Crucial Ballistix Gaming Memory DDR4-3200 MHz CL16 4x16 GB

[Black Haru]

Crucial's new Ballistix Gaming Memory has something for everyone. With options for both RGB and heat-spreader color, along with a wide range of specifications, the Crucial Ballistix Gaming Memory could be a classic in the making. Only time and our benchmarking suite will tell!

Show full review

 

[gamefoo21]

Just found a typo..

I was able to squeeze 3800 MHz out of the Crucial Ballistix Gaming Memory on my MEG X570 ACE with the XMP default timings. For timings, the "Fast" preset booted perfectly! Considering this is a 400 MHz overclock, and some significant timing changes, that's a very impressive result!

Shouldn't that be 600mhz, as the kit is a 3200mhz kit?

What were the primaries of the Fast preset?

Awesome review

Really like that you poked at the overclocking on the AMD side a bit more.

On the memory settings did you do the advanced preset settings? I've seen some decent bumps to performance by tinkering with the channel interleaving settings.

5/5

 

[Black Haru]

Just found a typo..



Shouldn't that be 600mhz, as the kit is a 3200mhz kit?

What were the primaries of the Fast preset?

Awesome review

Really like that you poked at the overclocking on the AMD side a bit more.

On the memory settings did you do the advanced preset settings? I've seen some decent bumps to performance by tinkering with the channel interleaving settings.

5/5

For the timings overclocking I set the frequency to 3600 MHz, so the max frequency and optimized timings are separate overclocks. I have edited the wording to make that more apparent. I don't generally deviate from Dram calculators recommendations. While I agree that there are gains to be had with further tweaking, I also have to keep my production schedule in mind.

 

[Taraquin]

Good review of a awesome kit

 

[olymind1]

Good review, thank you!

Though i would have liked to see how much those RAM OC settings matter in games on the AMD and Intel systems.

 

[TheLostSwede]

Interesting read, as I'm running 4x 8GB single rank modules in my system. I only got hit by 0.7ns increase in latency by going from two to four modules. It would seem using four dual rank modules adds more memory latency, although I guess it could also be partly due to the larger memory chips as well.

It would be interesting to see what these could do at 3800MHz, but it seems like the test CPU simply doesn't want to go past 3733MHz for the IF. Oh well.

I don't generally deviate from Dram calculators recommendations. While I agree that there are gains to be had with further tweaking, I also have to keep my production schedule in mind.

Maybe you can do a separate test where you do some manual tinkering with a few more moderately priced modules? It would be interesting to see what the average 3600MHz kits are capable of, as most seem to be going for well under $100 and at least for Ryzen 3000 owners, that's a really good starting point in terms of what to buy when it comes to RAM.

 

[Chrispy_]

Nice that you tested with the DRAM calculator.

AMD needs to get on the ball with memory vendors to get an XMP-like standard onto SPDs. Ever since 1usmus' first version of the DRAM calculator I've realised how terrible Inlel-optimised XMP settings are for Ryzen. Given that Zen1 and Zen+ are mostly happy to run 1:1 with FCLK at 1600 and Zen2 being all but guaranteed to hit 1800, you'd think there'd be a big market for DDR4-3200 kits with AMD-specific timings in their SPDs, right?

Oh, right. That reminds me - I was going to tighten up the timings on this HTPC - just picked up a 32GB clearance deal 2666 kit to replace the 16GB 3200 B-die and they're still running at XMP. Silly me!

I would have liked to see how much those RAM OC settings matter in games on the AMD and Intel systems.

Intel, not much. Every core in an Intel consumer CPU uses a monolithic cluster of cores and cache. Whilst an oversimplification, the speeds of those cores, caches, and interconnects between them is fixed, regardless of the memory clock - every core sees every core's cache directly. There aren't too many applications that are bottlenecked by faster dual-channel DDR4 so outside of synthetic tests, faster RAM isn't that useful for most consumer workloads. It's been a couple or so since I read up on it, but for Kaby Lake-era I believe DDR4-2666 was the budget sweet spot and anything beyond DDR4-3200 was almost pointless for Intel because of diminishing returns.

Ryzen is far more memory dependent because infinity fabric (IF) clocks (called FCLK) are loosely locked to the RAM clocks. In a 3900X, for example, there are two CCDs (8-core chiplets), each with two CCXs (quad-core clusters) and each of the three active cores in a CCX has its own L2 cache. Despite being on the same chiplet, even the two CCXs in a CCD don't talk to each other, they talk via the IF bus which relays data between CCXs L2 cache and also acts as the bus for all the cores to see the L3 cache in a CCD.

