Gautam Posted January 26, 2009 Posted January 26, 2009 IF you guys are going to ban MFT Tweak to assist SSds in write / reads , then you have to ban the use of i-Rams as it uses memory for this benchmark and not normal HDD platters. No, they are totally different situations. i-RAMS use DIMMS as storage. Just the sticks as their storage medium. Not system memory. MFT uses a software buffer in system memory. This is why it's scores so closely match a software ramdrive. This is what people are raising issue with. In essence it's a software layer in between which is written to. Not the SSD array itself. The usage of software is the issue, not the storage medium.
chispy Posted January 26, 2009 Posted January 26, 2009 No, they are totally different situations. i-RAMS use DIMMS as storage. Just the sticks as their storage medium. Not system memory. MFT uses a software buffer in system memory. This is why it's scores so closely match a software ramdrive. This is what people are raising issue with. In essence it's a software layer in between which is written to. Not the SSD array itself. The usage of software is the issue, not the storage medium. In escence i-Rams use DIMMS (memory meant to be used on motherboards and not hardrives ) HDD platters or normal HDD usage right ? Ok.
TaiwanAndre Posted January 26, 2009 Posted January 26, 2009 (edited) Sorry about hurting everyone's eye. This score is too bad. I know I can do 40K+. If hwbot still accept MFT. I promise 40K+ is coming soon. http://www.iamxtreme.net/andre/34000.JPG http://www.hwbot.org/result.do?resultId=815628 Any questions? Remove MFT or Not? Edited January 26, 2009 by TaiwanAndre
knopflerbruce Posted January 27, 2009 Posted January 27, 2009 platter-based HDDs only? THat's what I was thinking;)
mikeguava Posted January 27, 2009 Posted January 27, 2009 THat's what I was thinking;) MFT can be used there too
eva2000 Posted January 27, 2009 Author Posted January 27, 2009 Geez guys getting their knickers in a knot MFT reserves part of the usable target drive's (SSD or regular HDD) free space for it's work (~10% of SSD drive space). You should just allow software ram based benchies too and solve the problem. Software ram disks are accessible by everyone if they want it (sometimes at a price) which is no different to how liquid helium, dry ice and LN2 are available for folks who are willing to fork out dollars for pots, dewars and related expenses. You don't see folks complaining they don't have access due to price/costs to liquid helium, dry ice and LN2 and calling a ban to such usage - well okay there are some folks calling for separate categories to cooling so they can compete. Costs of ramdisks/ssd can't be a valid argument for not allowing them, as liquid helium, dry ice and LN2 costs for pots, dewars and related expenses can also be lumped into the argument for disallowing sub zero cooling. Honestly, have no idea why folks and futuremark are so strongly against allowing memory system be a factor in a benchmark that benchmarks the entire system and should reflect real world usage. Real world facts SSDs are in use for both os and non-os disk purposes Hardware raid controllers use memory caches to improve raid/read/write performance Hard drives have a memory cache to do the same Utilising system memory based caches exist to improve I/O performance is widespread in both consumer and corporate worlds. Operating systems utilise memory system for caching. Databases have query caching (MySQL) to reduce the load on database servers. There's also Memcached to basically move databases into ramdisk storage rather than on hard drive. PHP which is widespread and utilised by allot of applications can use PHP caches like xcache, Eaccelerator and APC to cache PHP to reduce cpu loads. [*]Benchers tune their system cache/memory performance to improve their pi and even 3dmark results Pcmark05/benching pcmark05 measures overall system performance and memory system is part of that so allowing such will open up competition to see who can tune their memory system and bandwidth the best - similar to super pi, everest bandwidth that folks like to see. It still takes into account the cpu and gpu so folks tweaking that + the drive system + mem system will still come out on top eventually we have folks running e-ram/ramdisk for when they bench super pi 32m we have folks clocking and tuning the memory system to get more memory bandwidth and faster memory latency to get better pi and even squeeze out a few more points in 3dmark Folks complaining, are you so stuck in the past that you can't see the future ? In the future, memory system tuning and benching will become even more important especially with both Intel and AMD parties moving to integrated memory controller. Every benchmark run will utilise the memory system and to differing extent improvements and gains in benchmark results will occur by tuning and tweaking that memory system - be it clocking memory higher, tightening main and subtimings etc. The future is here now!
dinos22 Posted January 27, 2009 Posted January 27, 2009 Sorry about hurting everyone's eye. This score is too bad. I know I can do 40K+. If hwbot still accept MFT. I promise 40K+ is coming soon. http://www.iamxtreme.net/andre/34000.JPG http://www.hwbot.org/result.do?resultId=815628 Any questions? Remove MFT or Not? how did you get a valid link with such high XP startup
Gautam Posted January 27, 2009 Posted January 27, 2009 You're behind the times dude lol. The cap got removed a couple of weeks ago, if it didn't we wouldn't be having this discussion.
