How long should i run prime95 stress test? some people say 24 hrs some say 12 hours and others say if there are going to be problems it will happen with in the first hour. So that's my question how long should I stress?
MONITOR: 27" Samsung SyncMasterP2770 1920x1080 1ms CASE: Corsair Carbide 400r PSU: Corsair AX850 MB: Asus P8P67 Deluxe CPU: i5 2500k @ 4.5Ghz COOLER: Corsair H100 RAM: Corsair Vengeance 8gb 1600 (blue)
GPU: 2x EVGA GTX 560 Ti 448 Classified SSD: Corsair Force GT 120gb (OS) HDD: Western Digital VelociRaptor 600gb (steam files and games) HDD: Western Digital Black 2TB (Media files) HeatWare ID: http://www.heatware.com/eval.php?id=79923 Origin ID: kanechil
|
I usually run it for an hour or two when i'm getting close to my "Cap". When i've found something thats run for at least an hour stable. I'll run it all night and while i'm at work, Which is around 15 or 17 hours.
Zalman Z9 Mid Tower - Corsair HX750 PSU - MSI 970A-G46 - PII X4 955 C3@ 4.0ghz - 8GB Corsair Vengeance Blacks - MSI GTX 560 Ti Twin Frozr II / PhysX: EVGA GTS 450 @ 955 / 2004
|
nick1551
CLASSIFIED Member
- Total Posts : 2647
- Reward points : 0
- Joined: 2006/07/14 16:04:02
- Status: offline
- Ribbons : 24
20 passes of linx 0.6.4 @ max ram with the latest binaries, will test stability better for i5/i7 cpu's.
Heatware Evga 8800GT 512mb Fan Speed BIOS   i7-920 @4.0 - Evga X58 Micro - 2x4gb Gskill - Evga GTX 760 ACX @1228mhzWD 1tb Blue - Corsair HX620 - CM690 Nexus 4 -- Beanstalk 4.4.2 - LTE+ART enabled
Moto G -- Stock 4.4.2
|
nick1551
20 passes of linx 0.6.4 @ max ram with the latest binaries, will test stability better for i5/i7 cpu's.
1+ I usually do the "High" test I think, which is 4GB of memory. But yeah either are very CPU intensive, and if it passes, it sure shouldn't fail while folding, gaming, etc.
|
ehau
SSC Member
- Total Posts : 717
- Reward points : 0
- Joined: 2004/07/06 06:01:30
- Status: offline
- Ribbons : 0
Depends on what you are using your computer for. For gaming stable, 30 minutes. And if you ever get a BSOD with a secondary processor interrupt not received error in intensive games like BF3, bump the vcore up a notch.
|
owcraftsman
CLASSIFIED Member
- Total Posts : 4418
- Reward points : 0
- Joined: 2005/04/16 16:41:02
- Location: Cape Coral, Florida
- Status: offline
- Ribbons : 57
I guess I'm old school but to consider yourself completely stable one is not good without the other. Prime is a real world everyday usage stress test vs IBT or LinX which is overkill. True LinX is a good indication of stability but you will never see temps that it produces in a real world scenario. Look no further than how high your temps get after hours of intense gaming. However unless you use the 27.4 beta of P95 you will not be using the AVX instructions native to the SB proc that the current ver LinX ver. 0.6.4 (10.3.7.012) uses. I have seen many run 20 passes of LinX only to find they can not pass all iteration available in P95. Currently with a SB there are 72 separate iterations available in P95 ver 26.6 and 82 with P95 ver 27.4 which includes the AVX instructions. By default both versions use 1600 MB of RAM which is not satisfactory when our systems have 8-16GB available. It's best to run a custom blend using 90-94% of your memory. For example I have 16GB of memory therefore plugin 13500MB to use while testing which puts me in the 93% range after what windows overhead uses. With proper memory utilization set for torture testing there is no better stress test than P95 for real world stability. Because P95 ver 27.4 is a beta you may want to avoid that, I don't, which makes using LinX with AVX instruction a good companion and no more IMO. Understand there is more to stress test that none of these programs stress. For example the PCI bus and the GPU for that matter remain largely untested with prime or LinX so putting your faith in either for complete stability is not advisable and there is still more that needs to be done not the least of which is real world usage. Bottom line if complete stability is your goal then 20 passes of LinX is not enough but be prepared for many hours of testing, benching etc to insure it. Most peeps just don't have the time or patients to get there and complain the loudest when things go wrong. Lost and corrupted files, constant annoying bugs and formatting & reloading windows is not my idea of fun and complete stability helps eliminate a failed overclock as the culprit of said fun. What I recommend for complete stability: - P95 ver 26.6 20hrs min or ver. 27.4 for 23hrs min.
