-
Posts
528 -
Joined
-
Last visited
Content Type
Profiles
Forums
Events
Blogs
Posts posted by IanCutress
-
-
Nope, I write code
I deal in performance first
I have written a couple of benchmarks (not on HWBot) that act like this, to see where L2/L3 cache plays a role as you increase problem set sizes. Sometimes it's better to test this way in order to maximise throughput and minimise out-to-memory transfers, and you get an overall peak value at the end. I change my problem set sizes over time as well.
If anyone has ever looked at n-body calculations, the complexity of a problem rises as a square of the number of bodies in a simulation. At some point your memory requirements will be more than L2, more than L3, and then require looking back to main memory. Keeping out of main memory is a lot quicker in simulation steps. Or a good coder can hide that latency
-
Law of diminishing returns does apply.You can flip to slow mode after a second and a half and save LN2 and beating on your hardware.
Law of diminishing returns in the context of code applies to performance.
You mean to apply it to hardware stress, which is valid but confusing without an additional qualifier in your phrasing.
-
Well it depends if performance decreases over time. Law of diminishing returns doesn't apply: the time to do a calculation should stay the same, as you don't need to throw more power to keep a game running at 30fps indefinitely. (3D analogy for 2D benchmark, but yeah same principle applies). But as we know with older 3D stuff like Aq, fresh boot does it best.
Also each bit of the benchmark could be calculating something different. This is where L1/L2/L3 caches come in, and the benchmark keeps ramping up the P95 size until it overflows L3 cache. It's going to affect different CPUs at different times. But in XTU's case it seems to be doing the same calculation over time and finding the peak score.
-
OK guys there are two ways to calculate a score in a benchmark:
1) Run through sequence. Calculations per second overall is score. Like overall frames per second.
2) Run through sequence bits, calculate per second in each bit, score is the peak in each bit over all bits. Like peak frames per second.
XTU clearly does number two. So technically you can still get a higher score further down the benchmark, it has to beat the peak bits (peak FPS) at the beginning to register.
-
Thanks, borandi. But I have some point about it.
The hybrid only add the discrete VGA at the end, running recent versions of GPU-Z, take a look: http://hwbot.org/image/1049323.jpg
But it makes possible to create a category (in the case, HD 7650D + HD 6570 Dual, or another setup, like HD 7660D + HD 6670, for example), because the setup is detected like this.
Considering the fact of Trinity and Richland APU can be used on hybrid setup, I suppose they can participate in the Country Cup, regardless of whether they have a special identification or not. I think the Trinity and Richland could participate.
Yup I know, that much is obvious. It's a small land of unqualified statements here where the ref only reveals the goalposts at various points after the kickoff when you've already started building the team!
To this extent though, you need one of the following here that creates a new name:
http://www.cpu-world.com/info/AMD/Recommended_graphics_cards_for_AMD_dual-graphics.html
-
What's the correct GPU set in HWBOT engine for hybrid HD7560D + HD6570? HD7560D x2 crossfire?
Massman has essentially stated that hybrid setups that do not change the name in gpuz will nnot qualify for the hybrid part of the contest. You need to go back to llano to get an apu that qualifies for that round.
-
Only if it makes a new name in GPU-Z. Chances are the 7xxxD/7xxxG processors do not.
-
Le sigh. That wasn't stated at any point. AMD considers 7xxxD combinations as hybrid. Please adjust the words in the contest to state your 'clarification'.
-
The 6xxxD hybrids get a new name, the 7xxxD hybrids do not. So I guess 7xxxD crossfire?
-
No Competition Background for Country Cup.
-
It didn't work on quad cores and up last time I used it.
-
This is what happens when people get that pop up window and just accept 'participate' after submitting.
-
Yeah, that wasn't expressed clearly at all.
What you mean is: Any Pentium/Celeron part must be a member of a family that includes an i3 part.
So anything based on Nehalem architecture and up (Arrandale, Clarkdale), but not Wolfdale.
-
It's just a bonded silicone, nothing fancy like a lotus epoxy.
-
Question for stage 3.
i5 750 for example not allowed? Only 655K?
That is an odd restriction - imo any i5 for stage three should be allowed, any generation. Though it's obvious some > others.
For stage 4, Pentium cpu's from LGA 775 socket are alright? Example Pentium E5200, E5400, E6300, E6500 , etc.Or it's about Pentium SB/IB/Haswell core?
The stage has limitations set to Core i3/i3 Mobile only right now, but any that are labelled Pentium by ark.intel.com should be ok on the rules as written in text. All in this link for Legacy Pentiums: http://ark.intel.com/products/family/78132/Legacy-Intel-Pentium-Processor/desktop
-
Working with old computers
-
my standerd speed of cpu is 4 ghz so looks like no overclocking in stage 8 only lots of tweaking
I initially read that as 'lots of twerking'. Bonus points for all submissions with photos/videos of twerking.
