Recently, I bought a used Lenovo Thinkpad T14 Gen3 laptop with 24 GB RAM in it. 24 GB is an unusual amount of memory to have because it is not a power of two. I assumed that there is an 8 GB soldered and a 16 GB replaceable memory stick in the laptop. I didn’t think much of it at the time of buying, I was just happy to have more RAM (in the end 24 > 16).

However when I already had the machine for a few weeks, I wondered, would it be better to have just 16 GB of RAM but have it run in dual-channel mode?

Flex mode - Intel® Flex Memory Access

When reading forums on the topic of dual-channel memory, I found out that there is a technology that enables different memory stick sizes to work partly in dual-channel mode. That is with 8+16 GB memory, the first 8 GB of each module would work in dual-channel mode while the rest 8 GB would work in single channel speeds. My Intel Core i7-1260P processor supports this feature based on the specifications. However, in other forum posts people warn others to use two exactly matching memory sticks to make sure dual-channel mode will work.

It did not help that there is not much clear information on flex mode on the internet. I could imagine that this is not a top priority because most tech savvy people simply just buy a memory kit, but with today’s memory prices one needs to think twice whether an upgrade is worth it or not in my case with a soldered memory chip that is just not possible.

Is flex mode working?

It was already not easy to find out whether flex mode is working or not. I am using Linux Mint now as my daily driver, so naturally I looked up how to check dual-channel function there. Most forum posts lead to the dmidecode command. Based on that I ran my machine in single channel mode (both memory module were in ChannelA as opposed to being in ChannelA and ChannelB separately). Anyways, I also checked what CPU-Z and HW-iNFO says on Windows. CPU-Z indicated 2 x 64-bit, while HWiNFO showed Dual-Channel, suggesting that flex mode is working.

Is it worth going dual-channel mode? How much does flex mode improve over a single channel?

So I thought, okay let’s measure the memory speeds when I take out the additional memory stick (single channel mode) and compare it to the flex mode with 24 GB RAM. After I took out the removable memory module, it turned out that the soldered memory in my laptop is 16 GB and the replaceable is the 8 GB one, the opposite of what I had thought. In addition, I measured the memory speed when inserting an extra 16 GB module (this one I took out from my mini PC) instead of the 8 GB stick, thus achieving real dual-channel mode.

Benchmarks

My memory modules are all DDR4-3200 with 1600 MHz clock, yielding the 3200 MT/s (MegaTransfer/sec) when multiplying by two (hence the name Double Data Rate - DDR). The maximum theoretical bandwidth is then 3200 MT/s x 64 bit = 25.6 GB/s, assuming a single channel. In order to measure the memory speeds, I used PassMark on Windows. Although I tried sysbench on linux but it did not give comparable values to PassMark probably because the memory reads were cached.

PassMark memory speed tests

Based on the above table it seems that the flex mode is not really better than just a single channel mode, at least based on this specific benchmark. Only the Memory Threaded (read) was faster in flex mode by 7% but surprisingly the Memory Write was slower by 13%.

The dual-channel mode definitely made visible improvements on memory read and write speeds. Strangely, the Memory Read Cached value did not improve but actually that was already not a reliable value because the theoretical maximum for the DDR4-3200 is just 25600 MB/s. The Memory Threaded is closer to that theoretical max speed for both single channel and flex mode, while the dual-channel is close to its double with 41600 MB/s = 41.6 GB/s. The Memory Write speed of the dual-channel mode also almost doubled compared to the single channel case. The Memory Read Uncached speed is also higher in dual-channel mode by around 15%.

Benefits in gaming

Normally the system memory speeds have little influence on gaming performance because the GPU has its own memory. That is not the case for integrated graphics though, which uses the system memory. In my case that is the Intel Iris Graphics G7 Xe 96EUs.

Using mangohud, I measured the average FPS (frame per sec) of two arbitrary games: The Forest and the Cult of the Lamb next to lowest graphical setting in Full HD resolution (with Anti-aliasing on in the latter game). The table below shows the measured averaged FPS’s as well as the percent speed-ups compared to the single channel baseline values in parentheses.

Altogether it seems that the dual-channel memory mode can improve fps in games by 20-40% compared to the single channel mode and seemingly the flex mode also exhibit a small performance gain. I have admit that there might be some uncertainty in the numbers, mainly in the case of The Forest, so it would be nice to make a more general benchmark, but the trend seems to be the same in both cases.

Edit: including technical information

I remembered that it would be useful to add my machine and software information because that influences the results and is important for reproducibility. Machine: Lenovo Thinkpad T14 Gen 3 (21AJS2VD00) CPU: Intel Core i7-1260P Integrated GPU: Intel Iris Xe 96EU Soldered memory: SK Hynix 16GB DDR4 3200 MT/s (HMAB2GS6CMR6N-XN) Memory stick #1: Samsung 8GB DDR4 3200 MT/s (M471A1G44BB0-CWE) Memory stick #2: Samsung 16GB DDR4 3200 MT/s (M471A1K43DB0-CWE) OS: Linux Mint 22.2 Linux kernel: 6.18.1

Screen resolution: 1920x1080 I used an external screen during tests, with the internal screen being turned off.