Up until 1800MHz, the FCLK is locked in a 1:1 ratio with the RAM clock, so when you put DDR4-2133 in your Ryzen board, you are running the CPU's inter-core cache bus (among other things) at 1066MHz. If you plop a DDR4-3600 kit in there, the crucial inter-core cache bus gets a huge boost to 1800MHz. At DDR4-3733, there's an automatic divider that drops the RAM clock and FCLK ratio to 1:2, resulting in a 933MHz FCLK which is awful. You need to jump to about DDR4-4400 before the RAM gains offset the penalty of the 2:1 divider. Either way, running FCLK as high as you can is really important for Ryzen because it's reducing cache latencies, and a HUGE cache is a big part of why Zen2 is so much better than Zen1 so any kind of cache upgrades obviously make a difference to Ryzen performance. It may not show up in average framerates, but cache misses are closely tied to your minimum framerates. More, faster cache = better minimum FPS and as a gamer I really appreciate that.

Why 1800MHz FCLK? Because that's what AMD reckon you can safely push the IF clock to. Some motherboards let you set a manual ratio higher, but on the boards I've tried it with, I haven't had RAM fast enough to test beyond 3600. I've also seen an R5 3600 that was only borderline stable at 1800 FCLK and dropped the RAM down to 3533 to prevent occasional cold-boot issues - so 1800 is likely but not 100% guaranteed, the "guaranteed" official spec is 1600 FCLK to go with DDR4-3200.

If you want to see what sort of gains you can get on Ryzen with faster RAM, just google for videos with titles like "Ryzen memory speeds tested in x games" and you'll see some games really benefit hugely from the faster FCLK whilst others DGAF. The next part is just my own opinion, but I feel that Zen 2's default PBO is good enough to clock those chips close to their limits regardless of your motherboard or cooler. Sure, if you spend lots of money on a premium motherboard with overkill VRMs and back that up with a $100+ AIO you will also get clocks up by maybe 300MHz but you are better off getting your RAM up to 3600 first; The difference between affordable 16GB DDR4-3000 kits and more premium DDR4-3600 kits is about $60. That's the best $60 you'll spend on a Ryzen - much more performance/$ than cooling or motherboard upgrades.

If you can afford it, do everything - get the DDR4-3600 and get a 280mm AIO to cool your all-core or PBO+ overclock, and get a premium motherboard to feed the overclock with beefy VRMs. But if you have the budget for only only one or two of those things, make sure you get 3600MHz RAM first. It's the lowest hanging fruit for a lot of consumer applications like gaming, encoding, and those 150 background processes your Windows installation is currently running in the background behind your 100+ Chrome tabs

If you want a better explanation of IF dividers and FCLK, look up buildzoid's channel on YT - That's who I learned from....

 

[Caring1]

Interesting read, as I'm running 4x 8GB single rank modules in my system. I only got hit by 0.7ns increase in latency by going from two to four modules.

That's related to the Topology of the Motherboard, some AMD based boards are said to actually have decreased Latency when all slots are used.

It would seem using four dual rank modules adds more memory latency

Obviously, due to the cycling.

 

[Abomlol]

I've this Crucial kit paired with an ASRock b450m pro4 and a Ryzen 5 3600. It will post and run 3600C14 Fast as the reviewer stated but it has soft errors once you get towards the end of pass 2 in MemTest86. This will manifest as audio and graphics stutter after long sessions. The 3600C16 Safe produced a completely error free setup. The kit will run 3200C14 for anyone looking for compatibility across a wider range. Gotta love Ryzen DRAM Calculator

 

[SirKronan]

Says 4x16 in the title, but the author's screens show 2x16. I was getting all kinds of excited seeing the kinds of numbers these sticks are capable of, thinking, "wow, for 4x16GB this is GREAT!!".

Then I looked a little closer at the screens posted.

 

[Abomlol]

View attachment 153952

Says 4x16 in the title, but the author's screens show 2x16. I was getting all kinds of excited seeing the kinds of numbers these sticks are capable of, thinking, "wow, for 4x16GB this is GREAT!!".

Then I looked a little closer at the screens posted.

You should be able to run it fine at 3200C14, 4 DIMMs if compatibility is your requirement. see my previous post. not as good as 4x16gb hehe
I take issue with the original article. I'd bet dollars to donuts there are soft errors on his overclocks during extended boot tests. That said, this is some straight sleeper ram. Micro Center has it for $75. I'm tempted to buy another set.

 

[Black Haru]

You should be able to run it fine at 3200C14, 4 DIMMs if compatibility is your requirement. see my previous post. not as good as 4x16gb hehe
I take issue with the original article. I'd bet dollars to donuts there are soft errors on his overclocks during extended boot tests. That said, this is some straight sleeper ram. Micro Center has it for $75. I'm tempted to buy another set.

