DDR4 vs. DDR5 on Intel Core i9-12900K Alder Lake

[beautyless]

I'm appreciate your excellent job as always.

 

[Why_Me]

DDR4 motherboards are not as "feature rich". Expect less of a "premium feel". As for example. Why does the ASUS TUF has 2 SATA ports when 4x is part of the chipset? Save one penny...

4 x SATA

https://www.asus.com/Motherboards-Components/Motherboards/TUF-Gaming/TUF-GAMING-Z690-PLUS-WIFI-D4/

6 x SATA

MAG Z690 TOMAHAWK WIFI DDR4

Powered by Intel 12th Gen Core processors, the MSI MAG Z690 TOMAHAWK WIFI DDR4 is hardened with performance essential specifications to outlast enemies. Tuned for better performance by Core boost, Memory Boost, Premium Thermal Solution, M.2 Shield Frozr,

www.msi.com

6 x SATA

Z690 AERO G DDR4 (rev. 1.x) Key Features | Motherboard - GIGABYTE Global

Lasting Quality from GIGABYTE.GIGABYTE Ultra Durable™ motherboards bring together a unique blend of features and technologies that offer users the absolute ...

www.gigabyte.com

8 x SATA

ASRock Z690 Steel Legend

Supports 14th, 13th & 12th Gen Intel Core™ Processors (LGA1700); 13 Phase Dr.MOS Power Design; 4 x DDR4 DIMMs, Supports Dual Channel, up to 5333+ (OC); 1 PCIe 5.0 x16, 1 PCIe 4.0 x16, 1 PCIe 3.0 x16, 2 PCIe 3.0 x1, 1 M.2 Key-E for WiFi; Graphics Output Options: HDMI, DisplayPort; Realtek ALC897...

www.asrock.com

 

[15th Warlock]

DDR4 motherboards are not as "feature rich". Expect less of a "premium feel". As for example. Why does the ASUS TUF has 2 SATA ports when 4x is part of the chipset? Save one penny...

The Asus Z690 Strix-F and the Strix-A D4 are the same exact board, down to the power delivery and ports, except the latter supports DDR4 and is $50 less expensive.

 

[ir_cow]

4 x SATA

https://www.asus.com/Motherboards-Components/Motherboards/TUF-Gaming/TUF-GAMING-Z690-PLUS-WIFI-D4/

Sneaky sneaky lol. I had the board in front of me for a few days. Didn't even see the ones next to the GPU. I think I need to stop staring at so many boards at once.

 

[gerardfraser]

Awesome review and a bunch of mind numbing work.No surprises for me and will be trying DDR4 DR 4000Mhz Gear 1 on 12900k and maybe up to 4400Mhz on my MSI board

 

[ir_cow]

Awesome review and a bunch of mind numbing work.No surprises for me and will be trying DDR4 DR 4000Mhz Gear 1 on 12900k and maybe up to 4400Mhz on my MSI board

People are trying Gear 1 for DDR4-4000. I am able to achieve 4200 1:1 single rank, but only 3200 Dual Rank. I think the most chips will do 3600 and thats it.

 

[gerardfraser]

People are trying Gear 1 for DDR4-4000. I am able to achieve 4200 1:1 single rank, but only 3200 Dual Rank. I think the most chips will do 3600 and thats it.

I bought the cheapest MSI motherboard and I tend to have good luck on these MSI boards with overclocking and Ram over any Asus motherboard I have.
I am just going to give it a shot,I only PC game at 4K so actual results will never make a real world diference,just fun trying stuff.

 

[chrcoluk]

Interesting data.

Seems my 3200CL14 configuration is a good spot in DDR4 land the 4000+ DDR4 didnt fair too well, with 3200 and 3600 trading blows depending on the workload. Although I think 3200 wins out as on some of the workloads the 3600 loses badly when it loses.

The DDR5 even with its reported worse latency seems to hold out in both latency and bandwidth loads, so is in a good place on most workloads.

Ultimately my original gut feeling I think is correct, those upgrading, keep your DDR4 for now, then maybe at a later date when DDR5 prices come down and better kits come out, upgrade then. Or even keep the DDR4 for as long as you use AL and only move to DDR5 on a newer gen chip.

 

[Taraquin]

I usually love you benchmarks, but this time I'm confused. You test one of the best DDR5-kits vs 4 DDR4 kits that are average at best. What about testing against 3600cl15-15-15, 4000cl17-17-17 or 4400cl19-19-19?

Interesting data.

