DDR4 vs. DDR5 Memory Performance For Intel Core i5-12600K Alder Lake On Linux

[AlB80]

What memory configurations were used in terms of ranks/banks?

Ok. Obviously both have 16Gbit x8 chips, so it's best case for DDR5 (32GB module) and worst for DDR4 (16GB). Why?
DDR4 memory controller have only 32 banks in total (2 ch x 1 rank x 16 banks per rank).
DDR5 memory controller have 128 banks in total (2 ch x 2 subch x 2 rank x 32 banks per rank).
DDR4 memory transactions are often stalled due to bank collisions, thus even DDR4 good timings are not enough.

[waxhead]

Nobody will know for sure until security researches start hammering DDR5. From what I've heard rowhammer is easiest to execute on laptops where DIMMs are run at the lowest voltages practical to reduce power consumption. They can still operate normally at those voltages, but are more vulnerable to rowhammer. The attacks can work on desktop DIMMs and even (in some cases) ECC RAM on servers, but they aren't as likely to succeed quickly.

Interesting , while one would initially think that higher voltage makes disturbances easier I never considered that lower voltages actually can cause fluctuations/flickering so that you can read/imprint data. Not even sure if this is what happens , but it certainly got my imagination running!

[hajj_3]

You really should update your benchmark suite to use a new build of 7-zip. v16.02 is 4.5yrs old. Lots of optimisations have been made since then.

[creative]

It will be. New memory always starts at around the same MT's as the previous gen, and climbs from there. These are the officially supported JEDEC speeds over each generation. Of course the aftermarket will push the limits a bit more as well:

DDR2 introduced at 400, eventually climbing to 1066.
DDR3 introduced at 1066, eventually climbing to 2166.
DDR4 introduced at 1600, eventually climbing to 3200.
DDR5 introduced at 4400, eventually climbing to ????.

Eventually it will most likely be DDR5 6400 before people start saying "This is where it is at." Specific use cases will benefit right now and certain things will be snappier with current modules but nothing earth shattering. I predict 6400 is going to be the performance rift maker. By then DDR5-6400 should have native platform support and will have some standardization along with those corresponding platforms.

[sarmad]

Not gonna lie, the only thing about DDR5 that gets me excited are the potential performance improvements that iGPUs will get. Most of that excitement is because Ram+MB+APU will probably cost cheaper or close to the price of a Worth-A-Shit-GPU. Basically, if the friggin Steam Deck APU ends up doing what my 4GB RX 580 does, we'll call it enhanced medium 1080p60, then I know I'll be upgrading sooner than later. If not, eh, I'm happy with 3600 DDR4 bumped up to 3800. It's fast enough for my end-user needs.

Indeed. I'd be happy with medium 1080p60 if it saves me from having to deal with the current Prime/Optimus chaos.

[caligula]

DDR4 introduced at 1600, eventually climbing to 3200.
DDR5 introduced at 4400, eventually climbing to ????.

My first DDR4 setup was clocked at 3200 (some kind of overclock gamer sticks, not some IBM/Dell branded shit sticks). I didn't even buy the most performant system, actually. I think my current, 2nd gen machine with DDR4 has DDR4-4200 or 4400 modules.

edit: yea, re-checked. The fastest ddr4 on the market in 5333.

[cl333r]

Not worth it at all at the moment...

Yeah, I recently upgraded from a slow HDD to a new NVMe SSD.
QtCreator used to compile in 25 minutes, now it takes only 12.

[Jabberwocky]

Not worth it at all at the moment...

New truck with twice the amount of torque has an increased top speed of 0.05% ... that's pretty much what these tests are telling me.

Now I agree not everyone is looking for increased torque but the review should be objective, for example show acceleration(latency), top seed(overall-perf), torque(bandwidth) and cost. Let someone decide what's important based on their particular needs. In my opinion Dav1d/PlaidML/Darktable are not "real-world workloads".

Where's the in memory db tests?
Where's the igpu tests?
Just for interest sake: ram drive tests?

Why is DDR5 inherently more expensive than DDR4? Hint: PMIC
How does DDR5 make systems inherently more stable (more so with overclockers and dodgy XMP profiles)?

You don't need to spend hours or days trying to figure out everything yourself, there's a lot of trusted information out about DDR5. Reading up on the questions should take 15 to 30min.

I am sure many would still say DDR5 is not worth it after going through all of that, but at least it would have been an informed choice. I'm sure some would be waiting for DDR5 to get to DDR4 prices and that would never happen (even if global shortages are resloved).

[darkbasic]

New truck with twice the amount of torque has an increased top speed of 0.05% ... that's pretty much what these tests are telling me.

Now I agree not everyone is looking for increased torque but the review should be objective, for example show acceleration(latency), top seed(overall-perf), torque(bandwidth) and cost. Let someone decide what's important based on their particular needs. In my opinion Dav1d/PlaidML/Darktable are not "real-world workloads".

Where's the in memory db tests?
Where's the igpu tests?
Just for interest sake: ram drive tests?

Why is DDR5 inherently more expensive than DDR4? Hint: PMIC
How does DDR5 make systems inherently more stable (more so with overclockers and dodgy XMP profiles)?

You don't need to spend hours or days trying to figure out everything yourself, there's a lot of trusted information out about DDR5. Reading up on the questions should take 15 to 30min.

I am sure many would still say DDR5 is not worth it after going through all of that, but at least it would have been an informed choice. I'm sure some would be waiting for DDR5 to get to DDR4 prices and that would never happen (even if global shortages are resloved).

Phoronix has NEVER given any kind of reasoning behind the data it provided, nor picked representative tests (or options) for every scenario.
You get the raw numbers and sometimes not even the ones you want/need.
That's its biggest limitation and I can understand it being a one man job because it would require a much greater effort to bring the content to the next level.

[F.Ultra]

You really should update your benchmark suite to use a new build of 7-zip. v16.02 is 4.5yrs old. Lots of optimisations have been made since then.

What's being benchmarked is the hardware and not the 7-zip application, and using the same version for 4.5 years means that we can get stable numbers for the differences between various cpu:s for that time period.