• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

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

I like that benchmark from RandallFlagg. It perfectly shows that DDR5 does significantly increase memory performance, but you need a use case for it.

I wonder what the results would be when gaming and capturing gameplay using CPU encoding with any of the slow presets (which do make a difference when live streaming at low bitrates).
 
Selaya said:
The question is: Who does that? Why not just use your GPU to do realtime transcoding?
(Yes, there's a time and place for software trancoding since it produces much smaller files, but you usually don't want to do that in realtime ...)
Click to expand...
I guess you mean realtime encoding of the video from the game. The benchmark @RandallFlagg posted is for gaming + transcoding something unrelated to the game, not in realtime but as fast as RAM allows.
 
why would you do that ... ??????
the only realistic use case i can see is realtime en-/transcoding ie, when you're streaming what you're playing, or otherwise recording it?
 
There are clearly a bunch of use cases that can make use of the bandwidth. This is why HEDT and professional workstations have quad channel memory.

Also at the risk of repeating myself, we don't know exactly what effect memory has on real world use cases. The benchmarks run at most sites are run in a vacuum, nothing else is running or being done in order to get consistent results. However it is very normal for people to run more than one application at a time, you don't game with virus scan off, cloud sync off, no browser open on the 2nd monitor, no music playing, and so on and so forth do you? That would be rather.. 2010ish.

This is not to say that current benchmarks are useless, just limited. When you start streaming music videos on monitor #2 while playing battlefield on monitor #1, what is your bottleneck? Perhaps.. it's RAM.

"The workloads are divided by application categories that include media and entertainment (3D animation, rendering), product development (CAD/CAM/CAE), life sciences (medical, molecular), financial services, energy (oil and gas), general operations, and GPU compute. "



1638233839469.png
 
RandallFlagg said:
There are clearly a bunch of use cases that can make use of the bandwidth.
Click to expand...
Not many and they are very specific use-case-scenarios. 99.5% of programs will NOT see such a dramatic bump in performance. Anyone who falls into that category of use-cases will want to get DDR5 right now if they need the extra performance. For everyone else DDR4 is the best choice and will be for the next 18 months at minimum.
 
lexluthermiester said:
Not many and they are very specific use-case-scenarios. 99.5% of programs will NOT see such a dramatic bump in performance. Anyone who falls into that category of use-cases will want to get DDR5 right now if they need the extra performance. For everyone else DDR4 is the best choice and will be for the next 18 months at minimum.
Click to expand...

It's hard to make people think about running more than one application at a time.
 
lexluthermiester said:
Not many and they are very specific use-case-scenarios. 99.5% of programs will NOT see such a dramatic bump in performance. Anyone who falls into that category of use-cases will want to get DDR5 right now if they need the extra performance. For everyone else DDR4 is the best choice and will be for the next 18 months at minimum.
Click to expand...
It's 2021 and Windows deserves a more capable task scheduler. It should be able to prioritise not just processor time but also memory bandwidth usage, as well as access to disk and other I/O. That would be a big part of the solution when some resource, not just CPU, is limited.
 
Wirko said:
It's 2021 and Windows deserves a more capable task scheduler. It should be able to prioritise not just processor time but also memory bandwidth usage, as well as access to disk and other I/O. That would be a big part of the solution when some resource, not just CPU, is limited.
Click to expand...
extremely well said. You notice this limit especially on work-laptops where the handling of specific tasks is pre-set by the administration and you can't really "interfere" with some of it and the lovely software-center starts plotting away in the background installing some updates. The task-scheduler is the weakspot of most usecases right now.

The task-scheduler still handles the CPU load decently well, but the HDD accesses are going as they feel like - which impacts a lot more work-tasks than you might think. Especially if you have some excels as well as Outlook open on one screen while presenting a ppt on the other screen in teams. Sounds like "super light workload" but you wouldn't believe the impact of the stupid update running in the background on the performance of your otherwise still fine work-laptop. Even though you have enough CPU available thanks to the scheduler and prioritizing your focus-application the installer doesn't care and if your HDD/SSD/whateveryourstoragesolutionis is bombarded with "completely unchecked" R/W requests in the background you'll notice it by far more than you might think.



Even though this is pretty unrelated to the DDR4/5 conversation it shows that not everything is HW related and some usecases that people call "far fetched" are closer to reality than they want to admit. Still most of those could be solved by a better SW solution just as well and don't require DDR5 right away / now. In the future - definitely. We'll see a similar trend as with DDR3/4 where it also wasn't a "you have to get it NOW right away or lose out" scenario but a gradual move. Now no one would say "let me get my 12900k with DDR3 please, as it's cheaper and the difference would be marginal" as by now with all the refinements and better as well as faster modules DDR4 is the go-to and no one would want to go back (not even to quad-channel DDR3, even though that served very well for a very long time ^^). The same thing will happen here and I don't get why people are surprised that right now the differences are not "worlds apart" or only in "very specific workloads" (which can also be found for a dual-xeon 269x v4 with quad-channel ddr3 where it outshines nearly everything up to intel gen9 and ryzen2 - but that would be cherry-picking, unrelated and unnecessary as well as not representative of real-world-performance and the "feeling" the user has while working/playing on his machine)
 
Arcdar said:
extremely well said. You notice this limit especially on work-laptops where the handling of specific tasks is pre-set by the administration and you can't really "interfere" with some of it and the lovely software-center starts plotting away in the background installing some updates. The task-scheduler is the weakspot of most usecases right now.

