• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.

CUDIMM in capacities over 32GB, Corsair Dominator

Thug Esquire

Thug Esquire

[H]ard|Gawd
Joined
May 4, 2005
Messages
1,543
I've seen press releases and paper launches around CUDIMMs in capacities beyond 32GB but nothing material or shipping right now. It seems impossible to buy. Is anything actually available to buy at those capacities?

Anyone know when Corsair might update their Dominator line with CUDIMMs and these bigger capacities?
 
Thug Esquire said:
I've seen press releases and paper launches around CUDIMMs in capacities beyond 32GB but nothing material or shipping right now. It seems impossible to buy. Is anything actually available to buy at those capacities?

Anyone know when Corsair might update their Dominator line with CUDIMMs and these bigger capacities?
Click to expand...
I want gskill to sell me 32 or 48GB cudimms for my next rig.
 
Ive been searching as well and cant find any info on 64gb CUDIMM kits at the moment. There are several 48GB kits available though....
 
Using a G.Skill 48GBx2 for my setup atm. May go to 192GB if the mood hits me. (Because i can :))
 
Not really understanding what they are doing, but that seem huge difference from tweaking:

285k_1080_bg3_oc.png
285k_1080_bg3_5poc.png


Like brute forcing some of the memory latency issues. Not having those tweaks part of the XMP profile making those super high priced kit look not that useful "stock", but maybe their target audience know all about all of this.
 
LukeTbk said:
Not really understanding what they are doing, but that seem huge difference from tweaking:

View attachment 717712View attachment 717713

Like brute forcing some of the memory latency issues. Not having those tweaks part of the XMP profile making those super high priced kit look not that useful "stock", but maybe their target audience know all about all of this.
Click to expand...
For the blue line, they have manually lowered the sub-timings. Techpowerup shows an image of the exact timings they achieved. Usually takes a little extra voltage, and trial and error, and some luck.

For the green line, they use the lower sub-timings from the blue line AND they also overclocked the CPU's Ring (Cache) speed, as well as the Die-to-Die (D2D) speed. Which is the connection speed for the tiles/chiplets. The memory controller is not on the same tile as the CPU cores and cache. So, overclocking the die-to-die and cache, has some pretty good benefits.

The author says what cache speed and die-to-die speed they used, in the comments after the article.
 
What I'd really like to see is some more testing with dual ranked DIMMs, particularly gaming tests. Phoronix did an Arrow Lake test with some CUDIMMs but one of the kits they were comparing to was a 2x32GB dual ranked kit, and the dual ranked kit traded blows with the high clocked CUDIMMs and beat everything single ranked with the same clock speed. The high clocked CUDIMMs were usually faster, but the dual ranked kit was usually right behind them and won a couple tests. Of course it's Phoronix so they're running Linux and not doing gaming tests.
 
zandor said:
What I'd really like to see is some more testing with dual ranked DIMMs, particularly gaming tests. Phoronix did an Arrow Lake test with some CUDIMMs but one of the kits they were comparing to was a 2x32GB dual ranked kit, and the dual ranked kit traded blows with the high clocked CUDIMMs and beat everything single ranked with the same clock speed. The high clocked CUDIMMs were usually faster, but the dual ranked kit was usually right behind them and won a couple tests. Of course it's Phoronix so they're running Linux and not doing gaming tests.
Click to expand...
I just bought this KLEVV 64GB kit to pair with my 265K. Arrived this morning from Amazon.
https://pcpartpicker.com/product/G9...32-gb-ddr5-6400-cl32-memory-kd5bgua80-64a320g

Lowest timings of any DDR5 6400 kit, and also only 1.35v. So I'm hoping for some decent headroom to tighten further and/or increase speed.
 
zandor said:
What I'd really like to see is some more testing with dual ranked DIMMs, particularly gaming tests. Phoronix did an Arrow Lake test with some CUDIMMs but one of the kits they were comparing to was a 2x32GB dual ranked kit, and the dual ranked kit traded blows with the high clocked CUDIMMs and beat everything single ranked with the same clock speed. The high clocked CUDIMMs were usually faster, but the dual ranked kit was usually right behind them and won a couple tests. Of course it's Phoronix so they're running Linux and not doing gaming tests.
Click to expand...

chameleoneel said:
I just bought this KLEVV 64GB kit to pair with my 265K. Arrived this morning from Amazon.
https://pcpartpicker.com/product/G9...32-gb-ddr5-6400-cl32-memory-kd5bgua80-64a320g

Lowest timings of any DDR5 6400 kit, and also only 1.35v. So I'm hoping for some decent headroom to tighten further and/or increase speed.
Click to expand...
aannnnnnnd its a bad set of RAM.

I swear, I have gotten so much bad hardware the past few years. Sheesh!

It was juuuuust stable enough to run Elden Ring and FF14 benchmark. Both games had very obvious issues due to the RAM. And so did Windows.

If the numbers are correct-----performance was a smidge lower than my 32GB 7200 Set.

Staring at a Wall in Elden Ring at 1080p max, no Ray Tracing:

64GB set was 157 average and 146 1% lows
32GB 7200 set was 161 average and 146 1% lows

FF14 was similar. A couple of spots I alway look at, were 6 and 4 fps lower on the 64GB set (I used my cellphone to record the runs, with Afterburner tracking the numbers. So that I could easily and simply compare the numbers at my favorite parts of the benchmark).

If I remember correctly, Aida 64 latency was 3-4ms higher on the 64GB set.


That's all I tested. I would have tested a couple more games, but my Windows install was barely hanging on, with this bad RAM. I know both games favor latency. So the numbers are in line with what I would expect. Something like Starfield may be a decent test to potentially flex the RAM bandwidth and/or dual rank bank interleaving, more.
 
Last edited:
Someone on the overclockers forums has a 9600 CUDIMM setting running at 9133 with CL38 (and some other timings tweaks). CL38 is typically seen on a good 8200 set.

Their Aida latency is 65ns. Arrowlake can be saved, after all >_>

They are running a bit over 1.5v on the RAM, though. And that's like a $400 set of RAM.

I'm hoping by summertime we see some significant price decreases on CUDIMM. Probably doesnt help that AMD didn't support it in Zen5...
 
chameleoneel said:
Someone on the overclockers forums has a 9600 CUDIMM setting running at 9133 with CL38 (and some other timings tweaks). CL38 is typically seen on a good 8200 set.

Their Aida latency is 65ns. Arrowlake can be saved, after all >_>

They are running a bit over 1.5v on the RAM, though. And that's like a $400 set of RAM.

I'm hoping by summertime we see some significant price decreases on CUDIMM. Probably doesnt help that AMD didn't support it in Zen5...
Click to expand...
Got a link handy? I'm curious to see it :).
 
You must log in or register to reply here.
Back
Top