How much performance am I leaving on the table with 3200 CL16 on the 5950X

[Remixt]

I managed to get a 5950x two days ago and it's running well. However my memory is a few years old.

I've got an Aorus Master B550 motherboard and I'm running 32GB from a 64GB kit of 8gb Corsair vengeance DDR4 3200 CL16 sticks.(I used to have a Threadripper system hence the 8 stick kit)

 

Am I losing a meaningful amount of performance with ram this 'slow' or are we talking single digits.

 

Also, is my motherboard holding it back at all?

 

 

[da na]

Most likely single digits, but depends on the use case.

[gal-m]

11 minutes ago, Remixt said:

Am I losing a meaningful amount of performance with ram this 'slow' or are we talking single digits.

That depends on whether you're working with latency bound tasks.

In gaming? Single digits, prob. about a couple of % @Remixt

[WereCat]

Depends... what version ov Corsair Vengeance? It should be written in the RAM sticks

[Somerandomtechyboi]

Just overclock the ram to 3600mhz and maybe with some extra voltage lower the timings of the ram, maybe you can reach cl14 timings

 

This way you wont be leaving anything on the table

[YoungBlade]

Since you are running 4 sticks, and therefore have dual-rank operation, you really aren't losing any performance according to Hardware Unboxed's analysis since you game at 1440p (assuming the montior in your sig is being used):

Now, if you switched that out for a 1080p monitor, then you'd be leaving performance on the table. If you game at 1080p, and you upgraded to DDR4-3800, you would see a massive 4.49% uplift to 1% lows, and an unbelievable 3.21% uplift to average FPS.

 

So, basically, you're fine.

[CryingWimp]

Apart from FCLK, having 2 ranks per channel is the most significant performance gain out of all the different ways to optimize ram performance. You already have dual rank by using 4x8GB.

 

At 3200MTs you've already got a significant chunk of the performance bonus that 2666MTs vs 3800MTs gives, most of this performance difference comes by raising FCLK, it is linked to ram speed. If you wish, you can overclock your ram to get the rest of the FCLK performance.

 

This just leaves timings left for any gains to be had by getting better ram. Timings give the least performance difference. A new kit of anything but the most expensive ram will probably give you a couple % at most in some applications.

 

If you spent a decent amount of time doing very custom subtiming tuning on very expensive ram you could potentially get up to 10% in perfect scenarios.

 

 

[Chris Pratt]

10 minutes ago, CryingWimp said:

Apart from FCLK, having dual rank is the most significant performance gain out of all the different ways to optimize ram performance. You already have dual rank by using 4x8GB ram.

I think you're confused. Dual channel matters the most, but that much is achieved with any two sticks of RAM, one in each channel. Ranks is entirely different. Each stick can be single or dual rank, and the optimum number of ranks for Ryzen is four, not two. That can be achieved with two dual ranked sticks or 4 single ranked sticks. 4x8GB gets you there, since all 8GB sticks are single ranked, resulting in four ranks total.

[CryingWimp]

2 hours ago, Chris Pratt said:

I think you're confused. Dual channel matters the most, but that much is achieved with any two sticks of RAM, one in each channel. Ranks is entirely different. Each stick can be single or dual rank, and the optimum number of ranks for Ryzen is four, not two. That can be achieved with two dual ranked sticks or 4 single ranked sticks. 4x8GB gets you there, since all 8GB sticks are single ranked, resulting in four ranks total.

Totally right, I forget that some people don't know to run in dual channel

[Chris Pratt]

34 minutes ago, Somerandomtechyboi said:

Just overclock the ram to 3600mhz and maybe with some extra voltage lower the timings of the ram, maybe you can reach cl14 timings

 

This way you wont be leaving anything on the table

I wouldn't say "just" here. There's no guarantee that you can just take a kit of 3200MHz CL16, and magically make it 3600MHz CL14, just by pumping a little extra voltage into it, especially since the XMP voltage is already relatively high to begin with. RAM is binned, so if the kit could run at 3600MHz CL14, it would have been sold as 3600MHz CL14. There's always a chance you get lucky, just like with any piece of silicon, but it's far from a given.

[Chris Pratt]

There's an incredible amount of bad info floating in this thread. Simply, yes, you are leaving performance on the table because the memory bandwidth is lower and the FCLK is only running at 1600MHz.

 

How much that matters is entirely dependent on how memory and/or CPU bound your given workload is. For gaming, it generally doesn't matter, because even if it's a particularly CPU heavy title, you're still not generally using the full horse power of the CPU, especially with a 5950X. Running at 1440p also helps, because that makes the work far more GPU bound than anything else.

 

When it comes to productivity, that's completely different. In as much as the application makes use of the CPU and memory, you'll see much worse performance than you would with something like a 3600MHz CL16 kit. It's all relative.

