . . .

And yeah, beating a penryn/kentsfield is a pipe dream but we can always dream.

formulav8 said:
I'll chime in later with some unlocked to dualcore Sempron 140 results.

Is that 140 of yours unlocked or you simply running it as a single core?
Click to expand...

Single core. It still won't unlock. Grumble grumble. Now if only I could get it stable enough to do a 3.9 ghz 32m run . . .

alyarb said:
http://www.xtremesystems.org/forums/showthread.php?t=252727&page=2
Click to expand...

That's Chew. Chew automatically wins.
 
AdamK47 said:
Why do people still use Super PI?
Click to expand...

Why not? It's like a common thread that connects CPUs from the present to the past. Like Idontcare said, has it been that long already?

Besides, Pi 32M is the only bench where a crappy $25 single-core CPU can sort-of compete. The fact that it can is fascinating in and of itself.
 
Yean, i7s kill all in 32m. I've seen an i7 in the 5.3 ghz range get around 7m+ in 32m, which is pretty scary stuff.

But I've also seen i3s get over 20m.

Lopri, I haven't bested 17m 38s yet despite more tweaking. Looks like you have beaten a crappy budget CPU handily. Woohoo! Seriously though, when I started seeing 955s and 965s in the 4 ghz range posting 32m times in the 18-19s range, I started to wonder exactly what was going on with those CPUs. Thuban should, at least in theory, turn in the same 32m times since the architecture is similar on a core-per-core basis. 16m with only DDR2 is pretty damn good!

Need to work on my NB some more . . .
 
Last edited:
It looks like I have to surrender the SuperPi 32m war to the good doctor.

My Deneb at 3.813GHz core / 2.706 GHz uncore only gets to 19 minutes (I screwed up the shot, the window was too small. Doesn't matter anyway, the performance was awful)

Our clocks are mostly the same, so I suppose L3 plays no role here, and I am limited by my RAM? Perhaps I am limited by having only 2GB of RAM? Or is it the speed of the RAM, at only 5-5-5-15 DDR2 @ 984 MHz? (I can't seem to get any closer to 1066 than that)

SuperPI32_OC.JPG
 
jvroig said:
Our clocks are mostly the same, so I suppose L3 plays no role here, and I am limited by my RAM? Perhaps I am limited by having only 2GB of RAM? Or is it the speed of the RAM, at only 5-5-5-15 DDR2 @ 984 MHz? (I can't seem to get any closer to 1066 than that)
Click to expand...

RAM is definitely going to play a role here. You might want to ask Lopri about the subtimings he used since he is using DDR2 on his Thuban, at a lower memory clockspeed than yours with identical timings. Granted, his NB is running 180 mhz faster and his total clockspeed is 200 mhz faster, but 2 minutes in 32m is a pretty big gap.

I have a theory that in apps with a huge working set, AMD's inclusive cache architecture hurts more than it helps when L3 is present. Lopri sort of shot that idea in the head with a rail gun. Sort of. Thuban may have optimizations that most of us don't know about yet.

Every Deneb I've seen running 32m has struggled compared to Propus and . . . Sargas, apparently.
 
Last edited:
Well, Pi tweaks are difficult to explain. Not to be obnoxious, but I don't exactly know what I do (but just do) and usually follow the instinct, and every run differs in results and I simply take screenshots. On OS level I go to msconfig and do a minimal startup, and set priority to "High". In the above 32M run, I assigned first two cores (0,1).

K10.5 is the most memory sensitive CPU I've ever experienced. Thanks to its inefficient NB (or very efficient, depending on how you look at it), clock frequency is no longer the king with this architecture (at least in current state of production silicon). A board's BIOS plays a big role as well. I observe FSB strap-like behaviors as HTT is being raised. This also varies per CPU revision, too. Some starts around as early as 220 HTT, and others around 240. Then it occurs again around 260~300 HTT depending on boards. Also my experience tells me there is a sync issue between HT multiplier and NB multiplier, which again varies per chipset/board/BIOS.

This is why I think the new 8 series chipsets better suited for Thubans than 7 series chipsets. I don't think board makers are interested in tuning BIOS for older models more than they have to. In my case I've had this board over a year now and kind of learned how it behaves, so that saves me some time.

This guy here seems like an expert.

http://www.clunk.org.uk/forums/revi...air-iv-formula-living-review-5.html#post93816

 
Last edited:
He also gets the same wPrime numbers as mine, with CPU frequency 250MHz less. (3824 MHz v. 4080 MHz)

 
DrMrLordX said:
All you Bloomfield/Lynnfield guys will kill me at this, so I already know you'll win. Can't stop you from humiliating me though . . .
Click to expand...

must resist slaughter... 😛

But SuperPI is uber favored on an intel platform.
 
Last edited:
lopri said:
Well, Pi tweaks are difficult to explain. Not to be obnoxious, but I don't exactly know what I do (but just do) and usually follow the instinct, and every run differs in results and I simply take screenshots. On OS level I go to msconfig and do a minimal startup, and set priority to "High". In the above 32M run, I assigned first two cores (0,1).
Click to expand...

Hmm, interesting. I haven't bothered streamlining the OS boot or messing with priorities. Best I could do was around 17m 32s.

K10.5 is the most memory sensitive CPU I've ever experienced. Thanks to its inefficient NB (or very efficient, depending on how you look at it), clock frequency is no longer the king with this architecture (at least in current state of production silicon). A board's BIOS plays a big role as well. I observe FSB strap-like behaviors as HTT is being raised. This also varies per CPU revision, too. Some starts around as early as 220 HTT, and others around 240. Then it occurs again around 260~300 HTT depending on boards. Also my experience tells me there is a sync issue between HT multiplier and NB multiplier, which again varies per chipset/board/BIOS.
Click to expand...

I haven't seen the strap issue before, but that is an interesting observation. Or, at least, I haven't seen it on a 790FX-GD70 which is the only AM3 board I've used. The sync issue . . . hmm. Haven't really noticed it either, but again, that's going to vary.

I *have* seen different BIOS revisions change OC limits.

This is why I think the new 8 series chipsets better suited for Thubans than 7 series chipsets. I don't think board makers are interested in tuning BIOS for older models more than they have to. In my case I've had this board over a year now and kind of learned how it behaves, so that saves me some time.
Click to expand...

While that may be true, your Thuban on a 7-series chipset kills a lot of Denebs I've seen run 32m. I've seen a guy with a 4 ghz Deneb do 19m in SuperPi, though without knowing more about his NB speed and RAM timings, it's hard to say why he did so poorly.

This guy here seems like an expert.
Click to expand...

I concur. That's a damn good time for only 3.8 ghz.
 
richierich1212 said:
1st run, no tweaks, Win 7 64-bit:
Click to expand...

Splat. I am pwned.

Where are the Deneb numbers? It's pretty obvious that Thubans are going to kill my arse just from clockspeed alone.

aigomorla said:
must resist slaughter... 😛

But SuperPI is uber favored on an intel platform.
Click to expand...

That's why I put the line in the OP about Bloomfield/Lynnfield. i7 has gotten down in the 7m territory, if not better. It's just no contest.

The thing is, both of my K10.5 chips have been awful at SuperPi 1M, and I think it's a cache issue. What surprised me is that the few 32m numbers I saw for Denebs floating around seemed a bit worse than mine. Then Thuban shows up and gives me a proper stomping . . . but in the process, Thuban chips seem to be stomping Deneb chips at 32m as well.

17m at 3.8 ghz? Tweaks or no, that's pretty impressive. Why is Deneb so curiously bad at 32m? Or are they just not being tweaked properly?
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top