Friday, October 22nd 2021

Galaxy Shows off First DDR5 AIDA64 Memory Benchmarks

Remember the "Lego"/"Tetris" DRAM modules that Galaxy was showing off last week? Well, now we have the first benchmarks of said memory running at 4,800 MHz/MT/s with standard timings of 36-36-36-76 and the results are pretty much in line with DDR4 at the same clocks, but with a CAS latency of 19. That said, this doesn't take the latency into account and this is where the issue lies, just as expected.

Despite the change in memory architecture by going to a dual 32-bit bus per DIMM, instead of a 64-bit bus, there is somewhat surprisingly no real gains to be had in AIDA64, even when using two DIMMs. Earlier AIDA64 DDR5 memory benchmarks have only been with a single DIMM, so it has been a bit hard to compare the performance. However, the latency is up by about 30 ns compared to equivalent speed DDR4 memory, which is quite a jump. This does admittedly happen every time there's a transition to a new DRAM technology, but the increase in memory latency has never been this high before. Hopefully RAM with tighter timings and improved CPU memory controllers will help reduce the latency over time, since it'll have an adverse affect on a lot of things, least not many games.
Sources: Galaxy, via @9550pro
Add your own comment

11 Comments on Galaxy Shows off First DDR5 AIDA64 Memory Benchmarks

#1
Metroid
Not impressive, my 32gb 4x8gb dual channel ddr4 2100mhz 16cl, pushed to 21cl 3266mhz is impressive.
Posted on Reply
#2
dj-electric
That's uhh... very Z270 era results.

I don't think its time to show off AIDA64 speeds just yet. DDR5 needs to continue speeding up.
What we see here is the equivalent of DDR4 1866MHz CL17 kits, right at the very inception of consumer grade DDR4
Posted on Reply
#5
TheLostSwede
Caring1On Windows 11?
They didn't specify, but hopefully.
Posted on Reply
#6
Punkenjoy
People only see the latency these days. If you look at most PC stock or using XMP (or people not running AIDA64 lol), those latency are probably in the same ballpark, for a kit that is probably running a JDEC, i would say this is far from the disastrous latency problem of people complaining about DDR5.

On the other side, a kit running at 3266 MT have a maximum theorical bandwidth of 49835 MB/s. Those bandwidth are impressive for a kit running a JDEC specification. Some APUs would love that. The maximum theorical bandwidth is 73242 MB/s for 4800 MT.
Posted on Reply
#8
Metroid
PunkenjoyPeople only see the latency these days. If you look at most PC stock or using XMP (or people not running AIDA64 lol), those latency are probably in the same ballpark, for a kit that is probably running a JDEC, i would say this is far from the disastrous latency problem of people complaining about DDR5.

On the other side, a kit running at 3266 MT have a maximum theorical bandwidth of 49835 MB/s. Those bandwidth are impressive for a kit running a JDEC specification. Some APUs would love that. The maximum theorical bandwidth is 73242 MB/s for 4800 MT.
I'm glad you noticed that.
Posted on Reply
#9
freeagent
Quad channel makes me moist.

Who runs JEDEC? And why?
PunkenjoyPeople only see the latency these days.
No they don't.
Posted on Reply
#10
TheLostSwede
freeagentQuad channel makes me moist.
Four narrow channels...
Posted on Reply
#11
tabascosauz
Doesn't mean latency is out of the picture entirely. If we're talking about these DDR5 timings, it certainly will leave a mark on iGPU 3D performance.

With the way the Cezanne iGPUs currently perform though, it doesn't just keep on endlessly scaling with bandwidth even though bandwidth is the low hanging fruit. iGPU core needs to get newer and bigger as well (stupid 8 ROPs), and hopefully some sort of big improvement is in the works for Infinity Fabric.

As for 4800, well-binned Comet Lake has been doing it for over a year now, albeit without an iGPU to take advantage of it. (10900K is not my screenshot)

Posted on Reply
Add your own comment
Jun 28th, 2022 01:44 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts