PDA

View Full Version : 5960X @ 4.8385 GHz - RealBench: 220,970



Gobe
03-15-2016, 08:03 AM
Went all out with my 5960X on X99-A USB3.1 for a RealBench score of 220,970 at 4838.5 MHz (47 x 102.9) which actually calculates out to 4836.3 MHz, but what's a couple MHz?)

The damn thing was jumping around like an out of balance washing machine on spin cycle, but somehow I got it to finish the benchmark without crapping the bed.

Image Editing: 269,548
Video Encoding: 332,224
OpenCL: 50,817
Heavy Multitasking: 271,946
Score: 220,970

I'll bet Arne's not going to stand for this! Xeons, we're coming for you!

Arne Saknussemm
03-15-2016, 09:02 AM
Nice! 5960X crown :D

Sure...I'll have another go one of these days...but there's worse in these waters than me LOL

Menthol
03-15-2016, 11:44 AM
Gobe,
Very good score, that is quite amazing passing RB at that speed with an AIO cooler and get your laundry clean at the same time

Gobe
03-15-2016, 03:04 PM
AIO yes, but with some help. Ice bucket... sub-zero mixture... dunked radiator... -10C. And voltage, lots of voltage. We're talking "you'd better have your finances in order" type of voltage.

I could complete the bench at 4.85 GHz, but only a small percentage of the time and not long enough to make the multiple runs necessary to get a run where all the benches hit the top end of the range at once. It's that 5 - 10K variation in the encoding and heavy multitasking scores that makes things interesting... and frustrating.

cekim
03-15-2016, 09:17 PM
Nice! and LOL on the voltage...

Gobe
03-16-2016, 01:55 PM
I'm trying to figure out why I've got such a large lead on the rest of the field in image editing. 270,000 is huge and I really can't see why I would be 25,000 points ahead of the field on similar hardware. I thought it might be the Samsung 950 Pro NVMe, but I've run from a SATA SSD and got a similar high image editing score. I looked at memory timings, but this also doesn't seem to be it. Could somebody who understands the factors that contribute to a high image editing score comment on it? If there's am image editing secret that I've stumbled upon, I'd really like to pass it along.

Not that I'm complaining or anything...

Chino
03-16-2016, 03:40 PM
Go enter the ROG REALBENCH - FIRST BETA CHALLENGE (http://oc-esports.io/#!/round/rog_realbench_beta) for a chance to win some prizes. With your score will put you in first place.

Gobe
03-19-2016, 07:10 AM
Got back from working on the west coast this evening and was sad to learn that my wife wasn't home when DHL tried to deliver the cherry-picked 5960X I bought last week. Probably won't see it until next week. Boohoo.

Got some consolation though and pushed the 5960X I do have a little harder and took down one of the Xeons and moved into 2nd place on the RealBench leaderboard at 223,096. I needed to boost my anemic OpenCL score (50,817) so borrowed another GTX-960 from another system and loaded up the Intel OpenCL support (thanks for that one Arne!) and got the OpenCL up to 106,894. Got the frequency up a bit from 4.84 to 4.86 GHz with a sub-zero ice bucket (got it to -5C) and scary voltage. Cache and RAM were screaming (4651 and 3382 MHz). Temperatures stayed below 80C even though I was north of 1.5 volts. Livin' dangerously!

My Image editing and video encoding were actually down from my previous submission, but OpenCL more than doubled and heavy multitasking was up a bit. Seems the extra punch on OpenCL was the magic bullet.

Does one get over this addiction?



Image Editing:
268212


Video Encoding:
327269



OpenCL:
106594



Heavy Multi-tasking:
275586


Score:
223096


System Spec



CPU:
Intel Core i7 5960X


CPU Frequency:
4860.7 MHz


Physical Cores:
8


Logical Cores:
16


Multi:
46.0


Bus:
105.7 MHz


Memory:
G.Skill


Memory Frequency:
1690.8 MHz (1:24)


Memory Channels:
Quad


Memory Capacity:
32 (GB)


Memory Latency:
13-13-13-28-1T



Motherboard/Laptop:
X99-PRO/USB 3.1



Firmware:
2101


GPUs:
NVIDIA GeForce GTX 960
NVIDIA GeForce GTX 960


OS Version:
Windows 7 Professional


Submission Date:
Sat Mar 19 2016 06:09:45

Arne Saknussemm
03-19-2016, 08:25 AM
Nice one Gobe! climbing the table! Good run!