- Linx 0.6.4 (10.3.7.012) 10 passes when not using P95 ver 27.4
- Furmark min. 2 hrs
- 3DM11 bench and compare to similar systems
- 3DMV bench and compare to similar systems
- Heaven bench and compare to similar systems
- PCM7 bench and compare to similar systems
- SuperPI 32m bench and compare to similar systems
- HCI-Design Memtest 200% coverage max mem two instances.
- CineBench 11.5 bench and compare to similar systems
If you pass all these at stock and overclocked speeds and your scores compare to similar systems, barring a hardware failure which could happen to any of us at any given time, you will be in excellent shape and worry free.
Phanteks Eclipse P400: | MSI Z170A G7 | 6700k 4.7x 1.345v H110iGTX | G.Skill F4-3000C15D-16GVR |2x EVGA 980 TI SC ACX in SLI | Samsung 950 Pro M.2 OS | WD2003FZEX Data | Corsair RM750i
Phanteks Enthoo Luxe: MSI Z87-G45 Gaming v1.9 bios | 4770k 43x1.24v XSPC Raystorm | G.Skill TridentX2133C9 2x8-16GB | MSI R9-290x Kyrographics Hawaii w/Active Backplate | 2x 250gb 840 EVO Raid-0 OS | WD20003FZEX (data) | Corsair HX850
Phanteks Enthoo Primo: | MSI Z97 Gaming 7 | 4790k 48x 1.254v EK-EVO | G.Skill TridentX2400C9 2x8-16GB | 2x Sapphire R9-290x ref (hynix) CF EK block Backplate & serial bridge | 2x 840 Pro Raid-0 OS | WD20003FZEX (data) | EVGA SuperNova 1000 G2
|
owcraftsman
I guess I'm old school but to consider yourself completely stable one is not good without the other. Prime is a real world everyday usage stress test vs IBT or LinX which is overkill. True LinX is a good indication of stability but you will never see temps that it produces in a real world scenario. Look no further than how high your temps get after hours of intense gaming. However unless you use the 27.4 beta of P95 you will not be using the AVX instructions native to the SB proc that the current ver LinX ver. 0.6.4 (10.3.7.012) uses. I have seen many run 20 passes of LinX only to find they can not pass all iteration available in P95. Currently with a SB there are 72 separate iterations available in P95 ver 26.6 and 82 with P95 ver 27.4 which includes the AVX instructions. By default both versions use 1600 MB of RAM which is not satisfactory when our systems have 8-16GB available. It's best to run a custom blend using 90-94% of your memory. For example I have 16GB of memory therefore plugin 13500MB to use while testing which puts me in the 93% range after what windows overhead uses. With proper memory utilization set for torture testing there is no better stress test than P95 for real world stability. Because P95 ver 27.4 is a beta you may want to avoid that, I don't, which makes using LinX with AVX instruction a good companion and no more IMO. Understand there is more to stress test that none of these programs stress. For example the PCI bus and the GPU for that matter remain largely untested with prime or LinX so putting your faith in either for complete stability is not advisable and there is still more that needs to be done not the least of which is real world usage. Bottom line if complete stability is your goal then 20 passes of LinX is not enough but be prepared for many hours of testing, benching etc to insure it. Most peeps just don't have the time or patients to get there and complain the loudest when things go wrong. Lost and corrupted files, constant annoying bugs and formatting & reloading windows is not my idea of fun and complete stability helps eliminate a failed overclock as the culprit of said fun.
What I recommend for complete stability:
- P95 ver 26.6 20hrs min or ver. 27.4 for 23hrs min.
- Linx 0.6.4 (10.3.7.012) 10 passes when not using P95 ver 27.4
- Furmark min. 2 hrs
- 3DM11 bench and compare to similar systems
- 3DMV bench and compare to similar systems
- Heaven bench and compare to similar systems
- PCM7 bench and compare to similar systems
- SuperPI 32m bench and compare to similar systems
- HCI-Design Memtest 200% coverage max mem two instances.