The task-scheduler still handles the CPU load decently well, but the HDD accesses are going as they feel like - which impacts a lot more work-tasks than you might think. Especially if you have some excels as well as Outlook open on one screen while presenting a ppt on the other screen in teams. Sounds like "super light workload" but you wouldn't believe the impact of the stupid update running in the background on the performance of your otherwise still fine work-laptop. Even though you have enough CPU available thanks to the scheduler and prioritizing your focus-application the installer doesn't care and if your HDD/SSD/whateveryourstoragesolutionis is bombarded with "completely unchecked" R/W requests in the background you'll notice it by far more than you might think.



Even though this is pretty unrelated to the DDR4/5 conversation it shows that not everything is HW related and some usecases that people call "far fetched" are closer to reality than they want to admit. Still most of those could be solved by a better SW solution just as well and don't require DDR5 right away / now. In the future - definitely. We'll see a similar trend as with DDR3/4 where it also wasn't a "you have to get it NOW right away or lose out" scenario but a gradual move. Now no one would say "let me get my 12900k with DDR3 please, as it's cheaper and the difference would be marginal" as by now with all the refinements and better as well as faster modules DDR4 is the go-to and no one would want to go back (not even to quad-channel DDR3, even though that served very well for a very long time ^^). The same thing will happen here and I don't get why people are surprised that right now the differences are not "worlds apart" or only in "very specific workloads" (which can also be found for a dual-xeon 269x v4 with quad-channel ddr3 where it outshines nearly everything up to intel gen9 and ryzen2 - but that would be cherry-picking, unrelated and unnecessary as well as not representative of real-world-performance and the "feeling" the user has while working/playing on his machine)
Click to expand...

I notice the same thing, tasks which by themselves are innocuous can become quite a strain.

Background software updates on corporate laptops and such work a lot like Steam updates but you have little to no control on when they kick off, virus scan can kick off, Teams when you have 10+ people streaming video / audio is not a super light load, outlook with
1639663790387.png
in your folder is not a light load when you search or filter or even try to delete or move, Powerpoint presentations can be a hundred meg+ and spreadsheets can have tens of thousands of fields to calculate from. All of these have auto-save and auto-updates running. What's more it is totally common to be using all of these things at once, and more.

These multi-use case hybrid workload really don't show up on any normal benchmarks. Even things like PCMark just single tasks a bunch of common office workloads, which nobody but a PC neophyte single tasks in the real world.
 
A year later I thought I should add. I have Corsair Vengeance DDR5 5600 RGB CL36 36 36 76. I set it to 1.3v from 1.25 and have it running at cl32 34 34 34 70 for the last 6+ months
 
Has anyone seen a Raptor Lake memory scaling review like this for DDR5? Not finding anything really.
 
lexluthermiester said:
That is because RaptorLake CPU's are not out yet.
Click to expand...
Ok um. FORMERLY Raptor Lake. 13th Gen Intel CPU's. ??

I've seen a few articles that you'll find at sites like WCCFtech quoted by Toms, claiming up to a 20% boost in multicore on 13th gen vs 12th gen w/ DDR5.
 
lexluthermiester said:
Oops, you're right. I was thinking of Meteor Lake.


You have a 5800X. There is no reason to consider 13th gen Intel. Your gains would be trivial.
Click to expand...
I haven't updated my sig yet. Gave it to my kid for Xmas.
 
Best memory scaling article, after reading other review sites where frequency doesn't matter this is the first and only site where I can see it matters so I'm glad I ordered the 32 GB DDR5-6000 CL36 kit (which has gone down quite a bit in price depending on which brand you buy).
 
It's been almost 1.5 years since this article was posted and DDR5 is already having a big impact.

We've gotten more powerful CPUs and GPUs, which means the minimum framerate floor has gone up significantly. Most recent reviews show big gains in 1% lows from high-speed DDR5 memory, not just in newest games, but especially in those with ray tracing.

AMD made the right decision in the long run. You have to pay more money now, but you'll probably have two more generations of CPUs supported by this platform. They'll benefit from DDR5 even more.

Intel's platform is a dead end, so it made sense for them to include DDR4 support to attract people with lower costs. But in 2-3 years, Alder/Raptor Lake will be so far behind AMD, where all you'll have to do is replace the CPU with the newest generation.
 
You must log in or register to reply here.
Back
Top