Best arrangement of stock and 3rd party RAM?
I bought the 3.8GHz 8-core 10th-generation i7 variety 2020 iMac, with the base config 8GB of 2666MHz DDR4 memory. I’m now trying to install 2x 16GB sticks of HyperX HX426S15IB2K2 (Kingston) and have discovered an impossible optimisation problem.
The stock RAM comes as two 4GB sticks. I tried the following configurations of RAM stick in each slot (remember slots 1 and 3 are DIMM 1 and slots 2 and 4 are DIMM 0):
- Empty
- 4GB
- Empty
- 4GB
Result: 8GB @ 2667 MHz
- 4GB
- 16GB
- 4GB
- 16GB
Result: 40GB @ 2133 MHz
- 16GB
- 4GB
- 16GB
- 4GB
Result: 40GB @ 2133 MHz
- 4GB
- 4GB
- 16GB
- 16GB
Result: 40GB @ 2400 MHz
- Empty
- 16GB
- Empty
- 16GB
Result: 32GB @ 2667 MHz
Here’s a photo of the last configuration to provide physical orientation. Up to towards the top of the computer.
In summary, it seems the only way I can retain the full 2667 MHz speed, is to leave some of the RAM out! Oddly, cross-matching them (putting different size sticks in a single bank) reported a higher speed than matching sizes in a single bank, but I suspect I’d lose out in other ways like parallelisation.
If I want to take full advantage of my RAM bandwidth potential, but would otherwise more RAM than less, what’s my best option? Is there a way to have it all? Is the drop to 2133 MHz real? Is the smaller drop to 2400 MHz a false hope?
In case it’s relevant, the HyperX RAM wasn’t my first choice, but I was steered by stock limitations. It’s not exactly “approved” for use in the iMac by anyone, but matches up spec wise. It has a slightly better CAS Latency of CL15 than the required CL19, but is otherwise much of a muchness. This thread raises interesting points, although nothing hard and fast: What Ram to order with the new 2019 IMac?
Update (08/20/2020)
Well whaddya know, here’s a very detailed thread on why this occurs: iMac 2019 - upgraded w 2 extra ram and the speed dropped to 2400 MHz?
Still nothing conclusive (whether it’s an XMP/overclocking issue, or a single/dual rank issue seems unresolved), and doesn’t really explain why removing one pair restores full speed, but good background nonetheless. Still looking forward to your advice on the best way to proceed (short of buying more RAM).
5 件のコメント
A configuration is missing in your expertise:
1. 16GB
2. 16GB
3. 4GB
4. 4GB
Result: 40GB @ 2667 MHz
k2103 さんによる
Even if the stated speed is 2667, don't you suffer from poor performance due to in ability to parallel? See comment here: Best arrangement of stock and 3rd party RAM?
Heath Raftery さんによる
The tests that I carried out with Geekbench 5 did not show any observable loss of performance with this configuration which is the only one to preserve the advertised speeds of my Crucial bars.
k2103 さんによる
Based on that I thought it worth a shot! Alas, reported speed drops to 2400, just like in the 4,4,16,16 configuration. I also ran a test with Novabench, just to provide a different data point. It reported 27527MB/s in the 32GB configuration and 24096MB/s in your suggested 40GB configuration. I take these numbers with a grain of salt, but that does line up with the 2400/2667 reported speeds.
Heath Raftery さんによる
This may be a good point for Crucial.
k2103 さんによる