Seems my 3200CL14 configuration is a good spot in DDR4 land the 4000+ DDR didnt fair too well, with 3200 and 3600 trading blows depending on the workload. Although I think 3200 wins out as on some of the workloads the 3600 loses badly when it loses.

The DDR5 even with its reported worse latency seems to hold out in both latency and bandwidth loads, so is in a good place on most workloads.

Ultimately my original gut feeling I think is correct, those upgrading, keep your DDR4 for now, then maybe at a later date when DDR5 prices come down and better kits come out, upgrade then. Or even keep the DDR4 for as long as you use AL and only move to DDR5 on a newer gen chip.

Single rank gear 2 vs dual rank gear 1 (3200 and 3600). No wonder the 'slower' ram beat the crap out of the 'faster'.

 

[F-man4]

I see the igpu has hdmi 2.1 in its specs, will it output 120hz 4k via hdmi, just for desktop use?

4K 120 output is not tested.
But someone tested that UHD770 can H/W decode 8K 60 AV1 HDR via YouTube without any frame dropping.
(UHD750, RX6900XT, DG1-4G drop frames seriously on 8K 60 AV1 HDR playback)

Besides that, UHD770 gets a larger benefit from DDR5 than from DDR4. That’s because GPU needs more bandwidth than latency.

 

[Space Lynx]

this is rather depressing how little ddr5 actually matters for gamers. ugh. lots of hype and no gains, as usual for the ram industry.

 

[lexluthermiester]

We ran 38 application benchmarks and 10 games at multiple DDR4 configurations to learn what performance to expect when using DDR4 vs. DDR5 on 12th Gen, and whether there's a point at which DDR4 performance can beat the much more expensive DDR5.

Those results are not at surprising. They match what happens everytime the industry transitions from one RAM standard to the next. They also show the age old true that tighter timings wins the day.

While this would just be speculation, do you think that the historical 18months of time before GenA overtakes GenB in overall performance will hold true this generational transition?

Conclusion 1: if you are upgrading your system, recycle your DDR4. You’ll save a lot of money and wont “notice”any performance loss

Conclusion 2: If you are buying a new system, buy DDR4 and with the money saved, get a better processor or GPU.

Anyone see it differently?

Spot on! 100% agree for the current time. In a year to 18 months the picture is likely to be different, but for now this is great advice.

DDR4 motherboards are not as "feature rich". Expect less of a "premium feel".

That's just not correct.

Why does the ASUS TUF has 2 SATA ports when 4x is part of the chipset? Save one penny...

And that's a minor, cherry-picked example..

this is rather depressing how little ddr5 actually matters for gamers. ugh. lots of hype and no gains, as usual for the ram industry.

This is historically what has always happened with RAM standard switch-overs. It'll take time and refinement before DDR5 outperforms DDR4. As shown in the above testing, even DDR5-6000 was only fractionally ahead of any of the DDR4.

As a general rule, if you already have DDR4 and what to carry it forward, buy a DDR4 based board. If you want to bet on the future and have the money to spend, go with the best DDR5 you can get and upgrade later.

 

[Jacek]

Then I would show the problem! I don't know if it's intentional to buy DDR5, but the delay is bad there too.
The minimum FPS is much worse than in the old system ....

z390

z690

 

[W1zzard]

Something i havent seen anyone test yet, is how it does vs high end ddr4, like 2x16 4000 cl14, for example, or 4x8 4000 cl14.

happy to test it if you send me such a kit

Intel or in general?

in general.. i dont see anything convincing here .. and that's lowest possible settings

AMD Ryzen 5 5600G Review - Affordable Zen 3 with Integrated Graphics

The Ryzen 5 5600G is AMD's most affordable entry to the Zen 3 architecture. Priced at $260, this six-core, twelve-thread processor achieves excellent performance that rivals older eight-core processors from both Intel and AMD. The integrated Vega graphics cores are much more powerful than even...

www.techpowerup.com

and run of the mill 4800 at run of the mill timings

due to gear 2 i doubt that can impress

What about testing against 3600cl15-15-15, 4000cl17-17-17 or 4400cl19-19-19?

4000 and 4400 won't matter due to gear 2. i'll see if i can get 3600 cl15 stable on my 2x 16 gb DR kit

 

[chrcoluk]

If you could pull of 3000CL12 that would be interesting. The latency on that beats out all common configurations I have seen on the net, I used it on my 8600k because the chip couldnt handle 3200mhz, and it removed stutters I had in final fantasy 13-2 when enemies spawned. Bear in mind I tuned secondary and tertiary as well, as more gains than may seem obvious come from those, but thats a lot of work to tinker with that stuff, I did require high voltage on the ram to pull it off (1.45v). It also had more throughput than 3200CL14.