You can slot in any GPUs I think....don't have to be the same or running SLI or anything...open CL will just run on anything so if you have an old 580 plug it in too for example


Livin' dangerously!

Livin' like a man who has another CPU in the post... ;)


Does one get over this addiction?

When there are no more CPUs in the post...

Gobe
03-19-2016, 09:32 AM
I'm dying to get my hands on that chip. It was binned by 8Pack over in the U.K. as being in the top 2%. It's billed as being bench stable at 4.75 Ghz and 24/7 stable at 4.65 GHz at probably around 1.30 - 1.33 volts on custom water.

My current 5960X that I've been abusing needs 1.25 volts at 4.5 GHz for bench stability and 1.275 volts for 24/7 stability at 4.5 GHz. Ice bucket cooling (-5C) gets it bench stable at 4.8 GHz at 1.4 volts. I've gotten it up to 4.86 GHz, but that's pushing it to the edge with 1.50 volts or higher (which is insane).

I can only imagine what the binned chip will do with an ice bucket. My goal? The 5 GHz club.

Arne Saknussemm
03-19-2016, 10:58 AM
My goal? The 5 GHz club.

With the 2%er for sure but you should be able to validate 5 with that chip..if it RealBenches4.8+....unless....I have seen some CPUs just stop scaling full stop no matter what volts...

Gobe
03-22-2016, 01:21 AM
Started working with the cherry-picked CPU today. Individual cores each hit 4.9 and passed RealBench H.264 with one core just a bit less solid than the others. Here's the weird part, overclocking with all cores blazing, I can't get a inch beyond 4.8 GHz. Weirder still, I was able to pass RealBench on my older 5960X (which has several cores that I couldn't run past 4.7 individually) at 4.86 GHz.

Maybe every CPU has it's own personality and I've not gotten to know this new one well enough yet? I'd given up on my old chip at 218K on RealBench then worked with it a few more sessions and benched it up over 223K. Still, I was hoping to just dial the cherry-picked unit right on up... nothing is so simple!

Menthol
03-22-2016, 02:17 AM
Gobe,
Reacting differently is about the only thing you can count on, I couldn't complete RB over 4.8 even using SS cooling, 5.0 2D and 4.9 3D benchmarks, I may have to connect my water chiller again to test the latest version of RB, I have just been using it with an AIO lately
You may as well enter the competition on HWBot, who knows you may win a prize

Arne Saknussemm
03-22-2016, 05:11 AM
Are you OCing cache at the same time? I have found over 4.8 you need cache to be bench stable....not 1 to 1 but say 42....44

1 to 1 cpu/cache is difficult at reasonable volts higher up...but as long as you can get it into 40s....

Gobe
03-22-2016, 02:47 PM
Menthol, different is right. I know it's got better cores than my 1st chip, so cooling is probably something to get in order before I go on with it. Got a more custom replacement for my AIO cooling on the way so maybe the ice bucket will work that much better once I get it installed.

I start with the cache running modest then push it as far as it will go toward the core frequency while walking the cache voltage up.

I'm still baffled over getting my 1st chip to pass RB at 4.86 GHz for the 223K score. I mean I'm not baffled at what I did, I know what I did... I'm just amazed that it actually worked. With the new binned chip, a single three hour session just isn't long enough and it takes the better part of a day to refreeze all those ice-packs. Gives me time to think about it between sessions!

Gobe
03-23-2016, 07:13 AM
Who else suffers from the impulse to try to push a new CPU to the edge as soon as you get your hands on it? I stepped back and set most voltages default and dialed in 47 x 100 for 4.7 GHz. Set the cache to run modest at 4.0 GHz and gave Vcore of 1.35 Volts and Vinput of 1.95 volts. Still got my Nepton 240M AIO cooler going though I've got some Phobya cooling goodness on the way. Set it up with the RB torture test and it's happily plugging away at 4.7 GHZ on 1.35 volts for the past half hour. I'm not thrilled to be up around 90C, but I'm not scared either. I'll call it a run at 60 minutes. I figure the Phobya loop should let me either trim a few millivolts or a few degrees.

