News G.Skill introduces CL28 DDR5-6000 for AMD Ryzen 9000 CPUs — world's tightest timings for a DDR5-6000 memory kit

Was it ever commonplace for RAM kits to NOT have XMP profiles? G.skill definitely seems to be snubbing Intel w/this move.
Nobody buying premium memory for an Intel platform should be buying anything that slow. That being said G.Skill has been doing separate XMP/EXPO for whatever reason so it's common for them. This is part of the reason that checking a QVL can be so important as many boards can support both XMP and EXPO, but I wouldn't get XMP for AMD unless the kit was on the QVL and the same with EXPO on Intel.
 
Nobody buying premium memory for an Intel platform should be buying anything that slow. That being said G.Skill has been doing separate XMP/EXPO for whatever reason so it's common for them. This is part of the reason that checking a QVL can be so important as many boards can support both XMP and EXPO, but I wouldn't get XMP for AMD unless the kit was on the QVL and the same with EXPO on Intel.
Actually for RPL (which survived degradation), for those who need 64GB+ for video/photo editing, for quite a long time DDR5 6000 (2R) 6000 is the best you can get, now I think I see some 6400 cl32. But 2 years ago the best was 6000 cl32, I managed to clock mine in Z690 cl30-36-36-72 1T.
 
  • Like
Reactions: bit_user
Actually for RPL (which survived degradation), for those who need 64GB+ for video/photo editing, for quite a long time DDR5 6000 (2R) 6000 is the best you can get, now I think I see some 6400 cl32. But 2 years ago the best was 6000 cl32, I managed to clock mine in Z690 cl30-36-36-72 1T.
Yeah that's not performance memory which is what this article is referring to. You've also been able to get 10ns 6800 32GB and 48GB modules for around a year and a half so I'm not sure what your point is here. That's all going to be limited by 2R and won't be changed until 32Gb memory IC is mass produced.
 
Yeah that's not performance memory which is what this article is referring to. You've also been able to get 10ns 6800 32GB and 48GB modules for around a year and a half so I'm not sure what your point is here. That's all going to be limited by 2R and won't be changed until 32Gb memory IC is mass produced.
hum maybe I am not speaking clearly, I mean when one wanting 64GB or above yet still wants high performance memory DDR5 6000 with low cl could actually make sense for Intel platform, just not for gaming or benchmarking
 
  • Like
Reactions: jp7189 and Makaveli
hum maybe I am not speaking clearly, I mean when one wanting 64GB or above yet still wants high performance memory DDR5 6000 with low cl could actually make sense for Intel platform, just not for gaming or benchmarking
It still doesn't unless you're trying to save money (which is certainly a valid reason) because 6800 CL34 (you can get some CL32 as. well) has the same latency as 6000 CL30. I think 48GB modules might tweak a little better than 32GB, but 2R definitely hurts performance a lot with DDR5.

Hopefully this year will change things for 32GB modules because 32Gb IC should mean both higher speeds and lower latencies available. JEDEC allows for up to 64Gb DDR5 IC, but I haven't seen anything above 32Gb from any roadmaps/announcements.
 
Last edited:
When I read about these low-latency DDR5 modules, I'm always suspicious one way they're reducing latency is by reducing the amount of on-die ECC, if not completely disabling it. That might be fine for gaming, but if the memory's stability (i.e. error rate) is compromised, then I think we should know that, before trying to use it for everyday stuff.

I really wish JEDEC hadn't left the on-die ECC thing as a hidden implementation detail, but instead had mandated an I2C protocol for querying it. That way, we could see how much ECC a DIMM has and look at the error rates to get a sense of how close to the margins a DIMM is being pushed.
 
When I read about these low-latency DDR5 modules, I'm always suspicious one way they're reducing latency is by reducing the amount of on-die ECC, if not completely disabling it.
Just a guess but I'm assuming just voltage since I've seen 6000 CL30 as low as 1.35v and CL28 1.4v, but those are a lot higher than JEDEC. A lot of high speed goes to 1.45v so that might be the situation here.

That of course doesn't guarantee the error rate wouldn't go up.
 
