Clarification in PCMark'05 about Tweaks // Tricks allowed

[Moose83]

In gen usage you cant beat areca with RST or Raid Expert;)

VS can get crazy on RST sometimes, but in good Moments areca can score 2G VS too.

What about showing disk Manager on Screenshot? Then its clear if RST has been used:)

[IanCutress]

VS can get crazy on RST sometimes, but in good Moments areca can score 2G VS too.

 

'Good Moments' are often called 'bugged runs' in the world of 3D. If this RST glitch happens 1 in 20, then anyone could set up a 3D run on a card that bugs 3D01 one time in 20. But in 3D we see it as a black screen and call it a bug. In PC05, it's not that easy to see, but if it's 1 in 20, then we should be calling it a bugged run due to the inflated score.

 

I still don't see how two SSDs can hit 1300 MB/s with RST without something going on. Even with write back and them being clean. That's technically faster than the two drives could do together.

[crustytheclown]

So every sata2 drive is limited to 300mb/s and sata3 to 600mb/s ? double them for raid0 with two of each?

then about 99.5% of the pcmark runs should be deleted...

[Moose83]

Its no bugged run then! Areca Cache never scored the same in each run! It differences, same like rst.

And SSDs on southbridge are faster than ramdrive, they are only fast with Controller.

With 2 SSDs 1,3G VS is easy without tricking anything but activate write back in RST! On P45 i score 1,2 with 2 and 1,5G with 4 SSDs;)

Edited November 14, 2012 by Moose83

[I.M.O.G.]

but in good Moments areca can score 2G VS too.

 

For the sake of accuracy, that is wrong. There are no 2G VS on Areca. If you see someone with a 2G VS and running an Areca, they have switched storage for the VS subtest to use RST/RaidXpert with onboard raid storage.

 

Of my top 5 PCM05 scores, my best Virus scan on Areca 1882IX with 4GB ram cache with 3-6 OCZ Vertex 3 MaxIOPS was 1171MB/s. My good scores vary between 900-1171MB/s Virus scan on this setup. (I went SSD, because I believed enough of the newer SSDs could compete against stevero's acard ram drives at lower cost - I was close, but I was wrong)

 

Out of Stevero's top 5 PCM05 scores, his best Virus scan on Areca 1880IX with ? ram cache and a gaggle of Acard ramdrives was 1243MB/s. He has 3 other scores between 1150-1243 in his top 5. One score in his top 5 has virus scan of 1812MB/s, but its safe to say that isn't normal for Areca, and he may have been using RST with onboard RAID for Virus Scan that run... Or the stars aligned and he got an exceptional result from the Areca storage. That may be the case, but SteveRo is as good as they come with extreme storage, and if it was done on the Areca, I believe it is the best Virus Scan score submitted on any Areca based storage on hwbot.

 

MikeCDM also runs Areca, maybe an 12XX version with 1GB onboard ram cache. Out of his top 5 scores, his best VS was 967MB/s.

 

Also worth noting that Areca Virus Scan performance is heavily mitigated by CPU capability during MTT3. Virus scan on single or dual core processors is much higher on Areca when running just Virus Scan, but during MTT3 other tasks suck away CPU time and lower the virus scan part of MTT3. The same effect is not observed for single/dual core using RST/RAIDXpert, the VS ran alone or ran during MTT3 scores about the same.

 

The above is just my awareness... I don't watch everyone's results, but these are some sample Areca results I had paid attention to. I might have missed others also running Areca. But if 2G virus scan were possible on Areca right now, I think Steve or I would have managed it. Areca's true power is in general usage, it is much harder to artificially inflate by RST/RAIDXpert.

Edited November 14, 2012 by I.M.O.G.

[Moose83]

Yes Steve scored 1,7G VS on areca, i only had 2G one time;)

[I.nfraR.ed]

I once scored 30000 in 3DMark03 with integrated HD 4290.

[SteveRo]

I think that 1812 run was the highest vs I ever had, I doubt I will ever be able to get close to that again.

 

edit - Probably done with Areca4GB/15xAcard but could have been onboard 4xC300

Edited November 14, 2012 by SteveRo

[Yamunsa]

and again 1732 vs scored with only 2 ssds with rst and 32k stripe taking wr 4 core.

[Moose83]

Yes but actually it is not forbitten to use RST;)

[I.M.O.G.]

*woosh*

 

The point he's been reiterating in his recent posts is that VS scores like that are bogus for onboard RAID SSDs, and the example WR score he provided is intended as a demonstration of why it should not be accepted and that should be clearly stated to move PCM05 forward.

[Gluvocio]

well guys... the issue, I think, is not to decide if it's legal or not a thing have been used for years.

and for my tests without rst I noticed Gen. usage is the subtest more affected by this utility.

Virus scan is high (about gbps ) even if there is not Write-back enabled or bios is made "bios mode" ( press Ctrl+I )

The way how the test works seem to allow this since when it born. I saw similar sequential read on crystalDiskMark or other software... maybe a little under 1000mb/s , but anyway they were high.

Guys, I've seen 150+ mb/s made by a single mechanical disk.

That subtest hits high score since the dawn of time.

With raid you began to see 1000+ scores and noticed this.

But I think it's the way this subtest works.

The real issue, dear mates, should be (onboard RAM) write-caching of this utility , that affect much Gen.usage subtest.

It's a kind of software-ram-caching.

Equaliti should be allow or disallow ram-caching on write.

I think if disallow, decision should be applied to new scores/new CPUs ...one guy that played 1 year+ ago with e6600 / 2500k / i7 920 , never thinked it could be disallowed.

I prefer it whould be allowed, to put under conditions to play nicely people w/out controller too...

But for the same reason there should be allowed write-caching softwares with single disk, for equality.

