Question 3200 vs 3600 for Ryzen 5000

[blckgrffn]

Hey All -

Have a B450 board that I am going to match with a 5800x. It seems like Ryzen 3 needs 3600 mhz memory? I want to go with 32GB and what I have is sticking with the 3600 into my wife's new build.

Found this for $135 (Cas 18 Oloy 32GB kit):

Are you a human?

www.newegg.com

¯\_(ツ)_/¯

Is it worth seeking lower latency? It seems like lower latency is easier to find with slower kits, which makes sense...

@DrMrLordX you already did some personal shopping for me but I can't find the thread or the recommendation. I'll keep looking though...

Found it! OOS B&H claiming 2-4 week lead time and will let me add it to cart... so... worth trying it at $200?

https://pcpartpicker.com/product/Qh2bt6/crucial-ballistix-rgb-32-gb-2-x-16-gb-ddr4-3600-memory-bl2k16g36c16u4rl

Question - What's the best RAM for a 3600 XT

My cousin is building a new system for his son around a 3600 XT chip and from what I've been following RAM makes a significant difference in performance. What is the best RAM for that chip? MB: x570, he's going top notch on this $350ish Card: RTX 2060 RAM: 16 to 32 gigs of RAM PS: 750 gold...

forums.anandtech.com

We are talking a ~$70 premium for CAS16 (and other timings) compared to CAS18. I like optimizing things but that is a lot of money if the difference isn’t pretty solid.

I also understand that we can expect memory prices to go up so now might a time to get the right parts.

TIA for your $.02!

 

[Shmee]

Zen 3 has the same IO die/memory controller as Zen 2 I believe, though obviously support will vary by board, UEFI revision etc. In general, 3200 with tight timings, or 3600 with looser timings, doesn't matter too much if you are willing to tweak the RAM settings. Really what matters is the type of chip used on the module (for instance Samsung Bdie has tight timings, is very overclockable, and works well with Ryzen in general.)

My recommendation, assuming you are willing to tinker with settings, would be to research what available modules have what chips, and how much the kit costs. Then, buy the optimal memory solution, considering chips used, how many ranks and DIMMs, and of course price.

If you want to just set XMP or use it out of the box, or just minimum tinkering, I would not worry too much about it, just try to get something on the QVL for a decent price, at either 3200 or 3600.

 

[killster1]

think you forgot to mention what your main goal of the system will be and if your 70$ could improve it in another way better. i went with 3600@cas18 32gb on one system and 3200@ cas14 16gb on the next but i have 2x5600x of the 5000 series so far and they both seem identical in speed ;/ (of course i havnt asked my boys to compare them while gaming with 1080ti's ) but for me they seem the same.

 

[Leeea]

3600 with loose timings is better then 3200 with fast timings.

Keep in mind by upping the hz by 400, they are effectively reducing the latency by 12%ish over the same timings on a 3200 part. Sure, you can get better timings, but enough to make up the 12% deficit? probably not.

So the latency ends up similar, but the 3600 can still move more data.

 

[blckgrffn]

Woah, thanks all. Somehow I wasn't getting updates on this thread even though I started it? Anyway, will read and respond today. Thanks again.

 

[blckgrffn]

So I think I am really going to find some decent branded (Crucial, Corsair, etc.) CL16 kit and call it. B&H had some coming in next week at $180 or so and since most decent 32GB kits are hitting like $150+ I think the premium is small enough to pursue.

And yeah, my days of messing with memory timings and voltages are in the past, I don’t have that much patience now. I’ll hit the XMP easy button, run a MS memory test pass or three and call it good.

And if I see an amazing deal on 32GB 3200 CL16 or 3600 cl18? I guess I won’t sweat it . Thanks for your concise advice.

And what else could I do with $70? I guess chase a 5900x? Not like I could find a GPU or anything else measurable. This is my gaming/vanity PC so... yeah. It’s a money pit.

 

[BonzaiDuck]

I'm an interloper on this thread, and I'm AMD-illiterate.

Do AMD processors and motherboard have parallel equivalents to the Intel VCCIO and VCCSA voltage settings?

 

[bigboxes]

I just spent a lot of $$ on ram (as well as the rest of a 5950X build). I went with 3200 CL14. Most benches have the 3200 CL14 faster than 3600 CL16. One more thing to note. Ryzen 5000 series gives an 8-10% increase in performance just by using 4 sticks as opposed to 2 sticks. Capacity doesn't matter. So, I went with 4 x 8GB modules.

 

[kschendel]

... One more thing to note. Ryzen 5000 series gives an 8-10% increase in performance just by using 4 sticks as opposed to 2 sticks. Capacity doesn't matter. So, I went with 4 x 8GB modules.

I'm pretty sure the benefit comes from dual-ranking on a channel, not 4 sticks specifically. You see the same thing on Ryzen 3000. In general, I'd expect fewer issues and better chance at top speeds with 2x16 (using dual ranked 16gb sticks) rather than 4x8; it's a little easier electrically to drive 2 sticks.

As for 3200 vs 3600 MT/s, the transfer rate increase and 200 MHz fabric clock increase don't appear to translate to much in the real world, at least for gaming. A few very memory-intensive, cache-busting programs might see a few percent, most things will see nothing to maybe a percent+. Plenty of benchmarks floating around if you want to look for something specific to your use case. Personally I'd pay a few bucks for the 3600CL18 stuff vs 3200CL16, but not much more than that.

 

[blckgrffn]

Good points all!