No issues in extended testing. Keep in mind, overclocking results are never guaranteed. On top of variations in the memory binning, you also have the memory controller on the CPU, and the motherboard. It is worth noting, that while the ASRock b450m pro4 is a good budget board, it is not up to the same standard as the MEG X570 ACE.

 

[olymind1]

Wow, very well written and detailed response!

If you want to see what sort of gains you can get on Ryzen with faster RAM, just google for videos with titles like "Ryzen memory speeds tested in x games" and you'll see some games really benefit hugely from the faster FCLK whilst others DGAF. The next part is just my own opinion, but I feel that Zen 2's default PBO is good enough to clock those chips close to their limits regardless of your motherboard or cooler. Sure, if you spend lots of money on a premium motherboard with overkill VRMs and back that up with a $100+ AIO you will also get clocks up by maybe 300MHz but you are better off getting your RAM up to 3600 first; The difference between affordable 16GB DDR4-3000 kits and more premium DDR4-3600 kits is about $60. That's the best $60 you'll spend on a Ryzen - much more performance/$ than cooling or motherboard upgrades.

If you can afford it, do everything - get the DDR4-3600 and get a 280mm AIO to cool your all-core or PBO+ overclock, and get a premium motherboard to feed the overclock with beefy VRMs. But if you have the budget for only only one or two of those things, make sure you get 3600MHz RAM first. It's the lowest hanging fruit for a lot of consumer applications like gaming, encoding, and those 150 background processes your Windows installation is currently running in the background behind your 100+ Chrome tabs

If you want a better explanation of IF dividers and FCLK, look up buildzoid's channel on YT - That's who I learned from....

I have a MSI B450 Tomahawk motherboard, a Ryzen 5 2600 with an Alphenfohn Brocken 3 on it, i like my components Cool'n'Quiet (pun intended ), and a Patriot Viper 4 2x8GB 3400 c16 kit, they have Nanya chips, currently clocked at 3200 cl16 it can't do cl14.

I could replace the Vipers with a Crucial Ballistix 3200 2x8 pair fairly cheaply now, and my current Ryzen would probably like the cl14 latency, and later Ryzens the higher 3600 clock speeds too, if they can OC that much. And i'm curious about the new Zen3 cpus too.

But that i know: i won't buy and bring water close to my PC and if i upgrade my CPU it would have to cost max 150 € in the future.

 

[Chrispy_]

I have a MSI B450 Tomahawk motherboard, a Ryzen 5 2600 with an Alphenfohn Brocken 3 on it, and a Patriot Viper 4 2x8GB 3400 c16 kit currently clocked at 3200 cl16 it can't do cl14.

I could replace the Vipers with a Crucial Ballistix 3200 2x8 pair fairly cheaply now, and my current Ryzen would probably like the cl14 latency, and later Ryzens the higher 3600 clock speeds too, if they can OC that much. And i'm curious about the new Zen3 cpus too.

You're fine for now. the R5 2600 is respectable and 3200 CL16 is likely as high as it will go. Zen+ FCLKs above 1600 aren't very common IME.
Anyway, at 3200 you are already most of the way there - any improvements are going to be marginal and therefore not a good performance/$ spend.

Zen 3 is supposedly going to have monolithic chiplets, so all 8 cores and their caches can talk to each other without needing the Infinity Fabric. No more CCX/CCD shenanigans so I'm speculating that Zen 3 will not care about RAM speeds, at least not for the single-chiplet models like the 6C and 8C models.

 

[SirKronan]

I know I've mentioned this once already, but am I the only reader who thinks it would be nice if the title of the article actually reflected what Nate tested and documented in the article?

 

[W1zzard]

I know I've mentioned this once already, but am I the only reader who thinks it would be nice if the title of the article actually reflected what Nate tested and documented in the article?

I checked with him, it's really 4 sticks, as pictured in the photos and reported in the benchmarks. Only overclocking was done with 2 sticks

 

[SirKronan]

Appreciate that!! So there's still no sticks that can run well on x570 in 4x16 configuration?

 

[Chrispy_]

Appreciate that!! So there's still no sticks that can run well on x570 in 4x16 configuration?

It doesn't depend on this sticks, it depends on the CPU.

Nate's particular RAM kit could overclock to 3800MHz at XMP timings, no matter what CPU was driving it at those speeds and regardless of how many sticks were used together. Each stick = 3800MHz.

Asking a Zen2 CPU to drive four dimms at 3800 is optimistic. You'll have better luck with the good-quality SKUs (like the 3800X and 3950X which are full-functioning dies) rather than the defects that are cut down and clocked lower (so the 3900X and 3700X). Even then, you may need to overvolt the SOC to reach those speeds.

