ROG Maximus Z890 Apex Achieves Record-Breaking Overclocking Performance

[TheLostSwede]

ASUS Republic of Gamers (ROG) today announced that new ROG Maximus Z890 Apex motherboards have been used to achieve 5 world records, 19 global first-place records and 31 first-place records. In the hands of some of the world's premier professional overclockers, the Maximus Z890 Apex has coaxed dazzling performance out of the latest Intel Core Ultra processor (Series 2) lineup and the latest high-performance memory kits.

Veterans of the overclocking scene will not be surprised to learn that these records were achieved with an Apex motherboard on the bench. This series has an undeniable pedigree. Since the very first model, ASUS has designed Apex motherboards for the singular purpose of helping the world's most talented overclockers shatter barriers on their way to new records.




The ROG Maximus Z890 Apex takes the series to new heights with a sizeable 22+1+2+2 power solution ready to take Intel Arrow Lake processors to the stratosphere. But the true strength of the Apex is in its memory layout. In order to fully optimize copper trace pathways to the integrated memory controller, the Apex features two DIMM slots instead of four, sacrificing raw capacity to enable record-breaking frequencies.

The outstanding records
The innovative layout of the ROG Maximus Z890 Apex is proving itself on the overclocking battlefield. The memory frequency achieved by BenchMarc deserves special recognition. Using the Intel Core Ultra 9 processor 285K and the ROG Maximus Z890 Apex, BenchMarc was able to push a memory kit all the way to 12,066 MT/s for a new world record.

Famed overclocker Elmor took an Intel Core Ultra 9 processor 285K to dizzying heights with the ROG Maximus Z890 Apex. Using an innovative overclocking setup, Elmor pushed this CPU to 7488.8 MHz. The overclocking was conducted with liquid helium using the AI LN2 pot, a project by 3D Systems, Diabatix, ElmorLabs, and SkatterBencher. This LN2 pot was designed with generative AI technology from Diabatix that allowed the team to rapidly explore a stunning range ofa design alternatives. The output of the Gen AI process was unlike anything else on the market, and it required the cutting-edge 3D printing tech of 3D Systems to bring to life; but the results speak for themselves.

As impressive as these results are, they are only a small selection of the overclocking records achieved with the ROG Maximus Z890 Apex and Intel Core Ultra processor (Series 2) CPUs. Take a look:



Congratulations to Elmor, BenchMarc, OGS, and CENS for these impressive results.

When two is greater than four
While most users will not be clamping a liquid nitrogen pot to their new Intel Core Ultra processors (Series 2), there are some lessons that PC enthusiasts can learn from these professional overclockers when it comes to extracting the most performance out of their new system.

One important insight concerns memory gear modes. In the past, a processor's memory controller operated at a frequency equal to memory speed. We call this 1:1 ratio Gear 1. But as DDR4 modules were developed with increasingly higher frequencies, a new approach was needed to enable the higher memory-module data rates. That is when Gear 2 was introduced-a mode in which the processor memory controller operates at half the memory speed. This approach enhances compatibility and provides greater potential for pushing data rates to higher levels.

Today's DDR5 memory kits are so fast that Gear 1 is no longer an option. With the latest Intel Core Ultra processors (Series 2), users have the choice between Gear 2 and Gear 4. Gear 4 operates the CPU memory controller at a quarter of the memory clock speed, and it is the default setting for most Z890 motherboards, thanks to its excellent compatibility.

However, Gear 4 comes with a tradeoff in the form of higher memory latency. In this regard, Gear 2 is demonstrably better, and that is especially true in the 6400-9000 MHz frequency range. For example, the Intel Core Ultra 9 processor 285K and DDR5-8000 have 18.8% lower latency when switched from Gear 4 to Gear 2, according to AIDA64.

Switching to Gear 2 instead of Gear 4 can even allow a memory kit to perform above its specifications. At DDR5-9000 with Gear 2, an extremely fast CUDIMM kit with the Intel Core Ultra 9 processor 285K recorded faster memory read speeds and 12.48% lower latency than with DDR5-9600 and Gear 4.

