DDR5-4800 trades blows with DDR4-3200 in leaked Alder Lake benchmarks

mongeese

Posts: 521   +111
Staff member
In context: Intel’s upcoming 12th-gen Core processors, codenamed Alder Lake, are rumored to support both DDR4 and DDR5. A battle between the two is brewing, but courtesy of Userbenchmark, we’ve got a preview of the potential outcome.

In the lead up to the next-gen processors’ launch, leakers have been dropping a healthy amount of links to public Userbenchmark and Geekbench results. Their efforts have culminated in enough overlap to distill some meaningful numbers.

Two Userbenchmark results both belong to 16-core, 24-thread Alder Lake processors, presumably the Core i9-12900 or i9-12900K. One was inside an HP Omen system, while the other was in an OEM test bench. They have slight differences, but they’re overall quite similar -- with one exception. One had 32 GB of Kingston DDR4-3200 memory, while the other had 32 GB of Micron DDR5-4800.

Disclaimer: this isn’t an apples-to-apples comparison. One CPU had an average clock speed of 3.05 GHz during the test, and the other ran at 3.65 GHz. They weren’t under controlled test conditions. But memory tests are fairly resilient to small changes, so the two scores can still be pitted against each other to create a baseline for what we should expect.

DDR5-4800 vs. DDR4-3200

  Single-core Multi-core
Read 15.4 GB/s 15.4 GB/s 33.6 GB/s 41.5 GB/s
Write 31.4 GB/s 30.2 GB/s 30.8 GB/s 37.0 GB/s
Mixed 22.4 GB/s 22.1 GB/s 30.2 GB/s 28.8 GB/s

The results for both were remarkably similar in the single-core test, which might indicate that the processor or motherboard was bottlenecking the memory. In the multi-core test, the DDR4 outperformed the DDR5 by a ~20% margin in the read and write tests, but fell slightly behind in the mixed test.

DDR5 did best in the latency test. At a 128 MB data width, the DDR4 required 136 ns to be accessed and the DDR5 required 87 ns. A surprising outcome, compared to past leaks.

All up, the DDR5 results were mediocre. They suggest that, broadly, DDR5-4800 will be a starting point for DDR5 that’s similarly performant to typical DDR4 kits.

Fortunately, DDR5 isn’t limited to 4800 MT/s. DDR5-6400 is inbound, DDR5-8400 is under development, and DDR5-12600 is being talked about. DDR5 could achieve superiority in just a year's time.

Image credit: Pedro Henrique Santos

Permalink to story.

 

QuantumPhysics

Posts: 5,282   +5,999
I'm gonna wait till DDR5 kits and motherboards are on the shelves with full compatibility to whatever generation intel Core i9 is available and then upgrade to a new desktop.

I'm sure I'll see significant speed nd stability improvements - although I'm doing fine right now.
 

Vulcanproject

Posts: 1,402   +2,497
In theory DDR4 spec started at 1600, but in reality nobody used anything less than 2133 even from the early days of the standard. Within a year of Intel's first consumer DDR4 chipset platforms (Z170) 3000-3200 was pretty much the sweet spot of price/performance and everywhere.

I expect the same here. You're probably looking at 6400 as the bottom end widely sold DDR5 kits in a years time and it'll go quickly up from there.
 

Nobina

Posts: 3,335   +3,418
Yuck, you used the ‚U‘ word….

This may be a dumb question, but how is DDR5-4800 performing identically to DDR4-3200 a good thing / sign for DDR5 ?
You won't be buying DDR5-4800, unless you're mental that is, because this is just the early version of it. DDR5 should, like DDR4 before it, become much faster over time. So this right now is nothing to worry about but it points that it should outperform DDR4 in the near future as it matures.
 

Gimp65

Posts: 41   +76
I'm gonna wait till DDR5 kits and motherboards are on the shelves with full compatibility to whatever generation intel Core i9 is available and then upgrade to a new desktop.

I'm sure I'll see significant speed nd stability improvements - although I'm doing fine right now.

Unless Zen 4 comes in and wipes the floor with alder/raptor lake, tbh I think that is going to happend. Just like Zen 3 did with both 10th and 11th gen intel cpu's
 

seeprime

Posts: 597   +745
Every new generation of DDRx has higher CL latency than later models. So, they usually are pretty poor in comparison. CL15 DDR4-3200 will blow away a CL40 DDR5-4800, regardless of what data they show. By the third generation when CL will be 20-25, the DDR5 will scream. Until then, I'm avoiding it.
 

Pastuch

Posts: 88   +89
Every new generation of DDRx has higher CL latency than later models. So, they usually are pretty poor in comparison. CL15 DDR4-3200 will blow away a CL40 DDR5-4800, regardless of what data they show. By the third generation when CL will be 20-25, the DDR5 will scream. Until then, I'm avoiding it.

This guy gets it! There's pretty much no chance DDR5 with high latency will compare to Samsung Bdie 3800 at cas 14 with tight trfc. Still, if the new Intel chips can get me to 280 fps in Warzone I'm buying. I'm cpu limited at 1080p
 

VitalyT

Posts: 5,935   +6,181
I suspected this would happen - too much hype over nothing, high prices + mediocre performance. I needed a new computer recently, and assembled one 1 month ago, with 64GB of DDR4-4000, which is very stable at 4Ghz, and should beat all such entry-level DDR5 modules easily. It probably will be a year at least before a worthy replacement can be seen.

