Since ryzen 2 (3xxx) supports ECC is it worth getting it over standard DDR4? There doesn't seem to be as high speed ECC available, so it would limit things a bit. I'm just thinking all those cores and threads, does there come a point at which the degree of parallelisation starts to need ECC more?
ECC for desktop and gaming use absolutely not.Originally Posted by ik9000
My housemate has 64GBs of non ECC running VM's without issue. Our use of the VMs is non critical, so loss of power/data corruption isn't an issue. But as yet not experienced any errors. 64GB sounds a lot but when using with VM's suddenly its pitiful.
If you're simply opening up a secure excel document every now and again ECC isn't worth your while. If you have a critical database held in memory then ECC is worth your while.
There never really will be high speed ECC when stability is the purpose.
https://www.pugetsystems.com/labs/ar...CC-Memory-520/
For a home computer playing games, browsing the internet and doing productivity, IMO no. ECC protects what is already written and I don't see why parallelisation has any bearing on error correction here or indeed error rate.
For Ryzen 3000, if you want to splash a bit on memory, 3600-3733mHz is the upper range that AMD said keeps Infinity Fabric at 1:1 so theoretically has the best overall effect on latency.
To use EEC RAM you'd also need support of the motherboard front as it needs extra traces, there's not many consumer boards that support EEC (i think ASUS does one).
But as others have said, if you needed it you'd already know.
Good. That's what i'd hoped. thanks all.
My next question is does single or dual rank matter? I've found conflicting advice on l'interwebs.
I would have loved ECC ram, but it was just too hard to find at a sane price when I bought mine. But then my machine is primarily work rather than gaming, and I use Linux which does report machine check errors if it sees a problem whereas getting that information out of Windows seems harder.
It is hard to get more than 2666 MHz in ECC, but I read somewhere is down to JDEC standards only going that high. Last time I looked the best way to get ECC was direct from the manufacturer's website (Crucial I think).
But in terms of performance, it doesn't buy you one jot. It does however mean that if you overclock your ram, you will know whether the overclock is actually working or not. Many boards support ECC ram.
I think we need to step back, what is the use case?
Matter in what way? From a running the RAM out of spec (overclocking it or the memory controller) then single rank is better as it puts less 'strain' on the MC, from a capacity perspective then dual (or higher) rank is better as you can fit more DIMMs on each stick.
which is what I thought (SR being better), but then I found some articles saying DR is better as it processes more simultaneously per cycle ergo greater throughput. This confused me and I thought they might mean running RAM in dual channel but no it insisted DR was better. I found another article supporting my original understanding, and thought I'd lob it out for the more tech savvy folk to confirm. I'll see if I can find the url for it, but I think the browser update scrubbed the history.
I'm fairly sure wherever you read that DR has higher throughput has it wrong, it's wrong because technically you have two busses (more if the CPU supports more than dual channels), with a stick of dual rank memory each bus is split into two on the actual stick itself and the MC sends a rank select signal to the stick depending on what rank it wants access to, the problem is only one rank on a stick can be selected at a time and if data from the same stick but on the other rank is needed there's a delay while the stick switches from on rank to the other.
If I've done a bad job explaining things the section on ranking of the wiki entry on DIMMs goes into a little more detail.
I suspect they were talking about more chips meaning you can have more pages open in the ram. It's like the dram controller can have lots of bookmarks for where threads were reading/writing so you can get back to that page faster. That makes going back to those pages faster. Two banks of chips means twice the open pages.
OTOH single rank is electrically easier to drive, making high clock speeds easier.
I have a slight leaning to single rank as it can help if you expand the ram, two dimms of dual rank is a lot for the memory controller to drive.
Frankly, even if you can measure the difference I doubt you can feel it, and it will be highly workload dependent on which is best if it does matter. The 3900 has 64MB of L3 cache to hide this stuff
if you could get 2x8GB b-die 3200 CL14 or 3600 CL16 which would be more sensible? Having RAM that will default to the mobo default, and which hopefully will OC higher, or getting the higher one and hoping the mobo doesn't grumble if it needs to run the RAM slower? It would be annoying to end up having to run it 3200-CL16.
The difference is £25 (3600 costing more) and only the 3600 has enough in stock to get 32GB. Usually I'd just get 16GB and wait to add more when I've saved up but the EOL of b-die suggests this might be a risky strategy
Workload is mainly sound editing and photo editing. Large multitrack files can get quite memory heavy. Latency matters. Then 3D CAD and FEM when I use it for work, though more and more of that is going VPN remote-in which makes things slightly easier. (hopefully)
Just when you can't sleep along comes ebay and someone selling 3600CL16-16-16-36 for a sensible price. Decision made. Only 2x8GB but will be fine for the time being. Hopefully someone else will sell a matching pair for a similar price in the future. From the reviews I've read it sounds like 3600 plays best with the infinity fabric and CL16 net latency is pretty close to 3200CL14 anyway, the true latency probably better overall.
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
LinkBack URL
About LinkBacks
Reply With Quote
