DDR4 Overclock Pushed To 5 GHz With 4266 MHz G.Skill Kit

[IInuyasha74]

G.Skill announced that an enthusiast overclock has succeeded in pushing the company’s DDR4 memory to 5 GHz.

DDR4 Overclock Pushed To 5 GHz With 4266 MHz G.Skill Kit : Read more

 

[turkey3_scratch]

Wow, that's fast.

 

[Fates_Demise]

and he only had to completely screw over his memory timings to gain that little extra...

 

[Fates_Demise]

Stock Timing is 19-26-26-46, for anyone too lazy to look

 

[bmguyii]

Is there any tangible gain for this? The memory is not actually any faster so latency shouldn't change. The bandwidth is likely exceeding the internal fabric bandwidth. What's a stream score gain for this? Probably not much so while you can brag, you're really just wasting power and burning up your memory.

 

[Fates_Demise]

Is there any tangible gain for this? The memory is not actually any faster so latency shouldn't change. The bandwidth is likely exceeding the internal fabric bandwidth. What's a stream score gain for this? Probably not much so while you can brag, you're really just wasting power and burning up your memory.

He is actually making the memory faster, but with those timings id be surprised if he had any gain at all overall.
And to answer your main question no there is no point in this. Benchmark tests have already shown that speeds over 2100mhz have very little gain in performance even for programs that utilize the bandwidth heavily. So short of maybe a server system its all for bragging rights.

 

[alidan]

Is there any tangible gain for this? The memory is not actually any faster so latency shouldn't change. The bandwidth is likely exceeding the internal fabric bandwidth. What's a stream score gain for this? Probably not much so while you can brag, you're really just wasting power and burning up your memory.

He is actually making the memory faster, but with those timings id be surprised if he had any gain at all overall.
And to answer your main question no there is no point in this. Benchmark tests have already shown that speeds over 2100mhz have very little gain in performance even for programs that utilize the bandwidth heavily. So short of maybe a server system its all for bragging rights.

Is there any tangible gain for this? The memory is not actually any faster so latency shouldn't change. The bandwidth is likely exceeding the internal fabric bandwidth. What's a stream score gain for this? Probably not much so while you can brag, you're really just wasting power and burning up your memory.

some games do gain frames due to memory speed, i can't say how many, for the longest time i always worked on the assumption how fast does not matter so long as you have it.

one place did a comparison from i thin 1800 to 3200 mhz all with the same cpu speed, only memory changed, and the difference was tangible, something around a 60 fps spread on the worst of the games.

not every game, but enough to make memory speed something to consider going forward.

 

[neblogai]

I was also sure there is no much difference between fast and slow DDR4- but then I saw what happens if your other components are fast enough. It turns out, fast memory may be important, and can add up to 50% performance extra (Fallout4):
http://www.techspot.com/article/1171-ddr4-4000-mhz-performance/

 

[none12345]

Ummm....cas31? What a waste of time....

Look it can go super fast....by going super slow.... heh....

 

[photonboy]

Fast DDR4 memory matters when used as GPU memory. When the top-end Zen/Polaris 14nm APU's come out we'll need pretty fast memory to avoid a bottleneck of the GPU.

4000MHz DDR4 at least for the best APU's in 2017.

 

[Fates_Demise]

Is there any tangible gain for this? The memory is not actually any faster so latency shouldn't change. The bandwidth is likely exceeding the internal fabric bandwidth. What's a stream score gain for this? Probably not much so while you can brag, you're really just wasting power and burning up your memory.

He is actually making the memory faster, but with those timings id be surprised if he had any gain at all overall.
And to answer your main question no there is no point in this. Benchmark tests have already shown that speeds over 2100mhz have very little gain in performance even for programs that utilize the bandwidth heavily. So short of maybe a server system its all for bragging rights.

Is there any tangible gain for this? The memory is not actually any faster so latency shouldn't change. The bandwidth is likely exceeding the internal fabric bandwidth. What's a stream score gain for this? Probably not much so while you can brag, you're really just wasting power and burning up your memory.

some games do gain frames due to memory speed, i can't say how many, for the longest time i always worked on the assumption how fast does not matter so long as you have it.

one place did a comparison from i thin 1800 to 3200 mhz all with the same cpu speed, only memory changed, and the difference was tangible, something around a 60 fps spread on the worst of the games.

not every game, but enough to make memory speed something to consider going forward.

Highly doubtful on that test. Toms did a same test from I want to say 800mhz to like 2800mhz and there was less than a 10% difference from top to bottom. After 1800 less than a 3% difference in fps.
Games do not need much memory speed its capacity that mostly matters for memory in games.

 

[Alan Caldwell]

The time has come when ram is clocked faster than processors! :lol:

 

[Fates_Demise]

The time has come when ram is clocked faster than processors! :lol:

Not quite... if comparing world record ram then vs world record cpu I believe is like 8.4ghz..or close to that

 

[beetlejuicegr]

I have to ask, will we ever see ddr4 or 5(later on) memories with low latency? like CL 9-11 or even less?
Is that an impossible thing even in theory? Or manufacturers see it easier and more cost effective to dish out DDRx+1 than make lower latency?