Outcome it's not my personal trouble'cause I've controller, but I'd like to contribute with my ideas on this bench I've played for years.

[Moose83]

Thats right RST has been used and accepted over years! When you have Raid Controller, you can use it for gen usage and RST for VS;)

Then all have the same Chance in VS. I dont see there an problem.

I only see problem in sector size change.

I think it must decide faster about RST, new submits use it again, and nearly 90% of all pcm05 submits use RST:)

[SteveRo]

I only see problem in sector size change.

 

Agree completely Mr. Moose, so you will delete your non-512 sector submits?

[sofos1990]

RST can't be forbidden cause as and other guys told above most of the results should be deleted.

 

As for the stripe size...I don't see a reason that we should block it. Everyone knows it and can change it. 32K is not the best on all the setups. To me it seems like a "light tweak" as for example the LOD on 3Ds. The overclocker should find the best size for his setup and bench with it.

[Moose83]

Not stripe size! Sector size is an other thing!

Stripe size is legal.

But we need decission with RST! Why forbit it? Its an part of System driver/Software;)

 

@ Steve, i allready said, i will deleted:)

I will do it in december when i rebench with high clocks.

[IanCutress]

Its an part of System driver/Software;)

 

There are lots of things we ban which are part of the driver/software. Mipmap for example. Please stop using this as an argument. It's done on a case-by-case basis.

 

I still reckon that having 1300+ MB/s for VS with only two SSDs is an anomalous result, when total sequential of two in RAID-0 is only 1000MB/s. The write back cache only affects the writing to the memory, though I fear that is what PCMark actually detects rather than the full write through. Would be interesting to see how much it actually effects by profiling the functions

[Moose83]

Rst was used many years! Want to delete over 90% of all results? Really?

Everyone knows it so i dont see there an Problem;)

And i allways use 4 SSDs with rst score between 1500 and 2000...

Also abnormal for you?

[SteveRo]

@ Steve, i allready said, i will deleted:)

I will do it in december when i rebench with high clocks.

 

Mr. Moose,

 

The rules say it is illegal when -

 

"altering the perceived speed of the benchmark program, tricking it to believe it ran faster ".

 

If you agree that changing sector size is "altering the perceived speed of the benchmark program, tricking it to believe it ran faster " - than you should delete those scores now, it shouldn't have anything to do with "... December when I rebench with high clocks."

 

_____________________

[Moose83]

Ok Steve i delete them on weekend;)

[pro]

Ok final decisions are made and they are no different than we have been saying all along

 

I will write a small press release in the next few days and release shortly after that

 

Do us the favor and remove all your sector size modified scores now

[Toolius]

is software ram disk allowed guys ? (Radeon Ramdisk)

 

The tweaks i used are the following..

 

SetLOD for 7970

Mouse movement

 

Are these still legal ?

Edited November 16, 2012 by Toolius

[SteveRo]

is software ram disk allowed guys ?

 

The tweaks i used are the following.. .

SetLOD for 7970

Mouse movement

 

Are these still legal ?

 

You might want to wait for Master Pro's press release but I will venture a guess (my guess - not official!) -

 

1. Ramdisk is a no-no (as it always has been).

2. Irst or amd raid expert are fine provided sector size remains at 512 bytes.

3. SetLOD ok

4. Mouse movement ok.

Edited November 16, 2012 by SteveRo

[Toolius]

Thank you stevero

Also just another question.. is the WMV Registry tweaking (your encoding score is 3~4 times higher), PowerToy tweak legal ?

[Yamunsa]

just to be clear regarding RST - yes it has been available to install and use since 2003 known formally as Intel Matrix RAID, MSM and from 2010 RST. as today, like it's previous versions it's intended purpose is firmware level raid featuring the option to select different areas (e.g. partitions or logical volumes) on the same disk and assign them to different raid devices (e.g. ICH device).

 

in 2011, intel brought out ver.10.5 of RST with a new component called SRT. quoting the wiki:

 

Smart Response Technology (SRT) (pre-launch name SSD Caching) is a proprietary caching mechanism introduced in 2011 by Intel for their Z68 chipset (for the Sandy Bridge–series processors), which allows a SATA solid-state drive (SSD) to function as cache for a (conventional, magnetic) hard disk drive.

 

SRT is managed by Intel Rapid Storage Technology software version 10.5 or later, and implemented in its device driver and the Z68 motherboard's firmware (option ROM). It is available only when the (integrated) disk controller is configured in RAID mode (but not AHCI or IDE modes) by implementing a style of RAID-0 striping. Write-back (Maximized mode) or write-through (Enhanced mode) caching strategy can be selected by the user. The maximum utilizable cache size on the SSD is 64 GB. Caching is done at the logical block addressing (LBA) level, not the file level.

 

link

 

first time around, SRT doesn't really do all that much. It's only on the second and subsequent instances when frequently accessed data is called that SRT kicks in and Windows reads directly from SSD cache, bypassing the hard disk entirely. When that triggers, the speed boost over identical data called from the hard drive can be enormous.

 

link

 

effectively pc05 submissions made before may 2011 are completely free of scrutiny.

 

noting sentiment that RST (SRT) should be allowed so that competitors who do not own a 3rd party controller can compete with those that do would make pc05 the only benchmark with this degree of leniency. rules are already in place that don't allow mip to be used on a 560ti so it can compete evenly with a 680. I would go as far as saying that apart from this discussion hardware owns on hardware bot.

 

with SRT still being in it's infancy lord knows how it will develop and whether hwbot will need to revisit this same issue in the future, (let alone the development of ms readyboost).

 

understanding a press release will be put together in the next few days i look forward to clarification on the official read on how hwbot see tricking benchmarks into giving a score that’s not real fits with software caching in pcm moving forward.