Hi Ronny,
Thanks for your attempts to clarify the situation.
So, that's what we have. According to the specs (see Intel® NUC Kit NUC7CJYH 126135, for example), Max Memory Bandwidth is 38.4 GB/s.
In practice, in the optimal Dual Channel configuration NUC7PJYH provides only 14.5GB/s, which is
38% of Max Memory Bandwidth 38.4 GB/s and 74% of Single-channel mode.
![Scr_ 105.PNG]()
Just for a brief comparison, a couple of tests randomly taken from the web for DDR4-2400 memory:
92% at Single Channel:
![681076664_3_644x461_kingston-8-gb-ddr4-2400-mhz-hyperx-fury-black-hx424c15fb2-8-komplektuyuschie-i-aksessuary.jpg]()
89% at Dual Channel:
![geil_evo_potenza_aida64_mem.jpg]()
That does look normal.
Another example: an ultra-budget 4-year old laptop with DDR3-1600.
90% of 12.8 GB/s bandwidth at Single Channel:
![SCR_ 014.PNG]()
These numbers (around 90% of max bandwidth) look reasonably normal, but 38% does not.
By the way, Ronny, in your demonstration of Celeron J4005 at Dual Channel mode the speed is 12.454 GB/s, (32.4% of bandwidth) and is just marginally faster then for Celeron 1007U at Single Channel mode. That does not look very encouraging.
I'd say, that something is wrong with the memory controller in Gemini Lake chips and that Intel specs are deceptive.
Actually, memory performance does not bother me very much. What bothers me more is that NUC7PJYH does not actually support HDMI 2.0 (not speaking of claimed HDMI 2.0a). I tested it with 11 (eleven) 4K TVs and not a single one was compatible with it at HDMI 2.0. As a result, a $400 pc (NUC+Mem+SSD) is just lying around for half a year and waiting for a compatible TV to appear. If I knew that there is no HDMI 2.0 support I would not buy it.
It would be nice of Intel to fix the product or, at least, to fix the deceptive specs.
Regards,
Peter