Prices on memory seem to be going up. I ended up with this 3600 MHz 32GB CL18 kit:

Check this out on @Newegg: Team T-FORCE DARK Za 32GB (2 x 16GB) 288-Pin DDR4 SDRAM DDR4 3600 (PC4 28800) Desktop Memory (FOR AMD) Model TDZAD432G3600HC18JDC01

It was on sale for $160 which is about where the 3200mhz stuff was and it’s XMP profile is supposedly AMD tuned.

Reviews said that the chips are about at their limit, but plenty of folks were getting basic CL16 timings with 1.35v hand tuning the XMP profile just a bit. I can probably invest that much time into it but I feel like my new 5800x/6700xt is pretty balanced out at 1440p and much more time invested won’t pay much dividends.

 

[Insert_Nickname]

I'm pretty sure the benefit comes from dual-ranking on a channel, not 4 sticks specifically. You see the same thing on Ryzen 3000. In general, I'd expect fewer issues and better chance at top speeds with 2x16 (using dual ranked 16gb sticks) rather than 4x8; it's a little easier electrically to drive 2 sticks.

This^^

But if you want maximum performance out of an AMD memory controller, you'll want 2x dual rank DIMMs per channel (4 ranks per channel). But there is a significant penalty in what frequency you can get out of such a setup. F.x. Ryzen 3000-series is labelled for 3200MHz with 1 dual rank DIMM per channel, but only 2666MHz with 2 per channel. So it's a trade-off between ranks or frequency. Pick what is best for your task.

Single rank DIMMs can achieve really high frequency, but performance suffers due to lack of parallelism. My Athlon 200GE can run a pair of bog standard Samsung 4GB 2400MHz DIMMs at 3533MHz f.x.

 

[bigboxes]

I'm pretty sure the benefit comes from dual-ranking on a channel, not 4 sticks specifically. You see the same thing on Ryzen 3000. In general, I'd expect fewer issues and better chance at top speeds with 2x16 (using dual ranked 16gb sticks) rather than 4x8; it's a little easier electrically to drive 2 sticks.

As for 3200 vs 3600 MT/s, the transfer rate increase and 200 MHz fabric clock increase don't appear to translate to much in the real world, at least for gaming. A few very memory-intensive, cache-busting programs might see a few percent, most things will see nothing to maybe a percent+. Plenty of benchmarks floating around if you want to look for something specific to your use case. Personally I'd pay a few bucks for the 3600CL18 stuff vs 3200CL16, but not much more than that.

No. 4 sticks perform better than two sticks across the 5000 series. Their capacity does not have an effect on that performance. As I said, 4 sticks have about 10% increase in performance in gaming and 8-10% in productivity.

 

[bigboxes]

 

[Magic Carpet]

No. 4 sticks perform better than two sticks across the 5000 series. Their capacity does not have an effect on that performance. As I said, 4 sticks have about 10% increase in performance in gaming and 8-10% in productivity.

It’s definitely easier to overclock 4 sticks of single rank memory. However, 2 sticks of dual rank memory has comparable speed, but you won’t find those in 8gb modules, so most people just put four sticks, if mobo allows or else stick 16gb x 2.

 

[kschendel]

No. 4 sticks perform better than two sticks across the 5000 series. Their capacity does not have an effect on that performance. As I said, 4 sticks have about 10% increase in performance in gaming and 8-10% in productivity.

Yes. Watch the video; they are comparing 4x8 to 2x8, *not* to 2x16. I have not seen anywhere that claims to test 4x8 against 2x16 where the 16GB sticks are dual ranked.

 

[bigboxes]

Yes. Watch the video; they are comparing 4x8 to 2x8, *not* to 2x16. I have not seen anywhere that claims to test 4x8 against 2x16 where the 16GB sticks are dual ranked.

The test was about number of sticks, not capacity. You watch the video again. The variable was not the capacity. It was the quantity of the number of sticks in relation to the performance.

 

[kschendel]

The test was about number of sticks, not capacity. You watch the video again. The variable was not the capacity. It was the quantity of the number of sticks in relation to the performance.

My point has nothing to do with capacity. The point is that dual ranking allows refresh interleave on a channel, and you can get a significant gain from doing that. You effectively get dual ranking for free when you run 4x8. You also get dual ranking from a single 16GB stick that's using 8gbit chips.

Again, this has nothing to do with capacity. If anyone made a dual ranked, 8GB, DDR4 stick, you could install 2 of them for a total of 16GB and still see the dual rank interleave speedup.

The quantity that gamersnexus was varying was stick count, but as a side effect, they were also varying the ranks per channel. There's a very good explanation for why dual rank might go faster (refresh interleave). There's basically no explanation as to why four sticks would go faster other than the fact that you automatically get dual ranking when you install 2 sticks on a channel.

 

[sniper29a]

Do not be screwed up by marketing.

I bought Corsair Vengance DDR4-3200 with official CL17. G.skill DDR4-3600 CL18, on paper faster...in reality slower because they write only first four "most important" timings for PC tuners...in reality your MB auto tune another 50 timings, and your RAM is in reality slower. Of course, you can use Ryzen RAM calculator and experiment with manual setting. I doubt there is enough people to bother about hours of fine tunning.

Corsair 3200 CL17 still possible to OC and it is faster than G.skill.

As someone already mentioned, it is all about chips and binning. The most expensive memory kits use the best silicon and tightest timing + OC capability. Question is, do you need 1% more performance for double/triple price?