Figure I'll collect myself and walk this baby up nice and slow.

Arne Saknussemm
03-23-2016, 07:56 AM
Honestly these days I wouldn't even consider an AIO cooler for OCing a 5960X past more than 4.2-4.4...I'll stick my neck out and guess your load temp will drop a huge amount...maybe 15-20 degrees on a custom loop.....obviously a whole lot more in an ice bucket!

I'd simply say your CPU is waiting for real cooling...and then you can walk it where you want....

Gobe
03-23-2016, 08:54 AM
As AIO goes, Coolermaster's Nepton 240M isn't a bad choice. Still the radiator is aluminum and at 120 x 240 x 25, it's kind of wimpy and I doubt the return water to the CPU is all the way back to ambient... not to mention flow rate and block engineering above the CPU cores probably aren't as optimized as a custom solution.

It didn't survive the full hour at 4.7 GHz/1.35 volts without a crash on the RB stress test. I backed the speed and voltage off just a bit to 4.65 GHz and 1.325 volts and just finished a successful one hour RB stress test. Hottest single core got up to 90C.

If someone told me that I could pass a one hour RB stress test with a 5960X at 4.65 GHz/1.325 volts on an AIO cooler, I would say, "yes, please". I'm quite looking forward to putting the Phobya UC-2 LT CPU block through it's paces along with the beefier pump and 120 x 360 x 60 copper radiator. Volume-wise, it's 3.6 time bigger than the Nepton radiator and there's no question the UC-2 LT is a top notch CPU block.

Menthol
03-23-2016, 11:04 AM
Gobe, if you don't mind what is the batch number, I and Arne for that matter have gone through several CPU's since X99 was released, when word came out that J batch processors clocked well on low voltage I went to Microcenter and had them search there stock to find one for me, it clocks to 4.6 on less than 1.3 volts but I have not had the hankering to bench in a while, instead I took my system off the bench table and placed it in a case with a 360 Predator AIO and purchased a couple fast drives and left pretty much alone and don't know if this one will clock as high as my last one. believe Arne has a fine CPU that isn't J batch, and now that X99 has been out for some time I don't hear anyone talking about batch numbers anymore

Arne Saknussemm
03-23-2016, 11:21 AM
I think J still have a rep for being good...where are they made?

Mine L batch...Malaysia

Gobe..over to you again :o
56535

Gobe
03-23-2016, 05:36 PM
Menthol, the cherry picked CPU is predictably a J batch. Don't recall where it was made. Won't be taking it off until the Phobya cooling shows up so I'll have a look then. The CPU I used for my RB score that Arne just passed for 2nd place is a Malaysia L batch.

As far as finding a quantifiable difference between the two CPUs, under air, I can enable the J batch cores one at a time and bench 4.9 on all of them. With the L batch, the individual cores benched between 4.7 and 4.9. However, once I pulled out the ice bucket, I got the L batch CPU up to 4.86.

Your 4.6 figure on less than 1.3 volts seems on target. If the Phobya doesn't show up today, I'll try to see how low I can take the voltage at 4.6. Considering I got 4.65 at 1.325 on AIO, the 4.6/1.3 figure might be obtainable. Once the Phobya cooling arrives, all bets are off since my J batch was binned for best of best.

cekim
03-23-2016, 08:57 PM
Menthol, the cherry picked CPU is predictably a J batch. Don't recall where it was made. Won't be taking it off until the Phobya cooling shows up so I'll have a look then. The CPU I used for my RB score that Arne just passed for 2nd place is a Malaysia L batch.

As far as finding a quantifiable difference between the two CPUs, under air, I can enable the J batch cores one at a time and bench 4.9 on all of them. With the L batch, the individual cores benched between 4.7 and 4.9. However, once I pulled out the ice bucket, I got the L batch CPU up to 4.86.