 

[none12345]

"I have to ask, will we ever see ddr4 or 5(later on) memories with low latency? like CL 9-11 or even less?
Is that an impossible thing even in theory? Or manufacturers see it easier and more cost effective to dish out DDRx+1 than make lower latency?"

Impossible in theory? Nope not at all impossible. However its likely as fast as its ever going to get latency wise, on silicone. Considering that ram really hasn't gotten any faster in 15-20 years. Sure the clocks have increased, but the CAS delay keeps going up in step. Latency has remained more or less unchanged for a long long time now.

I cant remember when pc133 sdr sdram came out, say arround 2000, maybe a year or 2 earlier. But pc133 cl2 sdram had a random seek delay of 15 nanoseconds

This record above has a random access time of 12.4 nanoseconds. This memorys stock timings put it at 8.9 nanoseconds. From a performance standpoint this is really an underclock(tho it would win in burst transfer, if its even stable, doubt it).

Burst transfer has certainly gone up ddr4 at 2500mhz would have a burst rate ~37 times faster then the pc133 ram i mentioned above; but latency has barely changed in 15 years. DDR4 is no faster then DDR2 in the latency department.

 

[neblogai]

I was also sure there is no much difference between fast and slow DDR4- but then I saw what happens if your other components are fast enough. It turns out, fast memory may be important, and can add up to 50% performance extra (Fallout4):
http://www.techspot.com/article/1171-ddr4-4000-mhz-performance/

 

[alidan]

Is there any tangible gain for this? The memory is not actually any faster so latency shouldn't change. The bandwidth is likely exceeding the internal fabric bandwidth. What's a stream score gain for this? Probably not much so while you can brag, you're really just wasting power and burning up your memory.

He is actually making the memory faster, but with those timings id be surprised if he had any gain at all overall.
And to answer your main question no there is no point in this. Benchmark tests have already shown that speeds over 2100mhz have very little gain in performance even for programs that utilize the bandwidth heavily. So short of maybe a server system its all for bragging rights.

Is there any tangible gain for this? The memory is not actually any faster so latency shouldn't change. The bandwidth is likely exceeding the internal fabric bandwidth. What's a stream score gain for this? Probably not much so while you can brag, you're really just wasting power and burning up your memory.

some games do gain frames due to memory speed, i can't say how many, for the longest time i always worked on the assumption how fast does not matter so long as you have it.

one place did a comparison from i thin 1800 to 3200 mhz all with the same cpu speed, only memory changed, and the difference was tangible, something around a 60 fps spread on the worst of the games.

not every game, but enough to make memory speed something to consider going forward.

Highly doubtful on that test. Toms did a same test from I want to say 800mhz to like 2800mhz and there was less than a 10% difference from top to bottom. After 1800 less than a 3% difference in fps.
Games do not need much memory speed its capacity that mostly matters for memory in games.

it is entirely dependant on the game,
https://www.youtube.com/watch?v=Er_Fuz54U0Y
that isnt the 60fps spread, and the range isn't as big in the ram, but it does show some games benefiting 20% or more in some areas and the bench is far more trustable.

it seems more recent games go faster based on ram, and i would love to know the reason why, especially because i'm stuck on 660~ mhz ddr2 for a few more months.

 

[photonboy]

alidan,
The memory speed requirements may have gone up, but i think it's probably still about the same if looking at the relative memory bandwidth vs CPU processing power.

Keep in mind DDR4 and DDR3 results are not comparable.

Also, FALLOUT 4 is a new game, but its an old, inefficient code path.

Memory bandwidth again should scale roughly with the amount of CPU processing you are using at a GIVEN MOMENT in time. If you max out all eight threads of the i7-6700K using HANDBRAKE then you see a big difference.

Conversely, if the game is using 25% of your CPU's processing power then there's less communication with the system memory, thus it doesn't need to be as fast.

 

[manleysteele]

For those who care.

4166 8GByte 19-26-26-46 $489.99 1.4V
4133 16GByte 19-21-21-41 $219.99 1.35V
3600 16GByte 15-15-15-35 $184.99 1.35V
3200 32GByte 14-14-14-34 $204.99 1.35V

How many of the 4266MHz kits do you want?

(These are all GSkill TridentZ kits. There are plenty of others available. I had to make a decision and decided to keep the comparison as small as possible.)

 

[alidan]

alidan,
The memory speed requirements may have gone up, but i think it's probably still about the same if looking at the relative memory bandwidth vs CPU processing power.

Keep in mind DDR4 and DDR3 results are not comparable.

Also, FALLOUT 4 is a new game, but its an old, inefficient code path.

Memory bandwidth again should scale roughly with the amount of CPU processing you are using at a GIVEN MOMENT in time. If you max out all eight threads of the i7-6700K using HANDBRAKE then you see a big difference.

Conversely, if the game is using 25% of your CPU's processing power then there's less communication with the system memory, thus it doesn't need to be as fast.

for me fallout 4 is a worst case scenario game.

a game built on gambero, which still has the quarks you find in morrowind, thats been expanded WAY passed the point of make a new damn engine already because this one is screaming kill me please, while im on a 280x factory oc playing a game with gameworks code paths.

apparently the 1.3 update redid those paths, and really increased performance across the board for amd but still.