ECC vs. Non-ECC gaming performance

[Stroopwaffle]

I had two sticks of 8GB 2666MT CL22 Kingston DDR4 ECC memory laying around, so I decided to test how much the performance drop would be in my PC versus the 3600MT 32GB CL16 G.Skill Ripjaws I have in my system now. When I was building my machine I was weighing whether or not to go ECC, and decided not to for performance reasons.

 

Now I have tested that hypothesis and these are the results:

My specs are identical throughout both runs (except for RAM obv.):

- CPU: AMD Ryzen 5800X3D

- GPU: Nvidia/Gigabyte RTX2070 Super

- Mobo: Gigabyte Aorus X570 Pro

 

I thought it might be helpful to post this for anyone wanting to make a purchasing decision. For me the choice is clear, if with this CPU the extra fast ram matters so little, I might as well go for ECC ram for the added stability. I use my PC mostly for games, streaming, office work and the occasional 3D modelling. If there are any other benchmarks you guys want me to run let me know.

 

[Somerandomtechyboi]

Overclock the ecc ram ( ͡° ͜ʖ ͡°)

 

though before doing that look at the ics aka black sqares first, you can send a pic to verify the ics unless theyre rebranded, in that case thaiphoon burner would be the best bet

[NF-A12x25]

12 minutes ago, Stroopwaffle said:

I had two sticks of 8GB 2666MT CL22 Kingston DDR4 ECC memory laying around, so I decided to test how much the performance drop would be in my PC versus the 3600MT 32GB CL16 G.Skill Ripjaws I have in my system now. When I was building my machine I was weighing whether or not to go ECC, and decided not to for performance reasons.

 

Now I have tested that hypothesis and these are the results:

My specs are identical throughout both runs (except for RAM obv.):

- CPU: AMD Ryzen 5800X3D

- GPU: Nvidia/Gigabyte RTX2070 Super

- Mobo: Gigabyte Aorus X570 Pro

 

I thought it might be helpful to post this for anyone wanting to make a purchasing decision. For me the choice is clear, if with this CPU the extra fast ram matters so little, I might as well go for ECC ram for the added stability. I use my PC mostly for games, streaming, office work and the occasional 3D modelling. If there are any other benchmarks you guys want me to run let me know.

 

I’d just go with the 32gb 3600mt/s ram. There’s no real use for ECC ram in a rig like yours; if it was a server or NAS rig, or a high-end workstation build for critical tasks I would recommend ECC ram but there’s no point in using it for gaming/productivity work. You’ll get better loading etc. with your 32gb of 3600mt/s due to both the higher speed and amount of RAM.

[porina]

I see the testing done so far primarily as a comparison of 2666 vs 3600. Because it is consumer Ryzen, there is no official support for ECC, so you don't know it is working. There is a big possibility it is running as normal ram. Also the CPU is the X3D, the large cache of which will reduce the influence of the ram anyway.

 

IMO if you really want ECC, you have to use a platform that fully and officially supports it end to end. Consumer AM4 does not qualify.

 

And finally, I'd also question the usefulness of 3DMark's benches as a test. Different games behave differently, so testing ideally would be expanded to a variety of actual games.

[WereCat]

While a good intention from your side, this offers no value to be honest. 

 

5800X3D scales very little with fast low latency DRAM because of its massive L3 cache and 3D mark is great graphics benchmark but it also scales poorly with DRAM. 

 

If you want to see DRAM speed scaling differences you should try Linpack or SuperPi benchmark. 

[Stroopwaffle]

19 hours ago, WereCat said:

While a good intention from your side, this offers no value to be honest. 

 

5800X3D scales very little with fast low latency DRAM because of its massive L3 cache and 3D mark is great graphics benchmark but it also scales poorly with DRAM. 

 

If you want to see DRAM speed scaling differences you should try Linpack or SuperPi benchmark. 

While I agree that 3Dmark doesn't tell the whole story, benchmarks like SuperPi don't really show me what I can expect when daily driving my machine. I'm sure there's a difference, but if I can't feel it in the applications I use often, then they don't really add any value for me.

[Stroopwaffle]

20 hours ago, porina said:

I see the testing done so far primarily as a comparison of 2666 vs 3600. Because it is consumer Ryzen, there is no official support for ECC, so you don't know it is working. There is a big possibility it is running as normal ram. Also the CPU is the X3D, the large cache of which will reduce the influence of the ram anyway.

 

IMO if you really want ECC, you have to use a platform that fully and officially supports it end to end. Consumer AM4 does not qualify.

 

And finally, I'd also question the usefulness of 3DMark's benches as a test. Different games behave differently, so testing ideally would be expanded to a variety of actual games.

There are ways to check if ECC is functional.

 

I agree 3Dmark's value is limited. I'll do some other benchmarks and report back.

[mariushm]

Use 7-zip to compress 10GB+ of random content  then also measure how fast it's decompressed.

 

Ideally, have the folder to compress on a nvme drive, or make a RAM drive and copy the files to the ram drive, to remove hard drive access out of the measurements.

[porina]

46 minutes ago, Stroopwaffle said:

There are ways to check if ECC is functional.

Don't have time to read that right now, but there was discussion in the past that seeing ECC being reported being enabled does not necessarily mean it is functional. At the least you'd need to do error injection and check it is repaired.