G.Skill Showcases DDR5-8400 CL44 128GB (2x 64GB) Overclocked Memory

[btarunr]

G.SKILL International Enterprise Co., Ltd., the world's leading brand of performance overclock memory and PC components, is showcasing large-capacity overclocked memory at DDR5-8400 128 GB (2x 64 GB) on the ASUS ROG Crosshair X870E APEX motherboard and AMD Ryzen 9 9950X3D desktop processor. This memory overclock milestone redefines what's possible for high-performance computing with the newly announced G.SKILL DDR5 64 GB memory module, built with high-performance SK hynix DDR5 ICs. With 128 GB of overclocked high-speed memory capacity, this DDR5-8400 memory would be able to empower content creators, researchers, and professionals with more performance and capacity to complete computing tasks faster and more efficiently.

This extreme overclock achievement showcases G.SKILL's continued advancement in pushing DDR5 memory to new limits. While high-frequency memory kits have traditionally been limited to lower module capacities, this latest demonstration demonstrates that enthusiasts and professionals can now harness both blazing-fast speeds and massive memory capacity in a single kit. It provides an ideal choice for AI model development, high-resolution content creation, machine learning, and other professional workloads that demand both large memory pools and rapid memory data speeds.



The screenshot below shows the DDR5-8400 CL44 128 GB (2x 64 GB) extreme speed being tested with MemtestPro on the ASUS ROG Crosshair X870E APEX motherboard and AMD Ryzen 9 9950X3D desktop processor.

View at TechPowerUp Main Site

 

[Dr_b_]

those timings aren't that bad considering the speed, but is that really worth the premium to run them on AMD @ 8K+ MTs? Would like to see a 2x64 kit @ 6000 with ~CL30 that runs on normal boards

 

[bonehead123]

"I feel the need.....the need for SPEEEEEEED !" -Maverick/Goose

 

[rv8000]

those timings aren't that bad considering the speed, but is that really worth the premium to run them on AMD @ 8K+ MTs? Would like to see a 2x64 kit @ 6000 with ~CL30 that runs on normal boards

It’s probably something the realm of 30 gb/s read/write higher than 6000 c30. If you’re actually doing something that require memory bandwidth and you have a dual CCD zen chip, yes.

My 8000 c36 setup is as fast or faster than my 6400 c28 setup depending on the use for an extra $40 (memory cost only).

8400 is slightly more unique, yea, but there are benefits to be had.

 

[HBSound]

Is there an expected release?

 

[igormp]

It’s probably something the realm of 30 gb/s read/write higher than 6000 c30. If you’re actually doing something that require memory bandwidth and you have a dual CCD zen chip, yes.

In theory that's the bandwidth increase you'd have (2400*128*8000=38.4GB/s), but is Ryzen able to actually make use of that extra bandwidth given the IOD's write/read throughput limitation?
6000MHz alone in theory should be able to cross the 100GB/s mark, but I've only seen a couple cases of heavily overclocked Ryzen CPUs with a really fine fabric tune and 8000MHz+ memory clocks to even reach that mark.
More often than not seems like Ryzen is stuck at the 60~90GB/s mark due to interconnectivity bottlenecks, so I'm not sure if such faster memories would be relevant unless you really want to dig into OC'ing stuff.

 

[rv8000]

In theory that's the bandwidth increase you'd have (2400*128*8000=38.4GB/s), but is Ryzen able to actually make use of that extra bandwidth given the IOD's write/read throughput limitation?
6000MHz alone in theory should be able to cross the 100GB/s mark, but I've only seen a couple cases of heavily overclocked Ryzen CPUs with a really fine fabric tune and 8000MHz+ memory clocks to even reach that mark.
More often than not seems like Ryzen is stuck at the 60~90GB/s mark due to interconnectivity bottlenecks, so I'm not sure if such faster memories would be relevant unless you really want to dig into OC'ing stuff.

Yea, a Zen 5 user with an apex isn’t touching fclk or memory .

6400/2133 fclk is simple to hit and youre already hitting 100 gb/s read write then. Anyone using this kit on an apex or board capable of 8200+ will be tuning fclk and actually be capable of higher bandwidth.

 

[igormp]

Yea, a Zen 5 user with an apex isn’t touching fclk or memory .

Fair enough, but are those users also interested in the extra memory bandwidth? I usually see most people focusing on such OC's/setups for games, not bandwidth sensitive tasks.
And for mosts bandwidth-sensitive tasks, I personally haven't met anyone that goes that hard with only a 2 DIMM motherboard, since they also often need capacity, and thus 4 DIMMs which makes clocks go to shit.

6400/2133 fclk is simple to hit and youre already hitting 100 gb/s read write then.

Really? Would you have any easily available example? Most AIDA64 benchs I've seen with people doing some really tight 6000~6400 tunes manage to get close to 100GB/s, but actually never reach this value. I guess max I've seen on overclock.net was like 6600MHz at 98GB/s.
And even the 8000+ setups usually stay in the 100~110GB/s range, I've never seen more than that, even in the AIDA64 thread here:

Thread 'Share your AIDA 64 cache and memory benchmark here'

Jun 27, 2013

Hey folks, how about sharing some cache and memory benchmark you get using AIDA 64?