-
That was discussed on the forum with no conclusion, but almost all results for the benchmark are one instance.
-
Dread to think what the backup score is. Is 2K going to be possible on 6x ?
-
Team account? Who are members of the team and do they have individual accounts? There was an issue a few years back: if you're going to make a team account then you should turn off the individual scores and not submit new ones on them
-
-
The mathematics works, but there are two issues:
1) The % mark is a bit misleading and not technically accurate.
It's not a unit that carries on after multiplication of a scalar quantity.
+54.34% means nothing in that regard, as it's not +54.34 percent of anything.
54.34 points would make more sense.
2) Super Pi:
SniperOz = 423.441 / 311.703 = 135.85%
Zzolio = 423.441 / 314.141 = 134.79%
You're changing the denominator, meaning that moving from 315 seconds to 314 seconds has less effect than 310 to 309:
e.g. 423.441 / 315 = 134.425%
e.g. 423.441 / 314 = 134.854%
difference = 0.429%
0.429% of what? you've changed the denominator
e.g. 423.441 / 310 = 136.594%
e.g. 423.441 / 309 = 137.036%
difference = 0.442%
0.442% of what? you've changed the denominator
It works when counting up, line in Cinebench:
e.g. 10.00 / 8.93 = 111.982%
e.g. 10.10 / 8.93 = 113.102%
difference = 1.12%
1.12% of what? 1.12% of 8.93
e.g. 11.00 / 8.93 = 123.180%
e.g. 11.10 / 8.93 = 124.300%
difference = 1.12%
1.12% of what? 1.12% of 8.93
Bottom line: The way of calculating the Spi results means the faster your result, the exponentially better your % is.
Next time perhaps. Or maybe it's better that the one who is in the lead deserves earning more overall points for their 1st. (Another reason the points system in F1 has changed over the past decade, so 1st places are more significant.)
-
I think those numbers were just done with one of the setup at default settings/speed which makes sense if you want to do Overclocking in percentage.Yes there were baseline scores done by MSI:
SuperPI 32M: 7min 03.441 sec
Cinebench R11.5: 8,93 points
3DMark Fire Strike: 9891 points
And then keep in mind that 32m and FS counted 40% each to the final score and Cinebench 20%.
This doesn't make sense looking at the numbers.
Final score for SniperOz = 54.34+28.65+58.25 = 141.24, meaning 100% on each
SniperOz's scores:
Spi : 311.703 seconds, +54.34%, meaning 311.703/(1-0.5434) = 682.66 seconds original calculating back
Cinebench: 12.79 points, +28.65%, meaning 12.79/1.2865 = 9.94 points original calculating back
3DMFS: 14404 marks, +58.25%, meaning 14404/1.5825 = 9102 points original calculating back
Final score for Zzolio = 53.92+29.25+54.6 = 137.77, meaning 100% on each
Zzolio's scores:
Spi: 314.141 seconds, +53.92%, meaning 314.141/(1-0.5392) = 681.73 seconds original calculating back
Cinebench: 13.06 points, +29.25%, meaning 13.06/(1.2925) = 10.10 points calculating back
3DMFS: 13500 marks, +54.60%, meaning 13500/(1.5460) = 8732 points original calculating back
There's something wrong with the maths - people are being compared to different scores and the % are all wrong. They're not even near the numbers you quoted. It follows all the way down the leaderboard - people are being compared against different numbers.
More than welcome for someone to correct my maths if they think I'm wrong.
-
This way it does not matter if you beat someone else in one of the single benchmarks. However you have to have very good and high scores in all three benchmarks to win.
But a 1st+1st+6th lost to 5th+9th+1st. Any way you slice it, that's just a little odd. If Alva had dived the last benchmark and had 10th then I'd understand, and as Cine was only 20% you could argue that the second round was more 1st vs 4.5th.
I appreciate that the way it was run is more compelling viewing (harder to predict, etc), and pulling nothing away from T0lsty because he had to show the skill and push the gear to get where he was. Anyone looking at the positioning, or new people to the scene not understanding how it works, is going to think 'wtf'. I know I did - I only caught the tail end of the results.
Kind of odd they didn't select:
Spi 32m: 7min
Cine: 9.00
3D FS: 10000
Why not round numbers? I bet if you stuck these numbers in, someone would swap a place.
In b4 'these are the rules, either work with or work without'. That's true, no harm in discussing/debating the merits, even if pointless/no impact
XTU tweak (big boys already know time to share)
in Intel CPU Overclocking
Posted
I was saying that to Scott's misrepresentation to diminishing returns on performance without the qualifier on hardware to which he was referring.