The latency difference between Gear 2 and Gear 4 is not just visible in synthetic benchmarks—users will notice it in games, too. Internal testing of Far Cry 6 and Cyberpunk 2077 with a two-DIMM memory kit delivered an FPS increase of up to 8.9% at 1080p when switching from Gear 2 to Gear 4.

Accordingly, ASUS has selected Gear 2 as the default setting for ASUS Z890 motherboards. This means that users will not have to tinker with their memory settings to enjoy the advantages of Gear 2.



Fully equipped for premium memory performance
Exciting advancements in the world of DDR5 - like new clock-unbuffered DIMM (CUDIMM) memory modules - have created considerable speculation among enthusiasts. Impressive memory performance is becoming possible even for users who are not resorting to exotic cooling methods. With the ROG Maximus Z890 Apex, this group of overclockers was able to achieve DDR5-10266 at a CAS latency of 46.

However, users are not limited to the ROG Maximus Z890 Apex motherboard when it comes to next-level memory performance. The ROG Maximus Z890 Extreme, ROG Maximus Z890 Hero and ROG Strix Z890-E Gaming WiFi all feature NitroPath DRAM technology. This cutting-edge redesign of the DRAM slot improves signal quality, increasing DRAM overclocking performance by up to 400 MT/s, depending on factors like memory speed, module and system configuration. With these motherboards, users can enjoy premium memory speeds along with the capacity upgrade options available with their four DIMM slots.

Users of select ROG motherboards will also have an easy time optimizing the latest CUDIMM memory kits. ASUS Z890 motherboards include ASUS Enhanced Memory Profile III (AEMP III). This advanced firmware feature empowers users to achieve unparalleled memory performance with cutting-edge CUDIMM memory. Through a two-phase comprehensive tuning process, AEMP III first optimizes the clock driver and then the memory frequency. This results in the best balance between performance and stability, even at cutting-edge speeds of DDR5-8000 or higheri.

Not just for professional overclockers
ASUS Z890 motherboards offer a wealth of advantages above and beyond robust support for CPU and memory overclocking. With a wide range of options across the ROG, TUF Gaming, ProArt and Prime motherboard families, users are sure to find an option that fits their style, performance needs and budget.

View at TechPowerUp Main Site | Source

 

[Ravenmaster]

Too bad the Z890 platform along with the Core Ultra 9 285K sucks a$$ when it comes to gaming. I cancelled my pre-order.

 

[bonehead123]

Speed may not kill you, but suddenly becoming stationary probably will

 

[azrael]

I do love how manufacturers are trying to upsell you motherboards with *less* DIMM slots (at a higher price, of course). I would at any given time prefer 4 DIMM slots over higher memory speed. But then again, I'm not an overclocker and my PCs are usually fast enough as they are.

 

[SOAREVERSOR]

I do love how manufacturers are trying to upsell you motherboards with *less* DIMM slots (at a higher price, of course). I would at any given time prefer 4 DIMM slots over higher memory speed. But then again, I'm not an overclocker and my PCs are usually fast enough as they are.

This is entirely for overclockers so they aren't going to use more than two slots and having less slots allows you to better design the board.

It's not for me either. I tend towards workstation boards with gobs of memory and you're not going to be able to run that amount of RAM fast.

Too bad the Z890 platform along with the Core Ultra 9 285K sucks a$$ when it comes to gaming. I cancelled my pre-order.

Gaming is jerking off. It's not productive. Most people aren't using their computers for gaming. Intel is moving in the right direction and laying the ground work for the future and this does do that very well. AMD did this with the first Ryzen and it took them a bit to sort that out. When nvidia rolled out RTX same situation but later it paid off massively in AI which was always the real goal. That these things helped in gaming was a side bonus but it wasn't the reason those things were done.