Last edited:
It still doesn't unless you're trying to save money (which is certainly a valid reason) because 6800 CL34 (you can get some CL32 as. well) has the same latency as 6000 CL30. I think 48GB modules might tweak a little better than 32GB, but 2R definitely hurts performance a lot with DDR5.

Hopefully this year will change things for 32GB modules because 32Gb IC should mean both higher speeds and lower latencies available. JEDEC allows for up to 64Gb DDR5 IC, but I haven't seen anything above 32Gb from any roadmaps/announcements.
But 6800 cl34 64GB pack still isn't available AFAIK, for rigs used for both gaming and editing, 64GB is kind of much preferred than 32GB or even 48GB kits, a while ago 6000 c32 was the best available in 64GB+, while now I think I saw corsair donminator titanium have 6600 and Gskill have 6400 kits C32. ATM the 6000 low latency ones are still quite preferred for non gaming only rigs which wants fast ram in intel setup.

When I read about these low-latency DDR5 modules, I'm always suspicious one way they're reducing latency is by reducing the amount of on-die ECC, if not completely disabling it. That might be fine for gaming, but if the memory's stability (i.e. error rate) is compromised, then I think we should know that, before trying to use it for everyday stuff.

I really wish JEDEC hadn't left the on-die ECC thing as a hidden implementation detail, but instead had mandated an I2C protocol for querying it. That way, we could see how much ECC a DIMM has and look at the error rates to get a sense of how close to the margins a DIMM is being pushed.
Not sure but for some reasons, especially for 4 dimm boards using 2 dimms in intel, I found lowering the timings much easier for clocking higher to pass memtest pro 4.0+OCCT memory test 1hr. maybe due to the blank slots interference stuffs
 
But 6800 cl34 64GB pack still isn't available AFAIK, for rigs used for both gaming and editing, 64GB is kind of much preferred than 32GB or even 48GB kits, a while ago 6000 c32 was the best available in 64GB+, while now I think I saw corsair donminator titanium have 6600 and Gskill have 6400 kits C32. ATM the 6000 low latency ones are still quite preferred for non gaming only rigs which wants fast ram in intel setup.
I mean it's not hard to find this stuff... perhaps where you are they aren't available?

https://www.newegg.com/g-skill-64gb-ddr5-6800/p/N82E16820374503
https://www.newegg.com/g-skill-96gb-ddr5-6800/p/N82E16820374465
 
well yea, maybe, coz I only search the local market and have been a while (I think 1month or so when I was looking around for friends) since I search for ram kits, back then the fastest available locally is like 6400 C32, and I really didn't check before commenting on this.
 
  • Like
Reactions: thestryker
But back on topic, for basically all system builds I've ever done, ram is the most easily over priced/paid part in the system, normally the wall of diminishing return got hit really quickly, personally I opt for higher capacity for both gaming or rendering usage and get the kits with XMP/expo with reasonably high spec and latency yet not hitting the skyhigh TOTL tax. So either the CL28 kits here or the DDR5 8000 32GB kits for Intel is kind of TOTL tax IMO.
 
But back on topic, for basically all system builds I've ever done, ram is the most easily over priced/paid part in the system, normally the wall of diminishing return got hit really quickly, personally I opt for higher capacity for both gaming or rendering usage and get the kits with XMP/expo with reasonably high spec and latency yet not hitting the skyhigh TOTL tax. So either the CL28 kits here or the DDR5 8000 32GB kits for Intel is kind of TOTL tax IMO.
Oh agreed completely it's very rarely worth the price of the highest end DRAM unless you have a specific use for it. This is especially true on 1R memory so long as you know what the memory IC is. When 7200 kits were selling for a very high cost Team Group was selling some relatively slow (5200 or 5600) DDR5, but it was guaranteed to by Hynix A die so if you were willing to manually tune you could get the same speed/latency as the far more expensive kits. The same is still true if you know what you're getting, or I suppose are willing to return kits if they're not the IC you want.

I eventually plan on getting a CUDIMM kit for my system, but certainly not at the current prices. When I was getting components I want to say it was ~$180 for the 7200 CL34 kit I got compared to ~$350 for 8800 CL42. While I understand the CUDIMMs are better binned memory kits period that still doesn't justify the cost for me so I'll wait until I'm putting together another DDR5 system of some sort and reassess then.
 
