• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Higher mem clock speeds vs lower latency?

1.4V but it's Micron B die, not Samhung.

you mean Micron E Die ? unless you have v4.32 of Corsair Vengeance LPX sticks, then you would have samsung B-dies.
 
Wish I had your patience and ram kit! LPX's @3800mt 18 21 21 21 40 Is the most mine do but I'm scared to push it too much. Need a working computer!
3800 does not work on 11gen ..... only 3733 3866 100:133 mode
 
3800 does not work on 11gen ..... only 3733 3866 100:133 mode
Gear 2, half the speed tho so kinda pointless, I've been trying to get it to post at 3866 gear 1 but no matter how much voltage I put on imc it will not post, tried 1.5v on both SA and IO2 this morning and even 1.65v on dram but no success. I'm reading up abit more on the secondary and tertiary timings. Maybe i can push some more out of these sticks
 
TPU has some good reviews on all of this


The summary comes down to:

Ryzen cares more that you have four ranks of memory

Using the wrong gear mode/not being 1:1 with your memory controller is a huge hit

Oh and 3600C14 dominates in DDR4, with CL16 being super close
Intel 12900K
1659251073518.png



Ryzen 5800x:
Note how the SR (single rank) kits tank massively. Either run 4x8 or 2x16 if you want zen 3 to perform right.
Even CL numbers also benefit ryzen, since it rounds up odd numbers.
1659251130255.png



12900K has some small gains from higher memory clock speeds, but you'll note good old dual rank DDR4 3600 still being pretty high up - because it can run the RAM controller at 1:1 unlike the 4000 kit tanking the charts
1659251247579.png



TL;DR: Get fast enough that you can run 1:1 with your memory controller, as low latency as you can afford/justify.
And get 32GB so you're dual rank. Worth it.
 
TPU has some good reviews on all of this


The summary comes down to:

Ryzen cares more that you have four ranks of memory

Using the wrong gear mode/not being 1:1 with your memory controller is a huge hit

Oh and 3600C14 dominates in DDR4, with CL16 being super close
Intel 12900K
View attachment 256523


Ryzen 5800x:
Note how the SR (single rank) kits tank massively. Either run 4x8 or 2x16 if you want zen 3 to perform right.
Even CL numbers also benefit ryzen, since it rounds up odd numbers.
View attachment 256524


12900K has some small gains from higher memory clock speeds, but you'll note good old dual rank DDR4 3600 still being pretty high up - because it can run the RAM controller at 1:1 unlike the 4000 kit tanking the charts
View attachment 256525


TL;DR: Get fast enough that you can run 1:1 with your memory controller, as low latency as you can afford/justify.
And get 32GB so you're dual rank. Worth it.
thats some solid info, Thanks! yeah as ive said before, only when im at 4600mhz gear 2 i have the same latency as with xmp 3200mhz gear 1 but well over 20.000MB/s on average more on read, write and copy speed. but in reality those gains are useless to me at this point, because everything is already running smooth, still at 1080p with rtx3060ti, my frames will be solid for awhile if i stay at 1080p.
currently at 3733mhz 19-19-19-43 1.45v, stable with extreme TestMem5 config its the same timings as with default 2667. so in theory i should be able to get lower latencies with 3600mhz. and maybe even lower with 3200.
 
Last edited:
thats some solid info, Thanks! yeah as ive said before, only when im at 4600mhz gear 2 i have the same latency as with xmp 3200mhz gear 1 but well over 20.000MB/s on average more on read, write and copy speed. but in reality those gains are useless to me at this point, because everything is already running smooth, still at 1080p with rtx3060ti, my frames will be solid for awhile if i stay at 1080p.
currently at 3733mhz 19-19-19-43 1.45v, stable with extreme TestMem5 config its the same timings as with default 2667. so in theory i should be able to get lower latencies with 3600mhz. and maybe even lower with 3200.
depends on RAM - mine clocks up the MHz easy enough, but refuses to lower primary latencies at all

The big thing these days is that RAM has sped up massively while IMC's have not
 
depends on RAM - mine clocks up the MHz easy enough, but refuses to lower primary latencies at all

The big thing these days is that RAM has sped up massively while IMC's have not
I see, so would I have a better chance of getting lower latency if I get an 11900kf ? Since that's an i9 those are usually higher binned and thus also better IMC perhaps ?
 