dinos22 Posted January 27, 2009 Posted January 27, 2009 oh shit didnt know that it seems RAMdrives are still not allowed right bastards hehehe
mikeguava Posted January 27, 2009 Posted January 27, 2009 why did my result not get accepted? I worked sooo long for it!
chispy Posted January 27, 2009 Posted January 27, 2009 Futuremark removed the 220cap for xp start up for a reason Guys , Let the fun begin All or nothing at this point
SF3D Posted January 27, 2009 Posted January 27, 2009 So everybody still think, that scores around 35 000 are still telling real truth about system performance This test is so badly biased to HDD tests, that there is no idea to use it anymore. I know it must feel good to see such a scores on screen, but come on. In general 24/7 use storage is not the most important part.. usually it is not. How many of us move or use really large (+10GB) files all the time? OS's need a lot of ram and that is making usage more fluent than any super speed HDD or SSD. I will get some SSD to my internet/gaming machine too, but still, I don't like PCmark05 anymore. I think at least MFT usage will have to be stopped /banned.
SF3D Posted January 27, 2009 Posted January 27, 2009 Yes jmke. That is real life use. Not some synthetic benchmark, which will give totally insane results with MFT.
chispy Posted January 27, 2009 Posted January 27, 2009 Geez guys getting their knickers in a knot MFT reserves part of the usable target drive's (SSD or regular HDD) free space for it's work (~10% of SSD drive space). You should just allow software ram based benchies too and solve the problem. Software ram disks are accessible by everyone if they want it (sometimes at a price) which is no different to how liquid helium, dry ice and LN2 are available for folks who are willing to fork out dollars for pots, dewars and related expenses. You don't see folks complaining they don't have access due to price/costs to liquid helium, dry ice and LN2 and calling a ban to such usage - well okay there are some folks calling for separate categories to cooling so they can compete. Costs of ramdisks/ssd can't be a valid argument for not allowing them, as liquid helium, dry ice and LN2 costs for pots, dewars and related expenses can also be lumped into the argument for disallowing sub zero cooling. Honestly, have no idea why folks and futuremark are so strongly against allowing memory system be a factor in a benchmark that benchmarks the entire system and should reflect real world usage. Real world facts SSDs are in use for both os and non-os disk purposes Hardware raid controllers use memory caches to improve raid/read/write performance Hard drives have a memory cache to do the same Utilising system memory based caches exist to improve I/O performance is widespread in both consumer and corporate worlds. Operating systems utilise memory system for caching. Databases have query caching (MySQL) to reduce the load on database servers. There's also Memcached to basically move databases into ramdisk storage rather than on hard drive. PHP which is widespread and utilised by allot of applications can use PHP caches like xcache, Eaccelerator and APC to cache PHP to reduce cpu loads. [*]Benchers tune their system cache/memory performance to improve their pi and even 3dmark results Pcmark05/benching pcmark05 measures overall system performance and memory system is part of that so allowing such will open up competition to see who can tune their memory system and bandwidth the best - similar to super pi, everest bandwidth that folks like to see. It still takes into account the cpu and gpu so folks tweaking that + the drive system + mem system will still come out on top eventually we have folks running e-ram/ramdisk for when they bench super pi 32m we have folks clocking and tuning the memory system to get more memory bandwidth and faster memory latency to get better pi and even squeeze out a few more points in 3dmark Folks complaining, are you so stuck in the past that you can't see the future ? In the future, memory system tuning and benching will become even more important especially with both Intel and AMD parties moving to integrated memory controller. Every benchmark run will utilise the memory system and to differing extent improvements and gains in benchmark results will occur by tuning and tweaking that memory system - be it clocking memory higher, tightening main and subtimings etc. The future is here now! Well said Eva , i could not explain it better myself. You are absolutly correct on your statement , hopefully people will understand better SSds technology now , i hope we dont get stuck back in time and hwbot move forward instead of backwards with this benchie bro. Yes the future is here now...
TaiwanAndre Posted January 27, 2009 Posted January 27, 2009 So everybody still think, that scores around 35 000 are still telling real truth about system performance This test is so badly biased to HDD tests, that there is no idea to use it anymore. I know it must feel good to see such a scores on screen, but come on. In general 24/7 use storage is not the most important part.. usually it is not. How many of us move or use really large (+10GB) files all the time? OS's need a lot of ram and that is making usage more fluent than any super speed HDD or SSD. I will get some SSD to my internet/gaming machine too, but still, I don't like PCmark05 anymore. I think at least MFT usage will have to be stopped /banned. I still see MFT score on hwbot. If hwbot agree MFT score, I would plan to test 40K+. I don't want to waste my LN2 for this stupid benchmark until hwbot agree MFT or not.
Massman Posted January 27, 2009 Posted January 27, 2009 I still see MFT score on hwbot. If hwbot agree MFT score, I would plan to test 40K+. I don't want to waste my LN2 for this stupid benchmark until hwbot agree MFT or not. We're still having the discussion in the crew forums, nothing has been decided yet.