I'd much rather have this than a 9800X3D to go with my RTX 4090. If I'm going to game, I'll turn on the Switch or the PS5. List of PC games I've played in the past six months.... Doom 2, Quake 1 1v1 DM duels, Total War Warhammer III. And even then... not that much

 

[Hxx]

It gives me joy to see these overpriced boards that no one buys (except for YouTubers and few overclockers) because intel chips suck ballz lmao. Let asus waste their money and let Newegg inventory stock up as high as it can go

 

[evernessince]

This is entirely for overclockers so they aren't going to use more than two slots and having less slots allows you to better design the board.

It's not for me either. I tend towards workstation boards with gobs of memory and you're not going to be able to run that amount of RAM fast.

The act of simply removing the 2 slots itself is what provides improvements to RAM signal integrity, not improvements to topology or layout.

At higher frequencies the pins in RAM slots act as antennas that interfere with the neighboring slots. That strength of that interference increases as frequency increases.

Any vendor claiming that removing two slots allows you to better design a board is misleading you. There may be a very small nugget of truth in there but the interference problem is the major hurdle at these speeds. They are trying to sell you that they did a lot of work when in reality they did very little.

You can get the same interference reduction with any motherboard sporting ASUS's Nitro path, which simply designs the pins in the memory slot to reduce interference (not that I recommend ASUS products, only that this issue is not as complicated as many will have you believe to sell you a product).

 

[MxPhenom 216]

The act of simply removing the 2 slots itself is what provides improvements to RAM signal integrity, not improvements to topology or layout.

At higher frequencies the pins in RAM slots act as antennas that interfere with the neighboring slots. That strength of that interference increases as frequency increases.

Any vendor claiming that removing two slots allows you to better design a board is misleading you. There may be a very small nugget of truth in there but the interference problem is the major hurdle at these speeds. They are trying to sell you that they did a lot of work when in reality they did very little.

You can get the same interference reduction with any motherboard sporting ASUS's Nitro path, which simply designs the pins in the memory slot to reduce interference (not that I recommend ASUS products, only that this issue is not as complicated as many will have you believe to sell you a product).

I think what you are looking for here is reflections caused by a high speed interface. But DDR interfaces on boards are usually terminated to accommodate the varying speeds an end user might have. It also avoids any issues on the interface between DATA and CK going across. Usually cheaply implemented DDR interfaces are non-terminated.

Also reducing to 2 slots does change topology from daisy chain to DPC (It is rare to find 4 slot DPC boards). Which does improve memory OC by a non negligible amount. I think its buildzoid who has a plethora of videos about this on his youtube.

2 slots also does simplify pcb trace layout to be as optimal as possible

 

[evernessince]

I think what you are looking for here is reflections caused by a high speed interface. But DDR interfaces on boards are usually terminated to accommodate the varying speeds an end user might have. It also avoids any issues on the interface between DATA and CK going across. Usually cheaply implemented DDR interfaces are non-terminated.

1) It's EMI
2) All empty slots are non-terminated regardless of the price of the motherboard. A signal is being sent to all slots whether a module is in them or not.

Der Bauer did a video on this very topic:

You can see in the comments of that video people who point out that they use blank RAM DIMMs to get the same effect, because they are not terminated by default even on very expensive motherboards (whether that be HEDT, server, or desktop).

Also reducing to 2 slots does change topology from daisy chain to DPC (It is rare to find 4 slot DPC boards).

It's not an improvement over 4 slot board with 1 DPC (it's effectively the same). These 2 DIMM boards are being sold to the ultra-enthusiast market so it comes with the expectation that it's going to be compared against the best we could do prior.

Which does improve memory OC by a non negligible amount. I think its buildzoid who has a plethora of videos about this on his youtube.

You are assuming we are comparing a T or Daisy chain topology with 2 DPC and not a 1 DPC 4 slot board.

Mind you that's really beside the point that above a certain frequency you won't realize those gains unless you address the EMI issue.