Hahaha if im really honest,im just suprised these mem sticks can even go that far with only basic adjustments of timings and voltages. Default is 2667mhz, 4500mhz is the highest speed I can get it stable, so thats still a massive overclock. Not sure if thats a ddr4 thing. My old ddr3 1866 sticks cant boot past 2133mhz, no matter what setting I adjust. so is overclocking improved alot with ddr4?
lol. it really depends. i can tell you a lot of ddr4 examples which DOESN'T OC. like literally. plain trash, xmp, or.. change your memory. no boot either with changed timings or speed, dependless of voltage...
and, you must have very good MB to allow your oc and also good cpu IMC. so, there are many factors that allow you either oc or great oc:D
PS i haven't seen any "good" ddr3 ram, but person i know told me he had ones that oc from 1600 or 1866 to 2400! ddr4 "high speed" is just more available now than it was with "higher than 1600" ddr3. heck, even hyped-out 1866 was RARE.
 
lol. it really depends. i can tell you a lot of ddr4 examples which DOESN'T OC. like literally. plain trash, xmp, or.. change your memory. no boot either with changed timings or speed, dependless of voltage...
and, you must have very good MB to allow your oc and also good cpu IMC. so, there are many factors that allow you either oc or great oc:D
PS i haven't seen any "good" ddr3 ram, but person i know told me he had ones that oc from 1600 or 1866 to 2400! ddr4 "high speed" is just more available now than it was with "higher than 1600" ddr3. heck, even hyped-out 1866 was RARE.
yeah there are alot of factors at play, i agree. But i do know that the IMC of my 11600kf is currently blocking my advance to 3866mhz (gear 1) and higher speeds. In gear 2, ive reached 4800mhz. The max stable clock on my 11600kf is pretty high, slightly better clockspeeds than an i9 11900kf max turbo boost frequency, myn just has just 2 cores less and also 4mb L3 cache less . So in terms of cpu and memory silicon lottery i won i guess, but the overclocker in me is still not satisfied with the current mem clock speed/latencies, then again, is it ever ? xD thinking about buying an i9 11900kf, perhaps the IMC of the higher binned 11th gen cpu's allow for higher mem clock speeds in gear 1
 
I see, so would I have a better chance of getting lower latency if I get an 11900kf ? Since that's an i9 those are usually higher binned and thus also better IMC perhaps ?
That's certainly a viable option for anyone on 11th gen platform but ask yourself, is it really worth it considering the costs of an i9? The money used on an i9 (I'm guessing new here) could go a good way to an Alderlake platform or if you can wait a bit, the Raptor lake platform.
 
That's certainly a viable option for anyone on 11th gen platform but ask yourself, is it really worth it considering the costs of an i9? The money used on an i9 (I'm guessing new here) could go a good way to an Alderlake platform or if you can wait a bit, the Raptor lake platform.
i can get an new 11900kf for 375 euro where i live, i sell my 11600kf for maybe 150 euro. so then the i9 technicly only costs me 225 euro right now. I just upgraded my whole platform like 2 months ago. my next full platform upgrade is going to be at least 5 years from now. my last setup also lasted 10 years (sandy bridge platform), 8 cores and 16 threads at 5.3ghz of the 11th gen i9 is more then enough for the next few years. im not a fan of the 8 E cores on alder lake, ive heard its a bitch with overclocking. Raptor lake have been labeled as "alder lake refresh" but still, wouldnt justify a 800 euro upgrade. maybe if meteor lake comes out in 2023/2024 ill upgrade because its build on the 4nm node! it all depends on how long my system can keep up i guess.
 
Last edited:
i can get an new 11900kf for 375 euro where i live, i sell my 11600kf for maybe 150 euro. so then the i9 technicly only costs me 225 euro right now. I just upgraded my whole platform like 2 months ago. my next full platform upgrade is going to be at least 5 years from now. my last setup also lasted 10 years (sandy bridge platform), 8 cores and 16 threads at 5.3ghz of the 11th gen i9 is more then enough for the next few years. im not a fan of the 8 E cores on alder lake, ive heard its a bitch with overclocking. Raptor lake have been labeled as "alder lake refresh" but still, wouldnt justify a 800 euro upgrade. maybe if meteor lake comes out in 2023/2024 ill upgrade because its build on the 4nm node! it all depends on how long my system can keep up i guess.
Hey, everyone's situation is different, but your money.. .your call! :)
 
Hey, everyone's situation is different, but your money.. .your call! :)
yeah still not sure, because even tho those i9 are higher binned, its not guaranteed the IMC will run lets say 4000mhz+ speeds in gear 1. so worse case scenario i would spend 200-250 for 2 extra cores and 4 mb of L3 cache extra. tough decision haha
 
Overclocking is not guaranteed. Makers do some binning, whatever comes out of that, that is what's being guaranteed.
Even when most people can get 20% above the baseline, it doesn't mean you can't draw the short straw get a SKU that will not get anything above 1-3%.
 
Overclocking is not guaranteed. Makers do some binning, whatever comes out of that, that is what's being guaranteed.
Even when most people can get 20% above the baseline, it doesn't mean you can't draw the short straw get a SKU that will not get anything above

Exactly, like I said, a higher binned cpu is gonna perform better than a lower binned cpu, and thus most likely, a better performing IMC in comparison with the i5.
 