Massman Posted January 27, 2009 Posted January 27, 2009 Well said Eva , i could not explain it better myself. You are absolutly correct on your statement , hopefully people will understand better SSds technology now , i hope we dont get stuck back in time and hwbot move forward instead of backwards with this benchie bro. Yes the future is here now... SSD is the future, there's no doubt about that. We're discussing the usage of MFT here: an application that partially uses the system memory to increase HDD performance.
TheKarmakazi Posted January 27, 2009 Posted January 27, 2009 Just wait until later this year when SSD's will easily be hitting 500MB/s with just one drive. Total SATAII saturation soon (without MFT etc)!! Which leads me to another point... If MFT actually reflected the performance of the drive accurately how can you possibly get 1000+MB/s transfer rates with current generation SSDs?? ICH10R craps out around 600MB/s so I think that shows it is at least partially unrealistic results. I tried some quick runs on ICH10R and got well above 600mb/s with MFT and an Iram. Im not "for" or "against" this MFT "tweak," just trying to be devils advocate. I will abide by whatever the community and hwbot staff decide is the way to go forward...
eva2000 Posted January 27, 2009 Author Posted January 27, 2009 (edited) SSD is the future, there's no doubt about that. We're discussing the usage of MFT here: an application that partially uses the system memory to increase HDD performance. and why is that a problem ? I'm not one of those that 'heavily opposed' software ramdisks, I've always been for them even before pcmark04/05 existed as they're used widely in the REAL world in server/linux environments to improve storage based I/O performance why i'm so pro SSD + MFT ? answer is easy I use it to improve my SSD performance for STORAGE related usage in the REAL world = video encoding scratch disk 4x32GB OCZ SSD Raid 0 + MFT on Highpoint 3520LF PCI-E controller with 256MB cache vs 750GB Samsung SATAII Configurations for System Clocks: 4009Mhz @19x211Bclk with 8x mem multiplier 3x1GB HCF0 @844Mhz 9-9-9-24 i7 920 with HT no Video cache (SATA HDD source drive) = 4min 50s i7 920 with HT with Video cache (SATA HDD source drive) = 4min 10s i7 920 without HT no Video cache (SATA HDD source drive) = 4min 34s i7 920 without HT with Video cache (SATA HDD source drive) = 3min 53s i7 920 with HT no Video cache (SDD source drive) = 4min 05s i7 920 with HT with Video cache (SDD source drive) = 4min 06s i7 920 without HT no Video cache (SDD source drive) = 4min 04s i7 920 without HT with Video cache (SDD source drive) = 3min 55s Note: Video and Audio caches use system's tri-channel DDR3 memory bandwidth which is faster than both SATAII and SSD Raid 0 transfer bandwidth and as I tested early if you disable HT I notice that your get more clock for clock memory bandwidth in Memtest86+ v2.11 than when you have HT enabled. Guess ultimately it comes down to definition of storage usage. The fact is with SSD and even software ramdisks (which were out there way before pcmark04/05), the definition of storage has changed. Now the question is whether FM/hwbot accepts this evolution or not. Anyway, it's up to you guys to decide whether MFT assisted SSD is allowed or not. It's your house so it's your rules either way. edit: Also if the argument is pcmark05 is to test hdd performance not system memory performance which software ramdisks can effect, then the argument can be the same for 3dmark2001/2003/05/06 which all are effected by system memory bandwidth to various extents. Classic is 3dmark2001 which can get pretty big boosts when memory bandwidth is tweaked. Accordingly, since 3dmark benchmarks are meant to test GPU performance and not system memory bandwidth, we should ban 3dmark2001/2003/05/06 as well. Edited January 27, 2009 by eva2000
Praz Posted January 27, 2009 Posted January 27, 2009 MFT is the best thing to happen to SSD drives for both server and desktop use. I don't think anyone is disputing that. When used with PcMark is the score an actual indication of system performance or is it some artificially inflated value? I get 6000 MB/sec in HD Tune on a single MFT enabled SSD. If PcMark is being affected in a like manner I don't think there is anyone that can justify it's use.
Massman Posted January 27, 2009 Posted January 27, 2009 MFT is the best thing to happen to SSD drives for both server and desktop use. I don't think anyone is disputing that. When used with PcMark is the score an actual indication of system performance or is it some artificially inflated value? I get 6000 MB/sec in HD Tune on a single MFT enabled SSD. If PcMark is being affected in a like manner I don't think there is anyone that can justify it's use. MFT disabled: 55MB/s MFT enabled: 1350MB/s
eva2000 Posted January 27, 2009 Author Posted January 27, 2009 what you comparing for 55MB/s? this is a quick run on non-MFT 4xSSD Raid 0
SF3D Posted January 27, 2009 Posted January 27, 2009 Massmans example is with 4x OCZ core SSD's. It is valid point. Increase in MB/s is huge with MFT. Your example is doable with I-rams/acard and that is nothing exceptional. With MTF you get 1600MB/s and that is crazy. I think Praz is only one so far who think the same way I do. SSD'S raid0 configuration + MFT is nothing from real world in this synthetic test. That is the point. Scores are artificially high in this particular test.
Recommended Posts