• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

I9 9900k

easyrider said:
I'm up and running but everytime i try and load the xmp profile I can't boot
Click to expand...
Are you using xmp 1 or 2 in BIOS? I understand 2 has looser timings so should be easier to set.

Failing that you may need to set the timings yourself or increase vccio and or vccsa a little.

What memory do you have?
 
Distracted said:
Are you using xmp 1 or 2 in BIOS? I understand 2 has looser timings so should be easier to set.

Failing that you may need to set the timings yourself or increase vccio and or vccsa a little.

What memory do you have?
Click to expand...


Corsair CMK16GX4M2B3000C15 Vengeance LPX 16 GB (2 x 8 GB) DDR4 3000 MHz C15 XMP 2.0 High Performance Desktop Memory Kit, Black
 
khemist said:
Thanks, what's a ballpark average voltage for 5.0ghz on these chips?.
Click to expand...

Distracted said:

Some updated figures regarding the silicon lottery.

Keep in mind their sample size and that it may not represent your own chances when buying a 9900k.

I think this is useful info for those overclocking. No info yet on why those particular voltages were chosen or any temperature data.

As of 12/07/18, 100% (no change) of tested 9900Ks were able to hit 4.8GHz or greater.
  • CPU Multiplier: 48
  • BCLK: 100.0
  • CPU Vcore: 1.275V
  • AVX Offset: 2

As of 12/07/18, the top 85% (+3%) of tested 9900Ks were able to hit 4.9GHz or greater.
  • CPU Multiplier: 49
  • BCLK: 100.0
  • CPU Vcore: 1.287V
  • AVX Offset: 2

As of 12/07/18, the top 41% (-5%) of tested 9900Ks were able to hit 5.0GHz or greater.
  • CPU Multiplier: 50
  • BCLK: 100.0
  • CPU Vcore: 1.300V
  • AVX Offset: 2

As of 12/07/18, the top 11% (-3%) of tested 9900Ks were able to hit 5.1GHz or greater.
  • CPU Multiplier: 51
  • BCLK: 100.0
  • CPU Vcore: 1.312V
  • AVX Offset: 2

Silicon lottery now have a larger sample size and as such the percentage of silicon meeting their requirements for each bin has changed.

As of 07/12/18 100% of 9900ks can hit 4.8GHz all core, while there has been an increase in the number of cpus able to hit 4.9GHz by 3% taking the number to 85%.

That's the end of the increases.

For 5.0GHz all core the percentage has dropped by 5% down to 41%. Almost 60% of 9900ks wont reach 5.0GHz all core, or roughly 3 out of every 5 9900ks are failing to reach 5.0GHz all core at the settings listed above.

Next a drop of 3% down from last month bringing the percent of 9900ks able to reach 5.1GHz all core to just 11%.
Just over 1 in 10 of the samples they are binning can hit 5.1GHz stable all core at their specified settings.

As always with these numbers there is no telling how representative of the 9900k supply as a whole the stock silicon lottery receives is. However as they bin more CPUs the numbers should more accurately reflect the silicon lottery for those considering buying retail or tray.

Finally, still no 5.2GHz bin listed. This doesn't mean they don't exist, there are probably 5.1GHz samples that will do 5.2GHz, but judging by the fact that no one is offering binned 5.2GHz 9900ks (when there are clearly people willing to pay for it) the number of stable 5.2GHz chips must be tiny.
Click to expand...

Right now there isn't a more accurate guide than the figures released by silicon lottery.
 
Silent_Scone said:
Nothing in that post mentions workload. Considering they're applying an AVX offset of -2 across the board, too.
Click to expand...

No, the testing they do is something they keep to themselves but since their reputation is built on stable overclocks they do themselves no favours by skimping on the workloads they test with.

As of right now, there is no other set of stats for "we are willing to sell the 9900k at this oc with these settings" being released. So while silicon lottery might not be the best, they are literally the only source we have for this info outside of anecdotal evidence.

Every cpu they test is done with the same settings and hardware and they test large volumes. That is as close as we are going to get for binning figures outside of someone like ocuk or caseking or intel releasing binning stats.

If they say they are willing to sell 5.0GHz at 1.3 and only 41% pass then to me that is a valid indicator of what the 9900k lottery is like.
If they could sell more cpus at 5.0GHz or higher they would, it would get them more profit. Likewise if a higher vcore was reasonable for the vast majority of the people they are selling to (aio owners for instance) then they would go that route.

The fact they can't gives us valuable info.


In context it is useful and you can apply some common sense to each of the figures to make it fit your own use case.

Custom loop - you can probably push the vcore much higher and still keep reasonable temps.

On air - vcore may need to be much lower before you hit thermal limits.

Use a higher avx offset like -3 or -4 - perhaps you can use lower vcore.

Use no avx offset - you will likely need much higher vcore.



The info is as useful as we make it and right now it is all we have.
 
Last edited:
Distracted said:
No, the testing they do is something they keep to themselves but since their reputation is built on stable overclocks they do themselves no favours by skimping on the workloads they test with.

As of right now, there is no other set of stats for "we are willing to sell the 9900k at this oc with these settings" being released. So while silicon lottery might not be the best, they are literally the only source we have for this info outside of anecdotal evidence.

Every cpu they test is done with the same settings and hardware and they test large volumes. That is as close as we are going to get for binning figures outside of someone like ocuk or caseking or intel releasing binning stats.

If they say they are willing to sell 5.0GHz at 1.3 and only 41% pass then to me that is a valid indicator of what the 9900k lottery is like.
If they could sell more cpus at 5.0GHz or higher they would, it would get them more profit. Likewise if a higher vcore was reasonable for the vast majority of the people they are selling to (aio owners for instance) then they would go that route.