I no longer run it as I felt having my ram at 1.45v 24/7 and the chips I have, have no temp sensors was not something I was comfortable with so have dropped back to lower voltage. Now back on XMP 3200CL14 for primary timings/clock and still tuned secondary timings.

I did take photos of bios timings I set, so may send you screenshots if you interested in testing it.

 

[W1zzard]

3000CL12

Not happening with any kit I have

 

[Taraquin]

happy to test it if you send me such a kit


in general.. i dont see anything convincing here .. and that's lowest possible settings

AMD Ryzen 5 5600G Review - Affordable Zen 3 with Integrated Graphics

The Ryzen 5 5600G is AMD's most affordable entry to the Zen 3 architecture. Priced at $260, this six-core, twelve-thread processor achieves excellent performance that rivals older eight-core processors from both Intel and AMD. The integrated Vega graphics cores are much more powerful than even...

www.techpowerup.com

due to gear 2 i doubt that can impress


4000 and 4400 won't matter due to gear 2. i'll see if i can get 3600 cl15 stable on my 2x 16 gb DR kit

I talked to some others and it seems several can do 4000-4133 in gear 1. What is the max speed in gear 1 your samples can do? 3800cl16-16-16 also might be a good option. And dual rank please as it boost perf vs SR a bit

 

[W1zzard]

I talked to some others and it seems several can do 4000-4133 in gear 1. What is the max speed in gear 1 your samples can do? 3800cl16-16-16 also might be a good option. And dual rank please as it boost perf vs SR a bit

Just checked for you, 3800 Gear 1 with DDR4 is the maximum, with +0.1V on System Agent

 

[Zubasa]

Then I would show the problem! I don't know if it's intentional to buy DDR5, but the delay is bad there too.
The minimum FPS is much worse than in the old system ....

z390
View attachment 224615

z690
View attachment 224616

Check if your new system is running in Gear1, the board might be defaulting to Gear2 at higher mem clocks, which halfs the IMC clock and increase latency.
It can also be the board defaulting to really loose sub-timmings. Check those in the bios.

 

[Arcdar]

Thanks again for a very detailed review with lots of background tests that show a "real" user picture compared to most marketing slabs.

I was a bit surprised about the general "low" differences between DDR4 & DDR5, even though I didn't expect massive differences to begin with (I expected this to come with the next generation) but that the general difference is this small is interesting.

Will be interesting to watch the changes with more and more DDR5 releases, better timings and pro's to fiddle with the specs (just look at the Ryzen-Timings-Optimizer which was a lot of work from a gifted individual and changed the performance for many day-to-day-users massively).

But again, kudos for the amount of work which was put into this review (as always, but still not taken for granted )

 

[Ferrum Master]

Ryzen-Timings-Optimizer which was a lot of work from a gifted individual and changed the performance for many day-to-day-users massively

I would not dare to call him gifted after seeing his patreon for few months and his emails, I would rather say vice versa. That calculator doesn't really work for most people really. Better RAM in general accepts more foolish settings especially with those high voltages it asks to set and you won't notice something doesn't work, try some cheaper worser binned RAM and then the rock n roll starts rendering the tool useless.

I would nitcpick that RAM primaries are not the only ones deserving attention. TRFC as usually is overlooked again that massively affects latency then FAW being the next, it is very RAM ic dependent manufacturer states the number in their IC datasheet even. Others are simply a sum of primaries, but tertiaries are an unknown territory for anyone so far I guess.

 

[W1zzard]

What about testing against 3600cl15-15-15

Best I can get is 3600 15-18-18-36. I don't think that's worth spending several hours of testing vs 3600 16-20-20-34

 

[yeeeeman]

ok, so you've confirmed that Intel engineers know what they are doing and that they actually freaking designed and tested the memory controller to work the best with DDR5 memory. Hope that all the nonsense talkers on forums end up their lectures on how they know better than the Intel engineers that DDR5 is gonna be very bad because of "those latencies".

 

[bug]

Interesting.
I would have expected AMD's current gen apu would do 1600x900 with respectable fps.

Superb review BTW.
Bet that was fun

In Minesweeper or Solitaire it can do 4k HDR, no sweat
What does "would do" mean to you? You need to be a bit more specific.

 

[Wirko]

DDR5 memory scaling article is coming next

When testing and comparing multiple memory configurations, could you calculate and list latencies in nanoseconds too, to make comparison easier? Like this: