Just any 3600MHz memory?

Soldato
Joined
25 Nov 2020
Posts
2,506
Hi all,

For my build I need at least 24GB to run my CAD programs. I think 32GB is safe for a few future releases.
I will have a 3600(possibly X or XT) CPU on an MSI X570 Tomahawk Mobo. I think the GPU will be 5700XT for now.
The idea is to use these stopgap CPU/GPU until I can get what I want, a 5600XT and RTX 3070.

So for memory, I understand 3600MHz would be great for the new Ryzen CPUs. What I don't know though, is which brands to stick to, which to avoid, and which variants of certain brands are winners and which are losers. I also don't have a clue what CL16 or 18.

Really I don't know much beyond the "32GB" aspect so any pointers would be appreciated. If you have a similar system and your RAM is amazing please say so. If you are a memory geek or can advise good value for money options then I am looking forward to your tips.
 
Soldato
Joined
30 Jul 2005
Posts
19,358
Location
Midlands
I think official ram support is 3200 on that cpu as per amd website. Anything higher is overclocking and if used for a work pc id not overclock ram without some serious testing. E.g 72 hour memtest run followed by 48 hour prime blend and 24 hour linpak run.
But each to their own.
 
Associate
Joined
22 Sep 2009
Posts
1,329
Now-a-days it's all about the "Die" used on the actual ram stick.

What you are ideally aiming for is 3600Mhz Cl16. This is achieved by enabling the XMP profile in your motherboards bios.

See it as 3600Mhz being the bandwidth and Cl16 the latency. Good memory have higher bandwidth and lower latency.

For instance, a 3600Mhz Cl16 kit is good, but not much better than a 3200Mhz Cl14 kit.

Now, depending on your budget, you have a few options.

You can go with the very expensive kits that use Samsung B-Die. These kits can obtain very high bandwidth at low latency.

Or you can go with Micron E-Die, which is rediculously affordable for what you get.

Have a look at the Crucial Ballistix kits. They come in a variety of speeds/latencies as well as colours (red, white and black).

They are also often on offer for sub £100 for 32GB kits.

The model number is BL2K16G36C16U4B
 
Associate
Joined
29 Jun 2009
Posts
549
Now-a-days it's all about the "Die" used on the actual ram stick.

What you are ideally aiming for is 3600Mhz Cl16. This is achieved by enabling the XMP profile in your motherboards bios.

See it as 3600Mhz being the bandwidth and Cl16 the latency. Good memory have higher bandwidth and lower latency.

For instance, a 3600Mhz Cl16 kit is good, but not much better than a 3200Mhz Cl14 kit.

Now, depending on your budget, you have a few options.

You can go with the very expensive kits that use Samsung B-Die. These kits can obtain very high bandwidth at low latency.

Or you can go with Micron E-Die, which is rediculously affordable for what you get.

Have a look at the Crucial Ballistix kits. They come in a variety of speeds/latencies as well as colours (red, white and black).

They are also often on offer for sub £100 for 32GB kits.

The model number is BL2K16G36C16U4B

This advice is spot on. Unless you want to manually tune your memory all kits are just as badly tuned at XMP/Auto as the next so you really wont see much difference between them. Crucial Ballistix are well built, reliable and cheap.
 
Soldato
OP
Joined
25 Nov 2020
Posts
2,506
Thank you all for your replies, I am much more informed now.
I'll wait and see if tomorrow-Monday brings any deals on the BL2K16G36C16U4B cards. I'll also keep an eye on other 32gb 3600MHz kits bearing in mind the information above.
I take on board the comment about overclocking a work PC. I have no knowledge of testing a computer in the methods @Cyber-Mav mentioned but I'll look into it. I'd like to run as many tests as necessary/recommended when my build is complete so I know it is solid.

Thanks again, much appreciated
 
Soldato
OP
Joined
25 Nov 2020
Posts
2,506
The BL2K16G36C16U4B kits are proving almost impossible to get hold of.

I have found another 32GB kit (2x16gb) for £145. It's Adata, I know nothing about them. The model is - XPG GAMMIX D10 DDR4 which is 3600MHz CL16.

Alternatively the only other kit at 3600MHz CL16 that I can find is on OCUK
My basket at Overclockers UK:
Total: £208.69 (includes shipping: £8.70)​

Should I settle for 3600MHz CL18 or go for 3200MHz? Any other suggestions?

My CPU is Ryzen 5800X if that helps.
 
Permabanned
Joined
11 Jan 2019
Posts
3,214
Location
bedlam
72 hour memtest run followed by 48 hour prime blend and 24 hour linpak run

it would would still crash when you open chrome...

Prime is dead now. put silly unreal load on components and is known to kill things.. i've had systems 24hr stable and crash when i open chrome.
Just last nite i was talking about ram overclocking, i have been using memtest and all was 18hr good ;), i was advised to use memtest5 and it found errors within 30 seconds
Overclockers have sold pre overclocked systems that have been long duration linpak stable and they have been RMD'ed on day of delivery because there unstable.


with your testing all you have done is load a system up for 144 hours and not tested it in any way, unless all you do is run prime all day?
you need to test a system with a load based on the use of the system.
 
Soldato
Joined
30 Jul 2005
Posts
19,358
Location
Midlands
Your testing is wrong if chrome is crashing your system and not linpack. My testing has shown linpack pulls the highest power from the wall for any cpu based test that can be done. If your crashing with just chrome then you got a very dodgey system and thats usually ram related. Linpack work best with max ram allocation. Iv seen it be useless when people ran it with 4gb assigned
Also whats memtest5? Is that memtest86+?
 