The fact they can't gives us valuable info.

The fact they test with an avx offset of -2 also means that we can expect higher vclores for anyone not wanting to use an offset.

In context it is useful and you can apply some common sense to each of the figures to make it fit your own use case.

Custom loop - you can probably push the vcore much higher and still keep reasonable temps.

On air - vcore may need to be much lower before you hit thermal limits.

Use a higher avx offset like -3 or -4 - perhaps you can use lower vcore.

Use no avx offset - you will likely need much higher vcore.



The info is as useful as we make it and right now it is all we have.
Click to expand...

It actually hurts users not telling them. If someone doesn't plan on using heavy AVX workloads, then the statistics may do more harm than good when setting out to find a baseline for stability. They may not be open about it here, but they have been in the past. Prime95 with AVX is part of their regime; if we remove this from the equation the margins for CPU hitting 5GHz would climb. This is because nothing pulls as much current as Prime does, not even x256. When offering factory or pre-binned overclocking, there is significant guardband in place. More so than individual users would likely need.

TLDR; The required VID for each individual CPU varies a lot, so showing figures based on what a set number of CPU can or can't do without a baseline of workload means you're best off not paying too much attention and finding your own baseline for the workload that suits you.

I recommend the HWBOT x265 benchmark with multiple instances for testing AVX workloads.
 
Silent_Scone said:
It actually hurts users not telling them. If someone doesn't plan on using heavy AVX workloads, then the statistics may do more harm than good when setting out to find a baseline for stability. They may not be open about it here, but they have been in the past. Prime95 with AVX is part of their regime; if we remove this from the equation the margins for CPU hitting 5GHz would climb. This is because nothing pulls as much current as Prime does, not even x256. When offering factory or pre-binned overclocking, there is significant guardband in place. More so than individual users would likely need.

TLDR; The required VID for each individual CPU varies a lot, so showing figures based on what a set number of CPU can or can't do without a baseline of workload means you're best off not paying too much attention and finding your own baseline for the workload that suits you.

I recommend the HWBOT x265 benchmark with multiple instances for testing AVX workloads.
Click to expand...

I'm not disputing what you are saying about how avx affects vcore needed and the guardband required.

However, if we know this is the case then we can reasonably assume that non avx figures will require lower vcores.

They are giving us information. We know they use an avx offset. We know they test with avx. If we as users want a guide, and that is really all it is, to what the average 9900k is capable of then we have it.

As i said above, using the info they give us we can make several assumptions about vcore and temps with differing workloads and coolers.

It isn't a set in stone list of stats, it is a guide based on carefully controlled testing which you can make useful for your own usecase by applying some common sense to the figures you see.

Khemist wanted a ballpark average for oc the 9900k to 5.0GHz and the info i provided can do that within the specified limits. Outside of those limits anyone can make the reasonable assumptions i mentioned and still come up with ballpark averages.

Giving anecdotal evidence is not accurate or a true representation of the lottery. Taking standardised testing of large volumes of 9900ks at known settings and applying it to your own situation is far more useful imho.

If you can't make the leap from their settings to your own situation then yes the stats can be misleading, but i assume a certain level of understanding when posting and do make sure i say the stats may not be representative of the actual lottery when sharing the info.
This is because there is some bias based on how they test. As you say, they will need higher vcores at a given speed due to the avx load and they will need a lower cut off on vcore at a given speed (due to temp) because they are testing with aios and not custom loops.

But again, it is the best info we have right now.
 
Giving someone a range of VID to try is less anecdotal than saying “x number of CPU can do something, but I can’t tell you what that something is.” lol

The stats quoted are helpful to the extent one can gauge roughly where their own sample stands. Not how to overclock it. The point about guardband required is from the vendors perspective, they need to guarantee that X sample can do a set frequency under a certain load condition. Normally, this will be with an AIO, too.

They do serve a purpose, but without a workload it’s as anecdotal as first hand experience.
 
Last edited:
Hang fire, Boot times and everything have improved since moving the sticks of ram over to the other slots....

I'll run some benchies and report back...

2015 in cinebench...Is this about right? stock 4.7ghz
 
Silent_Scone said:
Giving someone a range of VID to try is less anecdotal than saying “x number of CPU can do something, but I can’t tell you what that something is.” lol
Click to expand...
Neither is perfect. In an ideal world we would have absolute transparency from Intel and this conversation would be unnecessary.

Having seen people quoting different vcores from cpuid, hwinfo, bios, underload, with llc x on board y or z I find it hard to take any meaningful info from the stats people post.

Not to mention the inherent bias towards people with decent silicon posting vs those with poor clockers remaining silent.

Gathering a range of vcore to use from that quagmire of settings and willy waving is less useful, imo, than taking a large scale controlled test and using common sense to make reasonable assumptions.

But this is only my opinion and as you point out, the stats from silicon lottery are not a perfect representation either.

They release the info and I'll continue to post it since i think it is useful. But i agree, that it is limited in what it can tell us.
 
easyrider said:
Hang fire, Boot times and everything have improved since moving the sticks of ram over to the other slots....

I'll run some benchies and report back...

2015 in cinebench...Is this about right? stock 4.7ghz
Click to expand...
Sounds like your board has mce enabled or some other boost going on. Iirc that is 5GHz ish scoring.

Edit - that isn't a bad thing. But you may find that temps are higher with mce than dialing in a 5GHz allcore oc yourself because the boards tend to use higher voltages than if you set them yourself.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom