News GeIL's DDR5-4800 32GB RAM Now Available for $350

aetolouee

Prominent
Apr 30, 2021
10
7
515
with 40-40-40-77 timings
Holy latency can't wait to see its impact on frame times when gaming. Hopefully it will be nothing but damn. I guess its the price we have to pay for ECC
 

InvalidError

Titan
Moderator
Holy latency can't wait to see its impact on frame times when gaming. Hopefully it will be nothing but damn. I guess its the price we have to pay for ECC
Except these aren't ECC DIMMs, at least not on the bus-facing side of things. Most of the increased latency comes from the lowered voltage and the command buffering chip.

The first year of DDR4 also had much worse latency than budget-friendly mature DDR3 available at the time. Also, if you want to run today's crazier high-speed low-latency memory, you have to run those DIMMs at 1.35+V when DDR4 is nominally only 1.2V..

As far as games go, it'll probably be a wash - win some from higher bandwidth, lose some to worse latency. The win-lose spread should be interesting.
 

Moscato

Honorable
Jan 11, 2017
3
0
10,510
Holy latency can't wait to see its impact on frame times when gaming. Hopefully it will be nothing but damn. I guess its the price we have to pay for ECC


That latency is actually a jedec specification

That's like using cas 20 ddr4 2400 conceptually
 
The memory runs at DDR5-4800 by default with 40-40-40-77 timings and a 1.1V DRAM voltage so no overclocking is necessary.
$350 for 32GB of slow, bottom-of-the barrel DDR5? It's probably expected, given the typical pricing trends for new RAM generations, though I imagine a number of people will be tricked by the RGB lighting and higher frequency (compared to DDR4) into thinking this is a premium RAM kit. In reality, DDR5 starts at 4800 speed, so this is roughly comparable to the place 2133 serves in the market for DDR4.

That wouldn't be bad if the latency was in-line with existing RAM kits, but typical mainstream (OC) kits of DDR4 tend to have absolute latency of around 10-11ns, while this kit's latency works out to 16.6ns. That's slower latency than even budget DDR4-2400 and 2133 kits have. The frequency is obviously higher, but that can only help so much with latency that high. I wouldn't be surprised if a $120 32GB kit of DDR4-3600 CAS 18 manages to outperform kits like this in games and many other workloads.

People should probably look into comparisons of how different types of RAM affect the performance of Alder Lake prior to building a system around it. Both DDR4 and DDR5 high-end motherboards will apparently be available, but I suspect you will have to pay a lot more than $350 for a DDR5 kit that manages to perform better than DDR4 kits available for a fraction of the price.
 

InvalidError

Titan
Moderator
The frequency is obviously higher, but that can only help so much with latency that high. I wouldn't be surprised if a $120 32GB kit of DDR4-3600 CAS 18 manages to outperform kits like this in games and many other workloads.
Maybe, maybe not. DDR4 may have lower first-word latency but CPU caches usually work with cache lines. DDR5-4800 may take 6.7ns fetching the first word from a row but it completes 16-words bursts 2ns faster, which makes DDR5 only 4ns slower for single bursts.

Not so bad when you also account for the fact that read/writes are usually interleaved across different banks to hide most of the latency and increase data bus utilization.

Decently optimized games usually organize their data in chunks for efficient sequential processing and shouldn't get hurt too badly by the increased latency. Tasks with a higher random read/write pattern like code compiling might get a significant handicap.
 
Maybe, maybe not. DDR4 may have lower first-word latency but CPU caches usually work with cache lines. DDR5-4800 may take 6.7ns fetching the first word from a row but it completes 16-words bursts 2ns faster, which makes DDR5 only 4ns slower for single bursts.
Perhaps, though seeing as even a lot of Z690 boards will be DDR4 models suggests that DDR4 may still be the norm for most high-end PCs this generation, and I'm not sure that would be the case if DDR4 left much performance on the table. Even if it were to perform similar or slightly better, it really doesn't seem like a "low-end" DDR5 kit like this is going to be worth the massive premium over DDR4, at the very least. Of course, that goes for existing high-performance DDR4 kits as well, as the performance differences for paying two or more times as much for high-end RAM tend to be imperceptible in real-world workloads.