Rules: Use the latest stable(non alpha/beta) version of AIDA(currently 4.50.3000)

So here goes mine.

All the best.

Table as requested. In order of entry.... Only the best scores of an individual are taken. To check the overclock, refer to the screenshots below please.

To compare as per your needs, click on the respective field on the top and it will be sorted in ascending/descending order.

User

CPU model

...

• itsakjt
• Replies: 3,096
• Forum: Benchmarking

• 

Anyhow, bringing back the original point of yours:

It’s probably something the realm of 30 gb/s read/write higher than 6000 c30.

My point is just that with even 6000MHz one should in theory reach 96GB/s. 8400MHz should be allowing for ~134GB/s, but I don't think any Ryzen owner will be achieving this in any way, and is more likely to get stuck at the 100~105GB/s range, which is a marginal improvement from the theoretical speeds that even 6000MHz can bring, so...

If you’re actually doing something that require memory bandwidth and you have a dual CCD zen chip, yes.

No actual extra bandwidth, which makes this a bit moot.

Anyone using this kit on an apex or board capable of 8200+ will be tuning fclk and actually be capable of higher bandwidth.

Would you mind sharing any case of a Ryzen CPU crossing the 120GB/s mark? I'm really curious to see something like that.

 

[rv8000]

Fair enough, but are those users also interested in the extra memory bandwidth? I usually see most people focusing on such OC's/setups for games, not bandwidth sensitive tasks.
And for mosts bandwidth-sensitive tasks, I personally haven't met anyone that goes that hard with only a 2 DIMM motherboard, since they also often need capacity, and thus 4 DIMMs which makes clocks go to shit.


Really? Would you have any easily available example? Most AIDA64 benchs I've seen with people doing some really tight 6000~6400 tunes manage to get close to 100GB/s, but actually never reach this value. I guess max I've seen on overclock.net was like 6600MHz at 98GB/s.
And even the 8000+ setups usually stay in the 100~110GB/s range, I've never seen more than that, even in the AIDA64 thread here:

Thread 'Share your AIDA 64 cache and memory benchmark here'

Jun 27, 2013

Hey folks, how about sharing some cache and memory benchmark you get using AIDA 64?

Rules: Use the latest stable(non alpha/beta) version of AIDA(currently 4.50.3000)

So here goes mine.

All the best.

Table as requested. In order of entry.... Only the best scores of an individual are taken. To check the overclock, refer to the screenshots below please.

To compare as per your needs, click on the respective field on the top and it will be sorted in ascending/descending order.

User

CPU model

...

• itsakjt
• Replies: 3,096
• Forum: Benchmarking

• 

Anyhow, bringing back the original point of yours:

My point is just that with even 6000MHz one should in theory reach 96GB/s. 8400MHz should be allowing for ~134GB/s, but I don't think any Ryzen owner will be achieving this in any way, and is more likely to get stuck at the 100~105GB/s range, which is a marginal improvement from the theoretical speeds that even 6000MHz can bring, so...

No actual extra bandwidth, which makes this a bit moot.


Would you mind sharing any case of a Ryzen CPU crossing the 120GB/s mark? I'm really curious to see something like that.

Some random dual CCD 8000 result randomly pulled from the memory thread

https://www.overclock.net/posts/29421268/

Vs

Some random dual CCD 6200 result

AMD DDR5 OC And 24/7 Daily Memory Stability Thread

https://www.amd.com/en/technologies/expo Soon to be released EXPO AMD optimized DDR5 memory. Likely as with Intel memory, A-die will be preferred. M-die is a close second. It has been said that like 3800-4000 1-1 the sweet spot for AMD 5000 series CPUs, 6000 1-1 should be the sweet spot...

www.overclock.net

Chew and other users have posted significantly higher results closer to/at 120gb/s if memory serves correctly, but I’m not gonna dig through thousands of posts on my phone.

You’ve got some weird hangups, there is absolutely performance to be had, 8000 isn’t some unobtainable thing anymore either.

 

[igormp]

Some random dual CCD 8000 result randomly pulled from the memory thread

https://www.overclock.net/posts/29421268/

Vs

Some random dual CCD 6200 result

AMD DDR5 OC And 24/7 Daily Memory Stability Thread

https://www.amd.com/en/technologies/expo Soon to be released EXPO AMD optimized DDR5 memory. Likely as with Intel memory, A-die will be preferred. M-die is a close second. It has been said that like 3800-4000 1-1 the sweet spot for AMD 5000 series CPUs, 6000 1-1 should be the sweet spot...

www.overclock.net

Chew and other users have posted significantly higher results closer to/at 120gb/s if memory serves correctly, but I’m not gonna dig through thousands of posts on my phone.

Those 2 results point to what I had mentioned, reads for 6200MHz at 92GB/s, and 8000MHz at 103GB/s, both far from the theoretical values, specially the 8000MHz one.
Along with the fact that a 1800MHz increase should translate to almost a 28.8GB/s bump, not 11GB/s. Ryzen's memory scaling is really bad due to the IOD bottleneck.