Again these two slot boards don't do anything we haven't done before. They are selling you a 1 DPC board with 2 slots that at the same price point you could have gotten a 1 DPC 4 slot board before. They are side stepping an issue by going to DIMM that could be fixed mearly by them changing the slot pins or by providing dummy memory sticks.

2 slots also does simplify pcb trace layout to be as optimal as possible

Can we really say this is actually the case? You have to assume that modern 8 and 10 layer boards aren't enough to do things optimally in the first place.

 

[Dr. Dro]

I do love how manufacturers are trying to upsell you motherboards with *less* DIMM slots (at a higher price, of course). I would at any given time prefer 4 DIMM slots over higher memory speed. But then again, I'm not an overclocker and my PCs are usually fast enough as they are.

Speaking as someone who owns a Maximus Apex - this motherboard really isn't for everyone. However, if you're into advanced electronics, these things are a treat

 

[MxPhenom 216]

1) It's EMI
2) All empty slots are non-terminated regardless of the price of the motherboard. A signal is being sent to all slots whether a module is in them or not.

Der Bauer did a video on this very topic:

You can see in the comments of that video people who point out that they use blank RAM DIMMs to get the same effect, because they are not terminated by default even on very expensive motherboards (whether that be HEDT, server, or desktop).



It's not an improvement over 4 slot board with 1 DPC (it's effectively the same). These 2 DIMM boards are being sold to the ultra-enthusiast market so it comes with the expectation that it's going to be compared against the best we could do prior.



You are assuming we are comparing a T or Daisy chain topology with 2 DPC and not a 1 DPC 4 slot board.

Mind you that's really beside the point that above a certain frequency you won't realize those gains unless you address the EMI issue.

Again these two slot boards don't do anything we haven't done before. They are selling you a 1 DPC board with 2 slots that at the same price point you could have gotten a 1 DPC 4 slot board before. They are side stepping an issue by going to DIMM that could be fixed mearly by them changing the slot pins or by providing dummy memory sticks.



Can we really say this is actually the case? You have to assume that modern 8 and 10 layer boards aren't enough to do things optimally in the first place.

You know that reflections are a type of EMI right?

 

[Dr_b_]

At least someone is enjoying this new platform, and maybe you need to pump 1000W and pour LN2 in it/on it to make it competitive

 

[MxPhenom 216]

Speaking as someone who owns a Maximus Apex - this motherboard really isn't for everyone. However, if you're into advanced electronics, these things are a treat

I kind of want the Z890 to have some fun with but god damn that price

 

[Dr. Dro]

I kind of want the Z890 to have some fun with but god damn that price

Yeah, it stings a little. But it's a super solid board to build and tweak on, top tier and high-priority BIOS support makes it worth it. Getting 7600 out of my 6800 kit on my Z790 Apex Encore was hilariously easy with it too, like a 10 minute thing. Scored stable settings on the first try, too.

At least someone is enjoying this new platform, and maybe you need to pump 1000W and pour LN2 in it/on it to make it competitive

That seems a bit extreme. Sure it's disappointing that it's not a chart topper, but still. XOC isn't necessary to extract the best out of any architecture

 

[Tomorrow]

I would at any given time prefer 4 DIMM slots over higher memory speed. But then again, I'm not an overclocker and my PCs are usually fast enough as they are.

I tend to go the other way. Since 2x48GB is enough for me i want to maximise speed.

 

[Caring1]

If two DIMM slots is better, then iTX should rock.

 

[Squared]

I do love how manufacturers are trying to upsell you motherboards with *less* DIMM slots (at a higher price, of course). I would at any given time prefer 4 DIMM slots over higher memory speed. But then again, I'm not an overclocker and my PCs are usually fast enough as they are.

Tom's Hardware has a table listing the maximum officially-supported memory speed with every possible configuration.