- CineBench 11.5 bench and compare to similar systems
If you pass all these at stock and overclocked speeds and your scores compare to similar systems, barring a hardware failure which could happen to any of us at any given time, you will be in excellent shape and worry free.
Wow Thanks for the detailed response, i had no idea all of this would be involved, even though some may call it overkill, i like to have assurance. Even though my memory and gpu are running at stock do you think i should still test them? For instance running p95 and cinebench at the same time?
MONITOR: 27" Samsung SyncMasterP2770 1920x1080 1ms CASE: Corsair Carbide 400r PSU: Corsair AX850 MB: Asus P8P67 Deluxe CPU: i5 2500k @ 4.5Ghz COOLER: Corsair H100 RAM: Corsair Vengeance 8gb 1600 (blue)
GPU: 2x EVGA GTX 560 Ti 448 Classified SSD: Corsair Force GT 120gb (OS) HDD: Western Digital VelociRaptor 600gb (steam files and games) HDD: Western Digital Black 2TB (Media files) HeatWare ID: http://www.heatware.com/eval.php?id=79923 Origin ID: kanechil
|
I run ptime95 for 30 min on each mode
LinX for 20 min means same thing but i just find LinX too unrealistic
I've had an temporary 2500K running @ 5.3Ghz stable for 3 days i ran LinX and after 10 seconds the machine locked up
|
hellshealer
SSC Member
- Total Posts : 906
- Reward points : 0
- Joined: 2009/10/20 20:48:21
- Status: offline
- Ribbons : 1
warlord420
I run ptime95 for 30 min on each mode
LinX for 20 min means same thing but i just find LinX too unrealistic
I've had an temporary 2500K running @ 5.3Ghz stable for 3 days i ran LinX and after 10 seconds the machine locked up
that means it wasnt stable like my 980x @4.3 i can run LinX all day everyday at dcent temps but fail after 6 hours of prime. or 12-13 hours of folding. BEST WAY to test stability is to run SMP clients from F@H. they will absolutly test your system
mah rig: HAF X i7 980X @4.3 (143x30) H70 eVGA X58 4-way SLI Classified 6gb kingston hyperX 1436mhz@ 8-8-8-88-1T Vraptor 150gb OS installed win 7 WD black 1tb storage eVGA GTX470 2 way SLI evga GT430 (second monitor) Antec 1200watt true power
|
owcraftsman
CLASSIFIED Member
- Total Posts : 4418
- Reward points : 0
- Joined: 2005/04/16 16:41:02
- Location: Cape Coral, Florida
- Status: offline
- Ribbons : 57
Kanechil
owcraftsman
If you pass all these at stock and overclocked speeds and your scores compare to similar systems, barring a hardware failure which could happen to any of us at any given time, you will be in excellent shape and worry free.
Wow Thanks for the detailed response, i had no idea all of this would be involved, even though some may call it overkill, i like to have assurance. Even though my memory and gpu are running at stock do you think i should still test them? For instance running p95 and cinebench at the same time?
I usually run my new builds through all these test right out of the box. The benefits are multiple. If a piece of hardware is a lemon I want to know and you find out quickly while returning is easy. Most mfg only guarantee stock speeds so this verifies it's capable of a full load at stock clocks. Generally speaking if your hardware passes these test it's not likely to drop dead on you either which is some thing else I'd rather know sooner rather than later. Typically running these test at stock you revel windows installation & config related problems and hardware driver & config issue which is invaluable when diagnosing potential causes of a failed overclock since you have eliminated that as a potential cause. About running multiple stress test at the same time. I'm not a fan. It's hard enough to interpret a failure when one app is running let alone two. Then you have to be careful running two apps that are CPU intensive like Cinebench and P95 would be. I've heard of folks running P95 and Furmark at the same time which may not be to bad but still I would avoid it.