  • Like
Reactions: bit_user
Was it ever commonplace for RAM kits to NOT have XMP profiles? G.skill definitely seems to be snubbing Intel w/this move.
"Snubbing intel...."?

Are you serious? How anybody could ever consider Intel to be "snubbed" with respect to OEMs is beyond me.... Intel, thanks largely to their "Joint Development Program" where they literally just hand cash to OEMs to keep AMD hardware out of the most popular laptop and desktop models (Dell being one of the worst example), has used their money and power to "snub" AMD at every turn, going so far as being convicted of bribing OEMs to specifically keep AMD hardware out of their products in several different jurisdictions.

Just compare the number of motherboard models Intel gets versus AMD, and until X870, it was obvious that not only did OEMs provide more Intel options, but that the Intel options generally look to be of higher quality (just compare NZXT's X870 board to the Z890 board, it's obvious the Intel one received more "attention" and there are numerous examples across the board)...heck, Intel has their own form of DDR5 (cudimm) that doesn't even work on AMD platforms and somehow Intel is "snubbed"...?????
 
I thought you guys were talking about 64GB per dimm.. but since you're not let me gripe about it now... I wish I could pack 128GB in a single system and still get decent speed/latency. It's a toss up for which sucks worse.. 2x64GB or 4x32GB as most boards have to underclock pretty far to get stable at 2dpc.

On my last upgrade I split the difference with a g skill 2x48 6400/cl32 kit.
 
  • Like
Reactions: bit_user
I thought you guys were talking about 64GB per dimm.. but since you're not let me gripe about it now... I wish I could pack 128GB in a single system and still get decent speed/latency. It's a toss up for which sucks worse.. 2x64GB or 4x32GB as most boards have to underclock pretty far to get stable at 2dpc.

On my last upgrade I split the difference with a g skill 2x48 6400/cl32 kit.
64GB DIMMs are another thing which will come with 32Gb memory IC. Intel tends to do a little better with 2DPC on client platforms than AMD, but that's not really saying much as it still tanks compared to 1DPC. The only way to get real capacity with good speed and latency is workstation platforms.
 
Intel has their own form of DDR5 (cudimm) that doesn't even work on AMD platforms and somehow Intel is "snubbed"...?????
CUDIMM is a JEDEC standard. The fact that it's not supported by Zen 5 CPUs is probably due to AMD reusing the IO Die from Zen 4 and no other reason. Whenever AMD decides to update their IO Die, expect their CPUs to start supporting CUDMMs as well.

Maybe you got CUDIMMs confused with MRDIMMs? Here's where Intel partnered with a DRAM maker to release a spec and products ahead of JEDEC standardization. When the standard finally arrives, there's some risk that it will orphan the Intel version. For the time being, Intel got a nice performance boost and I doubt it's very common practice for most customers of server systems to try and carry forward RAM from their old machines. So, it's less of an issue if that DRAM spec is tied to Intel's 6th gen than if they did something like this in client computing products.
 
  • Like
Reactions: thestryker
Would CUDIMM help, or do you think that limit is mainly about power?
I'm not totally sure whether it's something with 2R or if it's power. With the DRAM kits I listed above the G.Skill are 1.35v which isn't bad for a 6800 XMP kit in general, but the Corsair is 1.45v and it has lower latency than the G.Skill. I'm also somewhat curious about whether or not cooling could be involved as well. TPU just published a DDR5 temperature test and it showed some relatively aggressive failure temperature for low latency.

The thing I'd really love to see tested with CUDIMMs is 2DPC. I wonder if the CKD has any impact on how high the memory clocks can get and this should be pretty simple to test.

While reading through some OC forums when looking for new ARL firmware results I did see several people disabling the CKD when overclocking memory. The current CUDIMM kits are better bins than most UDIMM kits and sometimes that helps. This leads me to believe the ARL memory controller is just a lot better than anything else on the market. It also makes me wonder how quickly we'll see improvements to the CKD.
 
  • Like
Reactions: 80251 and bit_user