The results that I'm getting on www.userbenchmark.com are:

SC Read 30.7
SC Write 27.5
SC Mixed 44.4

MC Read 62.6
MC Write 48.7
MC Mixed 47.9

Full results - https://www.userbenchmark.com/UserRun/45448533
 
Last edited:

Hardware Geek

Posts: 388   +446
I suspected this would happen - too much hype over nothing, high prices + mediocre performance. I needed a new computer recently, and assembled one 1 month ago, with 64GB of DDR4-4000, which is very stable at 4Ghz, and should beat all such entry-level DDR5 modules easily. It probably will be a year at least before a worthy replacement can be seen.
Exactly. I'm waiting for ddr5 to mature before investing in a new machine. I'm hoping I'll be able to find a good deal on a threadripper pro with 8 channel ddr4 memory support because I want the 128 pcie4 lanes primary for raiding a large number of cheap hard drives for cold data storage. I'm thinking in a year to two years when ddr5 is common ddr4 should be significantly cheaper. I plan to keep the machine for at least 10 years so I want a lot of expansion capability and plenty of networking IO for creating vms the kids can use on cheap thin clients for school work.

Also I'm trying to find any excuse to justify building an awesome workstation I can tinker with and it will be a great way to teach the kids how to work on computers.
 

Axle Grease

Posts: 218   +151
Every new generation of DDRx has higher CL latency than later models. So, they usually are pretty poor in comparison. CL15 DDR4-3200 will blow away a CL40 DDR5-4800, regardless of what data they show. By the third generation when CL will be 20-25, the DDR5 will scream. Until then, I'm avoiding it.

This is what I don't understand.

"DDR5 did best in the latency test. At a 128 MB data width, the DDR4 required 136 ns to be accessed and the DDR5 required 87 ns. A surprising outcome, compared to past leaks."

If, according to the latency test, the DDR5 RAM is faster to access, why doesn't that translate to faster R/W speeds than the DDR4 RAM? I would have thought there'd be a direct correlation.
 

Danny101

Posts: 1,848   +794
It's about what I would expect between high end DDR4 and entry level DDR5. I'll probably hold off on any upgrades until the 2nd to 3rd year chipsets for the platform. My these years went fast
 

captaincranky

Posts: 17,374   +6,125
In theory DDR4 spec started at 1600, but in reality nobody used anything less than 2133 even from the early days of the standard.
That's because Sandy Bridge CPU's IGP (HD-530), had a boot up limit of 2133 Mhz.

I'm sort of a wimp, and never run memory past the maximum boot speed of the onboard IGP. (You know, in case a video card blows up, you don't have to R & R the memory, just hook the monitor to the board. (I fibbed a bit. I'm wimpy, lazy, and cheap). ;)
 

captaincranky

Posts: 17,374   +6,125
Oops Mods, Sorry about that mess. The post button wasn't giving me any acknowledgement of response, along with a massive time delay.
 
Last edited:

Axle Grease

Posts: 218   +151
Oops Mods, Sorry about that mess. The post button wasn't giving me any acknowledgement of response, along with a massive time delay.
You'd think website admins would provide users the ability to delete their own posts. It's not only this website that lacks the functionality.

Edit: Oh, that's interesting. Although the message editor stays open after clicking "Post Reply", the message gets posted anyway.
 
Last edited:

hahahanoobs

Posts: 3,686   +1,767
I've doubled since DDR.
So on the surface, no one should be thinking DDR5 won't be the same.

Except there are other factors with DDR5 specifically, and these tests are not only not controlled, they not being performed on final hardware.

So as of now everyone should be aiming for 6400MHz minimum, but things may change for the better.

RAM just may stop being boring.
 

captaincranky

Posts: 17,374   +6,125
You'd think website admins would provide users the ability to delete their own posts. It's not only this website that lacks the functionality.
In this case it's for the best. Otherwise, I'd be falling on my sword thinking my browser was at fault.

That said, if I post something, it's something I meant to say, let the chips fall where they may.
 
Last edited:

captaincranky

Posts: 17,374   +6,125
I've also noticed some odd forum delays too.
I've gotten some truly bizarre errors. For example, I tried to navigate to the "AMD voltage thread", and got a "that page is not available". The error. It was classified as #500.

I was able to access the page with Firefox, further contributing to my paranoia that Opera was at fault.

I ran CCleaner, then opened Opera again, and got the 500 error when I tried to log in. I tried again and got, "you're already logged in"
 
Last edited:

Axle Grease

Posts: 218   +151
I've gotten some truly bizarre errors. For example, I tried to navigate to the "AMD voltage thread", and got a "that page is not available". The error. It was classified as #500.

I was able to access the page with Firefox, further contributing to my paranoia that Opera was at fault.

I ran CCleaner, then opened Opera again, and got the 500 error when I tried to log in. I tried again and got, "you're already logged in"

It's not your end. My Brave browser also displays the same error for the same thread.

Edit: Yet the thread displays fine in MS Edge...

It displays when I'm logged out.
 
Last edited:
This is what I don't understand.

"DDR5 did best in the latency test. At a 128 MB data width, the DDR4 required 136 ns to be accessed and the DDR5 required 87 ns. A surprising outcome, compared to past leaks."

If, according to the latency test, the DDR5 RAM is faster to access, why doesn't that translate to faster R/W speeds than the DDR4 RAM? I would have thought there'd be a direct correlation.

Onboard ECC