Phanteks Eclipse P400: | MSI Z170A G7 | 6700k 4.7x 1.345v H110iGTX | G.Skill F4-3000C15D-16GVR |2x EVGA 980 TI SC ACX in SLI | Samsung 950 Pro M.2 OS | WD2003FZEX Data | Corsair RM750i
Phanteks Enthoo Luxe: MSI Z87-G45 Gaming v1.9 bios | 4770k 43x1.24v XSPC Raystorm | G.Skill TridentX2133C9 2x8-16GB | MSI R9-290x Kyrographics Hawaii w/Active Backplate | 2x 250gb 840 EVO Raid-0 OS | WD20003FZEX (data) | Corsair HX850
Phanteks Enthoo Primo: | MSI Z97 Gaming 7 | 4790k 48x 1.254v EK-EVO | G.Skill TridentX2400C9 2x8-16GB | 2x Sapphire R9-290x ref (hynix) CF EK block Backplate & serial bridge | 2x 840 Pro Raid-0 OS | WD20003FZEX (data) | EVGA SuperNova 1000 G2
|
Prime on blend or w/e takes 12 hours to complete a full cycle of tests. So i would run for atleast 12 hours. add in a 20 pass of Linx at 25000 problem size. and throw some benches at it and normal everything things. Surprisingly enough ive found BF3 will actually weed out a low Vcore or any other instabilities. but thats just my 2 cents :)
|
owcraftsman
CLASSIFIED Member
- Total Posts : 4418
- Reward points : 0
- Joined: 2005/04/16 16:41:02
- Location: Cape Coral, Florida
- Status: offline
- Ribbons : 57
With all due respect and FYI @ Crushed I know you are a tweak hound and thought you may find this useful. For the i3 i5 & i7 processors P95 ver 26.6 64 bit has 70 separate test. Default blend between 8 - 4096k at 15 min per iteration it would take a min. of 17.5 hrs to complete a full circuit under perfect conditions. I have found that for what ever reason not all test take exactly 15 minutes but take on average 16.65 minutes per iteration which means it could easily take 19.5 hrs to complete a full circuit under normal conditions. 16.65*70=1165.5/60=19.425hrs You can check this quickly your self by doing a custom blend 1 core 1min per iteration and it will take up to 90 minutes to start repeating iterations over again. Conversely using 27.4 beta with AVX instruction has the same default 8 -4096k test but adds 15 tests and deletes 3 for a total of 82 total iterations to complete a full circuit. Doing the math it could take up to 22.75 hrs to run all test available. 16.65*82=1365.3/60=22.755hrs I would not be satisfied until I completed a full circuit with either version although I prefer using the AVX instructions and have found the additional overhead requires a bit more Vcore and or finely tuned CPU PLL V and VCCIO adjustments to complete w/o errors. I use 1344k, 1792 & 2688k custom 5 minute test to set those three voltages before attempting to run a complete circuit. Lower than default 1.8v CPU PLL between 1.5-1.7 seems to allow for a lower Vcore requirement and a higher than default VCCIO seems to help when using more than 8GB of memory and or all four dimm slots populated, or higher than 1600 clocks, or tighter timings and 1t vs 2t command rate. I'll use benchmarks like 3DM11 to fine tune my PLL looking for what setting gives me the best high avg score. I use Adia64 Cache & Memory bench to fine tune VCCIO and Vdimm. Once voltages are tuned I will attempt to run a full P95 custom blend test using 90-94% of memory for at least one full circuit.
Phanteks Eclipse P400: | MSI Z170A G7 | 6700k 4.7x 1.345v H110iGTX | G.Skill F4-3000C15D-16GVR |2x EVGA 980 TI SC ACX in SLI | Samsung 950 Pro M.2 OS | WD2003FZEX Data | Corsair RM750i
Phanteks Enthoo Luxe: MSI Z87-G45 Gaming v1.9 bios | 4770k 43x1.24v XSPC Raystorm | G.Skill TridentX2133C9 2x8-16GB | MSI R9-290x Kyrographics Hawaii w/Active Backplate | 2x 250gb 840 EVO Raid-0 OS | WD20003FZEX (data) | Corsair HX850
Phanteks Enthoo Primo: | MSI Z97 Gaming 7 | 4790k 48x 1.254v EK-EVO | G.Skill TridentX2400C9 2x8-16GB | 2x Sapphire R9-290x ref (hynix) CF EK block Backplate & serial bridge | 2x 840 Pro Raid-0 OS | WD20003FZEX (data) | EVGA SuperNova 1000 G2
|
