The official HWBOT Team Cup 2012 - SC6: "PCMark05 xCore" thread.

[Leeghoofd]

Question 1 for the experts: If it's RAM cache, shouldn't it boost other apps too like ASSD, CD, HD Tune ? When testing SSD cache for a review I always get an improvement after the 2nd-3rd run... does RAM cache work differently ? (not taking into account reboots ofcourse)

 

I borrowed Gamers setup : I got a boost in PCMark05 with my Sempron 145 from 18K to 22.5K by applying the appropriate strip sizes mentioned by VAPOR and utilising the AMD Expert tool for the RAID setup (this with 400Mhz lower CPU clocks... ) If I set 64K iso 4K format, my scores for Gen Usage and Virus scan get lower here...

 

I saw a virusscan of 3000 ( single test run ) with the Sempron at 4Ghz, but sometimes it was far lower...

 

Question 2: Does this only work for version 2.18 and maybe future releases ? Older versions don't give the gen usage and virusscan boost ?

Edited September 3, 2012 by Leeghoofd

[I.M.O.G.]

The RAID expert Software use ram for caching. This is the same like areca cards do.

 

That is correct. However you can use RAID expert and get "normal" onboard ram caching scores in general usage and virus scan if you don't modify cluster/sector size.

 

With Areca, the RAID controller has RAM onboard, and it will produce high gen usage and virus scan scores consistently regardless of stripe size, sector size or formatting... It produces those big scores because running the storage tests multiple times it puts a lot of the storage tests in its RAM cache. (you must pre-run storage tests to initialize/preload the ram cache on an areca... Without preloading the ram cache on Areca, scores are only "good" not "great")

 

The important difference, in my opinion, is that PCM05 whigs out with RAID expert and changing the cluster/sector size. Yes, Raid expert supports ram caching... But running the storage tests a second time does not increase the scores the way it does for Areca. It doesn't need to be preloaded for ram caching to boost the score. The cluster/sector size changes just produces really high scores every time, because PCM05 is detecting/reporting/calculating the speeds wrong.

 

Does the firmware/driver trick the benchmark or is it just boosting the one thing the benchmark is designed to test for?

 

Certain SSD firmwares perform much better in virus scan and gen usage. The firmware's aren't tweaked for PCM05, just certain drivers with certain optimized firmware happen to work much better for PCM05.

Edited September 3, 2012 by I.M.O.G.

[Moose83]

We need a clear answer, if allowed or not. Christian Ney says its allowed, but i take care myself and dont abload scores with this maybee Bug.

[Christian Ney]

We need a clear answer, if allowed or not. Christian Ney says its allowed, but i take care myself and dont abload scores with this maybee Bug.

Said this if it is actually a driver feature that works, if this is just abug under PCMark05 and has no effects at all under other storage benchs then we can't allow it like bugged drivers.

 

Massman will make an official statement on it anyway.

[Moose83]

I have tested it, only pcm05 get high storage, hdtune makes bluescreen or all sectors are bad.

I cant meassure speed with hdtune so i think its a bugged driver.

Actually im for banning all scores made with this driver because bugged.

[Christian Ney]

What about ATTO and CrystalDiskMark ?

[Moose83]

I must test, but with 12.6 driver it isnt possible.

Only if you change Cluster size to 4 or 2 it works. When it is Feature, why works not with other Cluster sizes? For me its bugged.

[Moose83]

Here are some pics, look at my Volume Size, 219GB.

Then look at HDTune, only 60GB... And read Test gives an Error, Error scan during 1s, all Sectors are damaged, haha

 

http://www.abload.de/image.php?img=1kvpbc.jpg

 

http://www.abload.de/image.php?img=216pmo.jpg

 

http://www.abload.de/image.php?img=3qmp0u.jpg

 

So the drivers are fully buged. And its no really Cache, that the driver uses, didnt need to preload Cache like areca, the Power is always the same.

Normally cache needs to preload, first run storage tests, than go for score.

[knopflerbruce]

It does indeed sound bugged, yes.

[Moose83]

And here is ATTO, with 4SSD Raid 0, this shows the true Power Sequencially:) I cant see the hidden PCM05 Power with 2,8G, haha:)

 

http://www.abload.de/image.php?img=4jprkm.jpg

 

Same in CrystalMark...

 

http://www.abload.de/image.php?img=5ssorz.jpg

Edited September 3, 2012 by Moose83

[Leeghoofd]

Moose you have a link for 12.6 RAID drivers plz ?

[Moose83]

Must be on official AMD site:)

[Leeghoofd]

Can only find 12.8 there, no legacy drivers... wow AMD expert beats the socks off Win7 RAID

 

Nevermind found them

Edited September 3, 2012 by Leeghoofd

[Moose83]

So what is final decission with bugged scores done with this driver and huge Virus scan?

[VapoR.]

not only huge VS , General usage is almost 2x bigger too.

[Massman]

After some testing, it seems the driver is not really bugged, but some of the caching options just boost the PCMark subtest score. You can try this out yourself. The 12.6 drivers only allows 512bytes sectorisation whereas 12.8 supports 1K, 2K and 4K as well. The results with RAID set at 512bytes is the same on 12.6 and 12.8; it's only when you go to 1K, 2K or 4K that you have a performance boost.

 

So, the scores of Vapor are valid.

 

It does give some food of thought for the PCMark05 benchmark in genera, though.

[Leeghoofd]

AMD PCmark05 enters a new dimension, cooooooooooool...

[Moose83]

So, then i will use areca for gen usage and AMD for Virus scan...

[I.M.O.G.]

After some testing, it seems the driver is not really bugged, but some of the caching options just boost the PCMark subtest score. You can try this out yourself. The 12.6 drivers only allows 512bytes sectorisation whereas 12.8 supports 1K, 2K and 4K as well. The results with RAID set at 512bytes is the same on 12.6 and 12.8; it's only when you go to 1K, 2K or 4K that you have a performance boost.

 

So, the scores of Vapor are valid.

 

It does give some food of thought for the PCMark05 benchmark in genera, though.

 

So it isn't a bug, it's just a performance boost that when applied, increases the score such that onboard raid outperforms $2500 hardware raid with 4GB dedicated ram cache. Understood.

 

Try similar tests on Areca or IRST, and you cannot duplicate it. Only AMD, only with raidxpert. And only with disk benchmarks that don't error out when trying to test after making the changes.

 

Doesn't make sense, but nothing with PCM05 does, so business as usual.