Exactly, like I said, a higher binned cpu is gonna perform better than a lower binned cpu, and thus most likely, a better performing IMC in comparison with the i5.
Not really, no. If all SKUs support the same RAM, you can bet all IMCs are binned the same. For AMD, the IMC is on a separate die even.
 
Not really, no. If all SKUs support the same RAM, you can bet all IMCs are binned the same. For AMD, the IMC is on a separate die even.
So you wanna tell me an i5 unlocked has the same quality imc as an i9 ? That's just not true. Because that would mean that all the chips in the same generation are the same quality. No lol haha
 
So you wanna tell me an i5 unlocked has the same quality imc as an i9 ? That's just not true. Because that would mean that all the chips in the same generation are the same quality. No lol haha
I am telling you if the IMC is supposed to support up to DDR4-3200 without overclocking, that's what it's binned for. It doesn't matter if it's part of an i3 or an i9. And again, for AMD, it's actually produced separately, meaning there's even less correlation between the performance of the CPU and the IMC.
 
I am telling you if the IMC is supposed to support up to DDR4-3200 without overclocking, that's what it's binned for. It doesn't matter if it's part of an i3 or an i9. And again, for AMD, it's actually produced separately, meaning there's even less correlation between the performance of the CPU and the IMC.
And you think I'm not gonna overclock that i9? We were talking about overclocking, not the default supported speed and I don't care about AMD, we were discussing intel, stay on topic please.

I am telling you if the IMC is supposed to support up to DDR4-3200 without overclocking, that's what it's binned for. It doesn't matter if it's part of an i3 or an i9. And again, for AMD, it's actually produced separately, meaning there's even less correlation between the performance of the CPU and the IMC.
Besides, binning process is done by how high the cpu can clock with stable temperatures and voltages, meaning, its all depended on the quality of the chip in general, that includes the INTEGRATED memory controller, so an i9 has a much better chance of reaching higher (mem) speed than an i3 or i5 or i7
 
Last edited:
Besides, binning process is done by how high the cpu can clock with stable temperatures and voltages, meaning, its all depended on the quality of the chip in general, that includes the INTEGRATED memory controller, so an i9 has a much better chance of reaching higher (mem) speed than an i3 or i5 or i7
Technically possible, but none of us can put numbers on that. That "much better chance" can translate to anything from 5 to 30%.
 
Technically possible, but none of us can put numbers on that. That "much better chance" can translate to anything from 5 to 30%.
ofcourse i agree. but the chances of an 11th gen i5 doing 5.3ghz all cores with 1.48v under load is probably less than 5%. those speeds are reserved for the i9 chips. i was really really lucky with this chip. i could be very unlucky if i upgrade to an i9. i really dont know if i wanna upgrade to the i9, maybe ill just get an 11700kf and just settle for 2 cores extra to future proof my system a little bit more.
 
ofcourse i agree. but the chances of an 11th gen i5 doing 5.3ghz all cores with 1.48v under load is probably less than 5%. those speeds are reserved for the i9 chips. i was really really lucky with this chip. i could be very unlucky if i upgrade to an i9. i really dont know if i wanna upgrade to the i9, maybe ill just get an 11700kf and just settle for 2 cores extra to future proof my system a little bit more.
I've found the opposite tbh -- the x700 series that are right below the i9 are really hard to oc (11700, 12700, 10850K for example) since they're binned worse. But the i5's since they're disabling cores seem to OC great, and they don't have heat issues. Of the last 3 12600K i5's we built every one could hit 5.2Ghz-5.4Ghz, for the 12700K if we touched the OC at all it was too hot and couldn't clock past ~5.1.

I would be curious to see if they're disabling the weakest cores, and if, because it's a volume product they have to send in some higher binned samples to meet demand.
 
I've found the opposite tbh -- the x700 series that are right below the i9 are really hard to oc (11700, 12700, 10850K for example) since they're binned worse. But the i5's since they're disabling cores seem to OC great, and they don't have heat issues. Of the last 3 12600K i5's we built every one could hit 5.2Ghz-5.4Ghz, for the 12700K if we touched the OC at all it was too hot and couldn't clock past ~5.1.

I would be curious to see if they're disabling the weakest cores, and if, because it's a volume product they have to send in some higher binned samples to meet demand.
Alder lake is a bitch for overclocking because of the useless E-cores even though its build on 10nm which is much more power efficient because of the increased amount of transistors on the die, it still draws additional power and thus, heat, increasing tdp massively, while gaining not much in comparinson. the 11th gen doesnt have E cores, but is the last gen 14nm, which as you might know, is incredibly power hungry, when i fire up prime95 or IBT or any other AVX(512) workload for the matter, the tdp of my cpu goes well over 200 watt. a friend has almost the same setup as me, but his cpu requires 1.57v to achieve the same clockspeed as my i5
 
Last edited:
Back
Top