Your 4.6 figure on less than 1.3 volts seems on target. If the Phobya doesn't show up today, I'll try to see how low I can take the voltage at 4.6. Considering I got 4.65 at 1.325 on AIO, the 4.6/1.3 figure might be obtainable. Once the Phobya cooling arrives, all bets are off since my J batch was binned for best of best.
My 4.8 capable chip does so at 1.36-1.37 (I really don't want to push it further than that anyway). The funny thing is that lower than 1.3 @ 4.5 or 4.6 (doesn't matter), I eventually (day to weeks). Even funnier, is that the crash occurs under light load (cold to first use speed step - first key-strokes to spin up after idling all night).

Wondering if maybe I am tweaking the wrong voltage to squash whatever it is that causes long-term instability?

Wondering even more if I'm still getting gremlins from SLI + 10GbE + M.2 under linux. None of those things has struck me as terribly well worn paths, less still all in one place. I had to do the PCIe dance a few times just to prevent random boot hang-ups for resource contention (m.2 and 10GbE card don't appreciate eachother's company AT ALL).

I may swap the strix GTX980's out for the 980ti lightning and walk those voltages down this weekend. It would be frustrating to lose the Cuda/openCL power, but good to know if I am pushing more voltage than required for the processor because my PCIe cards are fighting eachother.

Gobe
03-23-2016, 09:44 PM
Sounds like the speed step then [Edit: said with the confidence of someone who doesn't have to deal with the problem]. I turn all that stuff off since it only costs me around 40 watts at idle. I just turn off an extra light and the power company is none the wiser... Until I crank it up and the entire grid sags under the load.

cekim
03-23-2016, 10:22 PM
Sounds like the speed step then [Edit: said with the confidence of someone who doesn't have to deal with the problem]. I turn all that stuff off since it only costs me around 40 watts at idle. I just turn off an extra light and the power company is none the wiser... Until I crank it up and the entire grid sags under the load.

Could also be the PSU. Not the top-end but the bottom. These experiments take weeks as the failure takes that long to show...

As for speedstep, it isn't the money or the planet, just want to help ensure that when I retire the chip because its obsolete, not because I baked it doing nothing overnight when I didn't have to. Some nights, many nights, I do need to bake it.

Menthol
03-23-2016, 11:11 PM
So did Arne move back out in front? I bet that sneaky guy was holding back all along laughing at us

Gobe
03-24-2016, 12:28 AM
Yea, the little sneak moved back ahead of me into 2nd place. Was messaging with Arne earlier and he made an interesting observation about the Xeons up at the top. The leaderboard shows them both as 8 cores, but the two motherboards are DUAL XEON boards. Now we know where those monster H.264 scores come from!

Just made some interesting stress test runs on my cherry 5960X. At 4.6 GHz and 1.200 volts, it BSOD at the login screen. Bump it up to 1.253 volts and it chugs right through an hour of RealBench stress test at 4.6 GHz. I'm still trying to figure out what to make of a 53 mV difference between a BSOD at login and a successful one hour RB stress test stability slamming HandBrake among other things. Hottest core topped out at 77C on AIO cooling.

My older 5960X does 4.5 GHz at 1.300 volts, so 4.6 GHz at 1.243 volts is pretty sweet.

The guys who sold me the Phobya unit had it listed as in stock. I sent them a note today because I'd not yet received a shipment notification (paid for it last weekend). They wrote back to tell me that they didn't have the item in stock, but were having it shipped in and they'd notify me when they ship it out. Apparently the term "In Stock" in an item description is open to considerable interpretation these days. Oh well, it's not like I've never BS'ed a customer or two on getting stuff shipped here from Germany.

cekim
03-24-2016, 01:21 AM
Yea, the little sneak moved back ahead of me into 2nd place. Was messaging with Arne earlier and he made an interesting observation about the Xeons up at the top. The leaderboard shows them both as 8 cores, but the two motherboards are DUAL XEON boards. Now we know where those monster H.264 scores come from!

Just made some interesting stress test runs on my cherry 5960X. At 4.6 GHz and 1.200 volts, it BSOD at the login screen. Bump it up to 1.253 volts and it chugs right through an hour of RealBench stress test at 4.6 GHz. I'm still trying to figure out what to make of a 53 mV difference between a BSOD at login and a successful one hour RB stress test stability slamming HandBrake among other things. Hottest core topped out at 77C on AIO cooling.