Chew and other users have posted significantly higher results closer to/at 120gb/s if memory serves correctly, but I’m not gonna dig through thousands of posts on my phone.

Gave a quick look at chew's posts, he doesn't seem to do many bandwidth benchmarks, but still their best result seems to be a 106GB/s reads on 8800MHz, which only serves to show that IOD-imposed ceiling on bw.

 

[rv8000]

Those 2 results point to what I had mentioned, reads for 6200MHz at 92GB/s, and 8000MHz at 103GB/s, both far from the theoretical values, specially the 8000MHz one.
Along with the fact that a 1800MHz increase should translate to almost a 28.8GB/s bump, not 11GB/s. Ryzen's memory scaling is really bad due to the IOD bottleneck.


Gave a quick look at chew's posts, he doesn't seem to do many bandwidth benchmarks, but still their best result seems to be a 106GB/s reads on 8800MHz, which only serves to show that IOD-imposed ceiling on bw.

You’re missing the point. Going from 6000c30 to 8000c36 on a dual CCD chip you're gaining ~20 gb/s across r/w/c. Stop throwing a bunch of shit at the wall trying to obfuscate that there are clearly significant bandwidth benefits. If you don’t wanna spend the money or time, thats on you, and a different story.

 

[igormp]

Going from 6000c30 to 8000c36 on a dual CCD chip you're gaining ~20 gb/s across r/w/c.

Afaik, 103-92 = 11GB/s for the read numbers, which are the most relevant.
Let's not even dive into how bad AIDA64 is for such benchmarks, (mlc ftw), but you shouldn't be cherry-picking the write numbers to try to prove your point, given how unrealistic those are within AIDA64.

Stop throwing a bunch of shit at the wall trying to obfuscate

You're the one that seem to be throwing a fit and not backing your claims, after I asked for such because I was really wondering if it'd be possible, and it seems like not, from what you're showing so far.

there are clearly significant marginal bandwidth benefits

FTFY

If you don’t wanna spend the money or time, thats on you, and a different story.

I don't see how that's relevant to the topic, nor how your assumptions make any sense.

Once again, the claim that such modules are relevant from a bandwidth perspective doesn't seem to hold due to Ryzen's IOD limitations, but such 8400cl44 config may bring benefit in other scenarios still.

 

[rv8000]

Afaik, 103-92 = 11GB/s for the read numbers, which are the most relevant.
Let's not even dive into how bad AIDA64 is for such benchmarks, (mlc ftw), but you shouldn't be cherry-picking the write numbers to try to prove your point, given how unrealistic those are within AIDA64.

You're the one that seem to be throwing a fit and not backing your claims, after I asked for such because I was really wondering if it'd be possible, and it seems like not, from what you're showing so far.

FTFY

I don't see how that's relevant to the topic, nor how your assumptions make any sense.

Once again, the claim that such modules are relevant from a bandwidth perspective doesn't seem to hold due to Ryzen's IOD limitations, but such 8400cl44 config may bring benefit in other scenarios still.

In what world is increasing r/w/c by 12%/17%/19% respectively… marginal? And thats skipping over the fact those numbers are from 6200 vs 8000, not 6000 vs 8000 (as was explicitly stated).

I’d call marginal 3-5%, if that.

Again, exhaustingly… there are significant benefits to be had. If you want a 6000 c30 kit, you can go buy one. I’m otherwise unsure why you’re in this thread complaining about arch limits which no one brought up but yourself, and wasn’t being discussed in the original reply.

 

[igormp]

In what world is increasing r/w/c by 12%/17%/19% respectively… marginal?

Well, I do consider 12% to be marginal in comparison to the theoretical 29% it should've been. But that's indeed a me thing.
As I had stated, the other numbers from aida64 should not really be considered (specially because writes higher than reads don't make much sense to begin with).

If you want a 6000 c30 kit, you can go buy one.

I do have a 6400 one, that sadly couldn't manage to bring past 4800MHz, but it is what it is haha
I may give it a go at 5200MHz this weekend, I'd be happy if I could manage anything past 60GB/s, which would be around 70% of the theoretical bandwidth, on level with Zen 4 (and better than what Zen 5 usually does, as per chipsandcheese).

I’m otherwise unsure why you’re in this thread complaining about arch limits which no one brought up but yourself, and wasn’t being discussed in the original reply.

Because you claimed that:

It’s probably something the realm of 30 gb/s read/write higher than 6000 c30.

Which I genuinely wondered if it was achievable nowadays on Ryzen CPU, since I'd be really interested at this, specially at higher capacity setup such as the ones in the OP, but seems like it's not and the IOD bottlenecks are still withstanding, so I ended up finding out that your claim wasn't really a valid.

 

[Jnt MÉTODO$]

English please! Next post will be deleted if not in English. Thank you.

When will they be available for purchase?

Quando vão estar disponíveis para compra?

 

[Prima.Vera]

I never understood why does one need more than 32GB of RAM nowadays.
I always thought if you need more, just buy a server and run your VMs or your 3D design software and apps. Or simmilar.