Intel Core Ultra 9 285K Review: Intel Throws a Lateral with Arrow Lake

Some good, some not so good.

www.tomshardware.com

In short, 4 dual-rank modules will run at DDR5-4400. Moving to single-rank—which tends to be lower capacity—allows for DDR5-4800. Two modules of any type can reach DDR5-5600, and two CUDIMMs on a two-slot board can reach DDR5-6400.

Given this and given that 96 GB is supported with 2 modules, I would only consider Arrow Lake boards with 2 slots, because I want the best possible performance without overclocking.

 

[Dr_b_]

That seems a bit extreme. Sure it's disappointing that it's not a chart topper, but still. XOC isn't necessary to extract the best out of any architecture

It is for Arrow Lake
It's a disappointment that Arrow Lake loses to 3 generation old CPUs, and still can't beat any of them with power consumption, and it also has stability problems, this isn't just an issue with chart topping

 

[MxPhenom 216]

It is for Arrow Lake
It's a disappointment that Arrow Lake loses to 3 generation old CPUs, and still can't beat any of them with power consumption, and it also has stability problems, this isn't just an issue with chart topping

What do you mean it cant beat any with power consumption? What charts are you looking at?

Also of you are only looking at tpu charts they arent running with intels declared 8000 mhz memory sweet spot so i suspect its being held back some. Even then in non-gaming loads these new cpus are fine.

 

[Dr_b_]

What do you mean it cant beat any with power consumption? What charts are you looking at?

Also of you are only looking at tpu charts they arent running with intels declared 8000 mhz memory sweet spot so i suspect its being held back some. Even then in non-gaming loads these new cpus are fine.

GN and others did a power consumption tests, its bleak.

 

[MxPhenom 216]

GN and others did a power consumption tests, its bleak.

TPU did too and theirs is cpu only tests

This is bleak?

 

[Dr_b_]

TPU did too and theirs is cpu only tests

This is bleak?

Looking at that chart, compared to the 7800X3D, the 285K uses more than twice the power

 

[MxPhenom 216]

Looking at that chart, compared to the 7800X3D, the 285K uses more than twice the power

X3D chips are on their own. Even AMDs own non x3d chips fall behind on power largely due to clock frequencies not reach the same level and voltages being lower due to x3d limitations. Not exactly a fair comparison

 

[Hecate91]

The power consumption is underwhelming as Intel is on more advanced node, yet they aren't beating AMD on power efficiency. Even at the same clocks with non-X3D AMD is more efficient.

 

[Dr_b_]

X3D chips are on their own. Even AMDs own non x3d chips fall behind on power largely due to clock frequencies not reach the same level and voltages being lower due to x3d limitations. Not exactly a fair comparison

TPU's own graphs in the power section are a very bad look for intel, despite whether the comparison is fair, you can't really link a chart that shows the CPUs in a bad light but cherry pick which parts you can compare them to -- in the same chart -- so that they prove your point

From the TPU review:
"For gaming specifically, the Ryzen 7 7800X3D is the best choice for energy efficiency, though, using roughly half the power of the 285K while offering more FPS at the same time. You do have to consider that Intel is on a 3 nanometer process whereas AMD uses slightly less efficient 4 nanometer—both from TSMC."

• Gaming performance lower than expected, slower than Raptor Lake (LOL waaat? Intel released a "new" CPU that performs worse and far worse in some instances?)
• Some games and applications aren't currently performing well at all (Another what the...)
• Requires new motherboard with Socket LGA1851 (And this is going to be a dead end platform, gauranteed, or intel would have hyped it up, and the board partners want, nay, think, people are dumb enough to spend $500 for some of these motherboards?)
• No more support for DDR4 memory (Not sure why this is a con, DDR5 is faster, why would you want to use slower DDR4 RAM when it would be rate limiting the CPU just to save a few bucks)
• No support for AVX-512 (this one is a head scratcher, this is intel's own tech, and we have to wait until +5 years for it to get it back on desktop with AVX10, if intel is still in business)

Not sure how intel messed this up so bad.