DDR5 Memory Performance Scaling with Alder Lake Core i9-12900K

[W1zzard]

In this article, we take a closer look at the performance scaling you can expect for various DDR5 configurations. We test from DDR5-6000 all the way down to DDR5-2400 and compare CL30 vs. CL36 vs. CL40. Last but not least, we also consider these numbers in relation to what DDR4-3600 offers.

Show full review

 

[lightofhonor]

Would be interesting to see how this affects IGP performance, but looks like it's better to invest in good DDR4 at this point.

 

[W1zzard]

Would be interesting to see how this affects IGP performance, but looks like it's better to invest in good DDR4 at this point.

Heh, just today I was thinking "only testing at various TDPs left, then I can wrap up Alder Lake launch" .. guess there's more work to do

 

[cadaveca]

Heh, just today I was thinking "only testing at various TDPs left, then I can wrap up Alder Lake launch" .. guess there's more work to do

It’ll matter more with laptop iGPU, you’d think, with 96 EUs. The standard desktop parts are 24 EU only?

 

[ncrs]

In the fifth paragraph of the conclusion: "I also looked at memory timings, specifically DDR4-4800 CL30 vs CL36 vs CL40" - shouldn't this be DDR5-4800?

 

[mama]

DDR5 needs time to mature.

 

[FilipM]

Hmm, maybee give DDR4 a better change with something like a CL14 3600 setup?

Awesome review though, really nice to see the scaling

 

[HammerON]

Great review W1z!!! I appreciate you taking the time to do this

 

[dalekdukesboy]

Great review W1z!!! I appreciate you taking the time to do this

Indeed.....

Hmm, maybee give DDR4 a better change with something like a CL14 3600 setup?

Awesome review though, really nice to see the scaling

I'm currently on an x79 xeon setup I've had for ages, (but many upgrades/changes made over the years) but I am considering putting together a 12900k build and considering cost or non-existence of ddr5 at this point I'd probably just get best ddr4 board I can get. However that leads me to wondering what the best DDR OC'd and at best timings possible would do in a head to head with the various DDR5's out at the moment?


ALSO, what I'm really wishing is someone would show the bandwidth on a good set of said DDR4 on a 12900k in AIDA for read, write, latency etc....

 

[W1zzard]

In the fifth paragraph of the conclusion: "I also looked at memory timings, specifically DDR4-4800 CL30 vs CL36 vs CL40" - shouldn't this be DDR5-4800?

Fixed, thanks!

 

[Wirko]

You included memory speeds almost down to zero, which is great. However, the results are hard to interpret. The performance drop when going from DDR5-2800 to 2400 is precipitous in many games and applications. Also from 3200 to 2800 but to a lesser extent. It's not proportional (which could be explained by reduced bandwidth) but worse. I'm wondering what's going on in the IMC in such circumstances; it's almost as if it went to sleep often because of slow memory. Any ideas?

 

[swirl09]

Seeing this really does make it apparent how hard a good DDR5 kit will be to get for a while. Its not the first place that struggling to get much more beyond the 6000 CL36 mark.

I hope when I get some time I'll see what is needed to get me to Gear1 and if it benefits much. Im sure it'll be better, but if its actually worth the trouble is another story. Either way, DDR4 was the right call for now.

 

[nikoya]

thank you for this excellent study !

 

[mrthanhnguyen]

I have seen ddr5 6600cl28 1t and ddr4 430013 1t . Not sure all sample can run at those speed but they are very fast at the moment.

 

[mechtech]

Started reading review and all the data and thought to myself this must have taken hours and hours and hours
................. then got to last page
"I spent almost a week benchmarking—there are over 4000 benchmark scores, but it was worth it."
lol yep

One thing I found odd, and maybe I missed it was the high latency of DDR5 at 2400-3200 speeds, I would have assumed it could do the same latencies as DDR4 at the same speeds.

 

[AlwaysHope]

Big job there W1zzard, well done indeed! I'd be curious to see the effects of DDR4 when gear 2 kicks in around 3733Mhz & up on AL + RL compared to all that DDR5 stuff you did. Nice market now for really high end DDR4 sticks out there today. Plenty of kits in the 4,000 - 5300Mhz range available.

 

[Outback Bronze]

I haven't seen any reason at all to purchase a DDR5 board hence my DDR4 board purchase.

Gona wait for DDR5 to mature which could take me to Z790 with new CPU's.

 

[Garrus]

I haven't seen any reason at all to purchase a DDR5 board hence my DDR4 board purchase.

Gona wait for DDR5 to mature which could take me to Z790 with new CPU's.

The ram cost more than my motherboard. It was an easy choice to avoid DDR5

 

[potatosoup]

... DDR5 is implicitly Gear 2 across the board—a 1:2 divider between the command frequency and DRAM clock

Is this a limitation with either Alder Lake/motherboards or mostly for ease of testing? I would be really curious to see DDR5 in gear 1 at whatever the highest it will go at its lowest latencies. I would guess something like 3600 or 3733 (possibly require a voltage bump?). It seems like it might have some future-proofing benefits to get into the sparse DDR5 ecosystem with slow/small sticks then plan to upgrade them later as RAM is easier to resell typically than motherboard.

 

[Minus Infinity]

Wake me when DDR5 7200 CL32 is on par with DDR4 3600 Cl16 pricing.

 

[tussinman]

I'm still on DDR3. Games run fine and windows is quick.

DDR4 will be plenty fast for years to come, not much incentive to jump shit

 

[Outback Bronze]

The ram cost more than my motherboard. It was an easy choice to avoid DDR5

They were my thoughts exactly too matey : )

 

[W1zzard]

One thing I found odd, and maybe I missed it was the high latency of DDR5 at 2400-3200 speeds, I would have assumed it could do the same latencies as DDR4 at the same speeds.

I kept timings constant, so we see the effect of frequency alone, and not a mix of both. Interesting question though, let me test how far I can tighten timings at those frequencies

I would be really curious to see DDR5 in gear 1 at whatever the highest it will go at its lowest latencies. I would guess something like 3600 or 3733 (possibly require a voltage bump?)

I tried lower frequency and higher voltage, Gear 1 is still not working. Maybe I'm missing something, dunno

 

[Wirko]

I tried lower frequency and higher voltage, Gear 1 is still not working. Maybe I'm missing something, dunno

There's a lengthy description here: https://skatterbencher.com/2021/11/...ts-new/#Alder_Lake_Memory_Controller_Overview
Gear 1 not working may have something to do with "Gear-down mode", which apparently is a BIOS setting that needs to be enabled. Or disabled. Not that I understand much.

I also found this interesting bit here: "Note that the lowest Cas Latency supported by DDR5 is 20." Is that the absolute minimum, regardless of MT/s?
Gear 2 also means that the clock can be set in 200 MHz intervals, so speed can only be set in 400 MT/s intervals (if the base memory clock is 100 MHz). So, is it impossible to achieve 5000, 5400, 5800 etc.?

 

[mechtech]

I also found this interesting bit here: "Note that the lowest Cas Latency supported by DDR5 is 20." Is that the absolute minimum, regardless of MT/s?

I kept timings constant, so we see the effect of frequency alone, and not a mix of both. Interesting question though, let me test how far I can tighten timings at those frequencies

Hmmm so an apples to apples would be ddr4-3200 22-22-22 (jedec 1.2v) vs ddr5-3200 22-22-22??

Reminds me of the old AMD cpu that had a memory controller that could do 2 gens of ram.