 

Long and short, if all you care about is gaming, you're fine. If you're doing productivity work as well (and you shouldn't even have a 5950X if you're not), you might be hampering your performance. It would depend on the exact nature of the work you do, somewhat. However, again, for the type of workloads where a 5950X even makes sense, it's pretty much guaranteed that the RAM is too slow.

[Remixt]

4 minutes ago, Chris Pratt said:

There's an incredible amount of bad info floating in this thread. Simply, yes, you are leaving performance on the table because the memory bandwidth is lower and the FCLK is only running at 1600MHz.

 

How much that matters is entirely dependent on how memory and/or CPU bound your given workload is. For gaming, it generally doesn't matter, because even if it's a particularly CPU heavy title, you're still not generally using the full horse power of the CPU, especially with a 5950X. Running at 1440p also helps, because that makes the work far more GPU bound than anything else.

 

When it comes to productivity, that's completely different. In as much as the application makes use of the CPU and memory, you'll see much worse performance than you would with something like a 3600MHz CL16 kit. It's all relative.

 

Long and short, if all you care about is gaming, you're fine. If you're doing productivity work as well (and you shouldn't even have a 5950X if you're not), you might be hampering your performance. It would depend on the exact nature of the work you do, somewhat. However, again, for the type of workloads where a 5950X even makes sense, it's pretty much guaranteed that the RAM is too slow.

Thanks. I do lots of different work, but mostly in software development which involves compilation of large source code environments and I use Blender for 3d work on top of gaming.

 

[Chris Pratt]

4 minutes ago, Remixt said:

Thanks. I do lots of different work, but mostly in software development which involves compilation of large source code environments and I use Blender for 3d work on top of gaming.

 

Yep. You should get faster RAM. Not even really a question. I can't tell you exactly how much performance you're losing, but with an investment already in a 5950X, honestly, any amount of lost performance is unacceptable.

[freeagent]

Anything over C14 is slow

 

Unless you can run that C16 at  2K+ with flat 16s..

[CryingWimp]

Just keep in mind a lot of the performance gains in all the benchmarks you see include tuning subtimings.

 

A kit of 3600CL16 will make almost no difference compared to your kit OC'ed to 3600CL18. Even a great bin of Samsung B-die 3600CL14 without subtiming tuning will not make much more difference over your kit at 3600CL18. There is a chance your 3200CL16 kit can't do 3600CL18 though.

 

At most, worst case to best case is 25% in specific scenarios. Around 15% of that difference is from having 2 ranks per channel and overclocking to 3600MTs+ for 1800+ FCLK. Timings will make a couple % difference, the remaining difference is from highly tuning your subtimings.

 

To get all of the gains possible you will have to manually tune regardless of how expensive the ram kit you buy is.

 

If you don't plan on OC'ing your current ram than a new kit will make a digger difference because you will gain FCLK as well as better timings, and the higher MTs.

4 hours ago, Remixt said:

Am I losing a meaningful amount of performance with ram this 'slow' or are we talking single digits.

Without expensive ram we are talking way under 5% compared to your kit overclocked. Unless your very unlucky with your ram OC. If you don't OC, a new good kit would maybe be 10% at most.

 

With expensive ram and fine tuning you could gain 10% in the scenarios it matters. Whether it is worth it depends, how much does great ram cost compared to your fun money budget, or how much money do you make with your PC.

 

If the cost of great ram won't affect your life in anyway, then it is an easy "why not"

[MadGoatHaz]

Just my 2 cents here... 

 

5950x + 32gb on micron E-die in my rig running 3800 @ 16-20-16. The performance gained is mostly in the fclk (1900). Mild PBO that keeps temps under 70c paired with a 3080ti, I couldn't be happier with the setup. Of course I could push it more, but why when I have a quiet system that has plenty of grunt (and then some) for everything I want to do with it... 

[Guest]

8 hours ago, Remixt said:

I managed to get a 5950x two days ago and it's running well. However my memory is a few years old.

I've got an Aorus Master B550 motherboard and I'm running 32GB from a 64GB kit of 8gb Corsair vengeance DDR4 3200 CL16 sticks.(I used to have a Threadripper system hence the 8 stick kit)

 

Am I losing a meaningful amount of performance with ram this 'slow' or are we talking single digits.

 

Also, is my motherboard holding it back at all?

 

 

That board has the same VRM as the X570 Aorus Xtreme. It will run a 5950X without breaking a sweat.

[freeagent]

Realistically you guys won’t see a difference. You might pick up a few frames here and there, but it won’t be earth shattering. Performance will still be good.. if not excellent. 

[xg32]

11 hours ago, Remixt said:

.

 

basically 5-10% compared to cl14 3800 tuned, but since it's a 5950x, it's not insignificant, cause the price difference really isn't much when factoring the whole build (but since you had existing sticks just oc them if you want). motherboard could be a dark hero (which isn't guaranteed to hit 3800) for further tuning but it's less of an impact on performance than ram is.