My older 5960X does 4.5 GHz at 1.300 volts, so 4.6 GHz at 1.243 volts is pretty sweet.

The guys who sold me the Phobya unit had it listed as in stock. I sent them a note today because I'd not yet received a shipment notification (paid for it last weekend). They wrote back to tell me that they didn't have the item in stock, but were having it shipped in and they'd notify me when they ship it out. Apparently the term "In Stock" in an item description is open to considerable interpretation these days. Oh well, it's not like I've never BS'ed a customer or two on getting stuff shipped here from Germany.

What cache freq/voltage, if I may ask?

Menthol
03-24-2016, 03:27 AM
Gobe, where do you live if I may ask

Gobe
03-24-2016, 05:49 AM
cekim, Cache was running (I think) 4.2 GHz set to auto. I've been able to run the cache and cores both at 4.8 with stability, but that takes a lot of cache voltage and extra heat to deal with for not so much advantage. Right now I'm really frustrated with the AIO cooler I'm running while I wait for the Phobya. Sure, I can get an ice bucket nice and cold, but if the pump/block don't assault the thermal interface with a brutal water flow and smart engineering, it's all for naught. AIO just isn't up to the job where the rubber meets the road. I'm sure if I somehow cracked the AIO block open, my first word would be "wow", and not a good "wow".

Menthol, USA, Southeast, though any given week I can be working anywhere in the US or occasionally Germany.

Arne Saknussemm
03-24-2016, 09:51 AM
I'm sure if I somehow cracked the AIO block open, my first word would be

I did this to a H100 back in the day...after it leaked...my first words were "what a load of ba**s! Where's that EK link to build a real loop!"

cekim
03-24-2016, 03:17 PM
I did this to a H100 back in the day...after it leaked...my first words were "what a load of ba**s! Where's that EK link to build a real loop!"
Chuckle... $100 water cooling loop, what did you expect? ;-) I'm very pleased with mine, but I set my expectations appropriately. I'm equally pleased with with my EK loop, but it was a lot more work and money (and worth every penny).

cekim
03-28-2016, 06:53 PM
So, 4.5, 4.6 and now 4.7 @ 1.3V. 4.875 required 1.38v, but I can't beat my or Gobe's Realbench without pulling some dimms (this is with 8x16=128G at 2800). Can't get 3000 much less 3200 running with all 8 populated.

You can see it got a little warm under prolonged stress (a few hours), but still manageable.

Mind you, this is without "cooling heroics" (beyond an EK monoblock, which while awesome, requires no ice or other shenanigans.

cekim
03-28-2016, 06:55 PM
Oh and someone had mentioned ramdisks to speed up realbench not working. I saw the same. I suspect you are just increasing the chatter on the IMC as a result of reading/writing ram as if it were a disk. Better off just letting the cache do its job and/or have a really fast disk in the first place.

Gobe
03-30-2016, 07:20 PM
Got the custom loop and 5 GHz (http://valid.x86.fr/tdhvyw).

Arne Saknussemm
03-30-2016, 07:56 PM
cekim...Gobe Nice! :cool:

cekim
03-30-2016, 08:37 PM
got the custom loop and 5 ghz (http://valid.x86.fr/tdhvyw).

nice!

Menthol
03-31-2016, 02:31 AM
I for one prefer shenanigans and think they should be mandatory :)

Good stuff going on here I like it

cekim
03-31-2016, 02:38 AM
I for one prefer shenanigans and think they should be mandatory :)

Good stuff going on here I like it

Only if they are light-hearted and fun. If they are tragic and cruel then they aren't even really shenanigans are they?

cekim
03-31-2016, 02:40 AM
My latest is in the other thread - 5GHz club official on this cpu now at 1.35 which is just dumbfounding. It's not stable enough to bench there, but can browse around in windows indefinitely.

As far as stability, 4.8 @ 1.32 made it 1:06 in aid64 - and then stopped complaining of an error. That has one core hitting 97C, so it looks like baring cooling shenanigans, 4.7 is the daily-driver limit for this one...

However... given that 4.8/4.9/5.0 are all relatively stable at ludicrously low voltages, there is some fun to be had here beyond daily driver.