Last edited:
Soldato
Joined
30 Jul 2005
Posts
19,358
Location
Midlands
Prime is dead now..

prime i agree is not an actual cpu tester, digging around for my old screenshots and found them here:

over 73 hours prime stable no issues
prime-stress.png


same setup failed in under 1 hour and 15 mins of linpack. and thats with 3 gig ram allocation.
linpak-fail.png


linpack with large ram allocation finds errors much faster since more ram means problem size is bigger so computations last for longer between iterations.
again these are not ram testers and cpu only. dedicated memory testing is to be done before i even boot the os.
 
Permabanned
Joined
11 Jan 2019
Posts
3,214
Location
bedlam
over 73 hours prime stable no issues

you didnt read my post did you..
i'm not saying your system wasn't 72hr prime stable(mmm..). what i'm saying is many people do massive prime tests to see if there sysme is stable and after all that they open a program and boom... BSOD.
Prime is for testing dinosaurs not newer hardware.

EDIT: i see your was testing an i5 750. TBF that a dinosaur. but to recommend a 72hr test now days is madness
 
Soldato
Joined
30 Jul 2005
Posts
19,358
Location
Midlands
hahah yea my old nahelem platform, lasted me 10 years, still going strong but fancied a change. also some games not working due to lack of cpu instruction sets think avx.
yes these days a prime blend is ok for some imc testing but linpack and cinebench loops do well even winrar overnight run is handy.
im sure i got a screenshot somewhere of a 150+ hour linpack test (left it running while i went on holiday) on my old xeon.

cant believe how easy it is on ryzen. its just a multiplier increase. no faffing about with fsb or base clocks and worrying about memory and pci dividers its bliss. should have gone multiplier unlocked years ago.
 
Soldato
Joined
30 Jul 2005
Posts
19,358
Location
Midlands
you didnt read my post did you..
i'm not saying your system wasn't 72hr prime stable(mmm..). what i'm saying is many people do massive prime tests to see if there sysme is stable and after all that they open a program and boom... BSOD.
Prime is for testing dinosaurs not newer hardware.

EDIT: i see your was testing an i5 750. TBF that a dinosaur. but to recommend a 72hr test now days is madness

also i did that prime vs linpack test back in the day to prove to people that prime is no where near as good a stability tester as people think it is.
 
Associate
Joined
29 Jun 2009
Posts
549
Also whats memtest5? Is that memtest86+?

I think he means Testmem5 - click for instructions. Extreme Anta777 preset is the hardest to pass, it hammers cache, ram and puts some cpu load on too. In my experience if its stable for an hour or two of that its solid. I'd recommend a fan over your ram while running this stress app, the temperatures it generates are way higher than any you would normally face, its a true worst case scenario test.
 
Permabanned
Joined
11 Jan 2019
Posts
3,214
Location
bedlam
Also whats memtest5? Is that memtest86+?

I think he means Testmem5

yer that's the one, its really speeds up the testing prosses, made more progress in 1 day with Testmem5 that i did is a week with memtest86+

got my VERY budget Edie 18, 22, 22, 22, 42 kit down to 16, 19, 10, 14, 30

the tCRD will not be below 19, this is common with E-die Testmem5 shows error within seconds if i drop it to 18, memtest86+ gave me a 15hr clear run.
 
Soldato
Joined
30 Jul 2005
Posts
19,358
Location
Midlands
I think he means Testmem5 - click for instructions. Extreme Anta777 preset is the hardest to pass, it hammers cache, ram and puts some cpu load on too. In my experience if its stable for an hour or two of that its solid. I'd recommend a fan over your ram while running this stress app, the temperatures it generates are way higher than any you would normally face, its a true worst case scenario test.

managed to download that will let it run for a few days and see if it picks up any errors. since my ram is running below its rated frequency i doubt its gonna find any errors unless they are with the cpu's memory controller.
 
Associate
Joined
29 Jun 2009
Posts
549
yer that's the one, its really speeds up the testing prosses, made more progress in 1 day with Testmem5 that i did is a week with memtest86+

got my VERY budget Edie 18, 22, 22, 22, 42 kit down to 16, 19, 10, 14, 30

the tCRD will not be below 19, this is common with E-die Testmem5 shows error within seconds if i drop it to 18, memtest86+ gave me a 15hr clear run.

Yeah TM5 is my preferred tool for testing a memory overclock. You still need to validate with some 3D testing and a Realbench run afterwards, but so far for me it finds all the things that other apps can miss and quickly too.

managed to download that will let it run for a few days and see if it picks up any errors. since my ram is running below its rated frequency i doubt its gonna find any errors unless they are with the cpu's memory controller.

Don't run it for days, 4 hours is usually more than sufficient. I like to give it at least 4 completed loops before calling a final setting stable, though I tend to just run it for 20 minutes at a time while im still tweaking things downwards.
 
Soldato
OP
Joined
25 Nov 2020
Posts
2,506
I am going to have to find out which tests I should run when I finally build my machine. I'll not be fidgeting too much as it's a work PC but it would be reassuring to see it pass some tests.
 
Soldato
Joined
16 Aug 2009
Posts
7,728
also i did that prime vs linpack test back in the day to prove to people that prime is no where near as good a stability tester as people think it is.

Back in the day Prime was for testing efficiency of cpu coolers not ram stability wasn't it, what furmark was to gpu's if it could stand up to an hour of that without bursting into flames it was considered a pass....
 
Back
Top Bottom