Kingston and MSI Set Memory OC Record of DDR5-12196

[btarunr]

Kovan Yang set a new memory frequency record of DDR5-12196 (6097.6 MHz). This was possible with an Intel Core Ultra 7 265K with 2P+0E configuration running at a low 400 MHz, an MSI MEG Z890 Unify-X motherboard, and a single Kingston FURY Renegade 24 GB CKD module. The sub-timings achieved were 48-120-120-127-2. Not much else is known about the feat, such as the cooling or DRAM voltages used. The 5°C CPU temperature reading in the CPU-Z validation page suggests that an extreme cooling solution is being used. The Core Ultra "Arrow Lake-S" processor and Z890 platform, when paired with CUDIMMs (DDR5 DIMMs with client clock drivers) yield extreme memory overclocking capabilities, with several initial memory OC records breaching the 12000 MT/s mark. The HWBOT page can be accessed here, and the CPU-Z Validation page here.



View at TechPowerUp Main Site

 

[Gigaherz]

Great result. Guy knows what he is doing.
Wish he would also OC in private. Looks like he only does sponsored runs.

Btw the HWBot sub states LN2 was used both for cpu and mem (to solve the cooling mystery)

 

[codex5600x]

for wath if the CPU run only at 400 mgh

 

[Woomack]

Now wait 2-3 days so a sponsored guy makes 5MHz more on G.Skill RAM ... we have already seen 4 or 5 press releases with world records in the last two weeks, and each one was a few MHz better.
Nowadays, Hwbot is all about sponsored overclockers and marketing. This is why so many people left it for good, including me.

 

[Gigaherz]

Hwbot is all about sponsored overclockers and marketing. This is why so many people left it for good, including me.

That is just not true. I wish there was more sponsoring for more fancy events ngl.
PS: Overlocking is always about "the last 5Mhz".

Also, wow! Impressive resume you got there.

 

[lexluthermiester]

While this "record" is interesting, it's not at all impressive. They heavily gimped the CPU and ran at only single channel mode? They might as well be cheating..

 

[maxfly]

CUDIMM the new shiny.

 

[Woomack]

That is just not true. I wish there was more sponsoring for more fancy events ngl.
PS: Overlocking is always about "the last 5Mhz"

Read sponsoring as the top of the list and the most active overclockers are mainly those sponsored or people who work for leading hardware brands. I don't want to argue, but I was very active there for many years, top 50 global/hardware, etc., and a moderator for two years. I saw how many people left and for what reasons, and I saw how many teams went inactive.
Most competitions were always like "buy a $400 motherboard and risk killing it for a chance to win a $300 motherboard". Most who won were sponsored guys with unlimited LN2. I won some stuff in my time, too.

 

[lexluthermiester]

CUDIMM the new shiny.

True! Testing is showing some improvements, but refinements and tweaks are still needed to optimize things.

 

[bonehead123]

"I feel the need...the need, for SPEEEEED" - Maverick & Goose

'nuff said !

 

[Wirko]

It's surprising that he didn't have to loosen the CAS latency to 60 (10 ns) or so.

 

[AnarchoPrimitiv]

I think overclocking would be more interesting if they actually used these frequencies to perform some real world task.....to better relate the extremity of it to regular people.

 

[Woomack]

I still had no time to perform a proper comparison, but 8800+ is around where you have to switch from Gear 2 to Gear 4, and in AIDA64, it's already ~2-5GB/s bandwidth and 10-12ns latency worse, so to match the same latency, you have to pass ~9400 G4 or push sub-timings. It seems like the maximum reasonable can be ~8800 G2 and tight timings as the bandwidth is not scaling well past 8400, so above some point, you mainly fight for latency.