The only way to get speeds significantly higher than that is to use LN2 and change the resistivity of the silicon. I believe the record for 2 DIMMs is 5900MHz or so, and no LN2 overclocker is ever going to bother with 4 DIMMs.

 

[Black Haru]

Appreciate that!! So there's still no sticks that can run well on x570 in 4x16 configuration?

Sorry about the confusion! I have added a line in the overclocking section to clarify.

There are plenty of 4 stick kits that will do great on X570. As others have said, you are usually playing the silicon lottery with this. You can either gamble with a value kit like these, or pay for a more expensive kit that is rated for the speed that you want.

 

[SirKronan]

Sorry about the confusion! I have added a line in the overclocking section to clarify.

There are plenty of 4 stick kits that will do great on X570. As others have said, you are usually playing the silicon lottery with this. You can either gamble with a value kit like these, or pay for a more expensive kit that is rated for the speed that you want.

By "run well" I guess I should specify that I mean run all four sticks at their spec'd XMP profile without having to set to 2T.

 

[Chrispy_]

By "run well" I guess I should specify that I mean run all four sticks at their spec'd XMP profile without having to set to 2T.

The XMP profile GUARANTEES that all four sticks are capable of running at their spec'd XMP profile without having to set to 2T. It's not about the memory sticks, it's about the memory controller built into each CPU driving them.

I guess what you're asking is "how fast can a specific CPU drive 4 DIMMS at once?", and that would be a great thing to see in CPU reviews. Sadly it's not something that is commonly tested. People have to waste time, effort, and money finding out for themselves - or dig around on the web where the answers are often only posted in the context of overclocking, so rarely (if ever) relevant to a 4-DIMM configuration.

 

[SirKronan]

I checked with him, it's really 4 sticks, as pictured in the photos and reported in the benchmarks. Only overclocking was done with 2 sticks

The XMP profile GUARANTEES that all four sticks are capable of running at their spec'd XMP profile without having to set to 2T. It's not about the memory sticks, it's about the memory controller built into each CPU driving them.

I guess what you're asking is "how fast can a specific CPU drive 4 DIMMS at once?", and that would be a great thing to see in CPU reviews. Sadly it's not something that is commonly tested. People have to waste time, effort, and money finding out for themselves - or dig around on the web where the answers are often only posted in the context of overclocking, so rarely (if ever) relevant to a 4-DIMM configuration.

Yup. I got 4x16GB of the 3600MHz Ballistic memory arriving tomorrow. I guess we will see how it runs. Perhaps the memory controller on my 3900x is garbage. That would be very depressing... It's looking like it's my CPU and not the memory. I know I can do 2x32GB for the same price and probably run in 1T, but I do very much like the look of all 4 slots being full.

 

[Chrispy_]

Yup. I got 4x16GB of the 3600MHz Ballistic memory arriving tomorrow. I guess we will see how it runs. Perhaps the memory controller on my 3900x is garbage. That would be very depressing... It's looking like it's my CPU and not the memory. I know I can do 2x32GB for the same price and probably run in 1T, but I do very much like the look of all 4 slots being full.

2x32 won't be much better since each of the 32GB sticks is going to be dual-rank, meaning that it is interleaved by the memory controller in much the same way that two seperate sticks of 16GB would be.

All things being equal, a pair of dual-rank sticks is easier to drive than four single-rank sticks, but both of those options are in the same ballpark in terms of how the CPU's IMC is driving them.

 

[SirKronan]

2x32 won't be much better since each of the 32GB sticks is going to be dual-rank, meaning that it is interleaved by the memory controller in much the same way that two seperate sticks of 16GB would be.

All things being equal, a pair of dual-rank sticks is easier to drive than four single-rank sticks, but both of those options are in the same ballpark in terms of how the CPU's IMC is driving them.

So there may not be a good way for me to run 64GB regardless of what I do? And it has nothing to do with the motherboard or RAM sticks, but my CPU? That is somewhat depressing...

Is there a bit of a silocon lottery with Ryzen 9 integrated memory controllers, and I just got very unlucky, or do most Ryzen 9 CPUs have sucky IMCs?

 

[Chrispy_]

So there may not be a good way for me to run 64GB regardless of what I do? And it has nothing to do with the motherboard or RAM sticks, but my CPU? That is somewhat depressing...

Is there a bit of a silocon lottery with Ryzen 9 integrated memory controllers, and I just got very unlucky, or do most Ryzen 9 CPUs have sucky IMCs?

On the contrary, Ryzen 9 should be getting the better yields of IO die, simply because the twin-chiplet Ryzen 9 processors require the better yields of IO die for two reasons:

1. to meet TDP requirements at stock speeds - the Ryzen 9 TDP is harder to meet with two chiplets, so they need to waste as little of the power budget on the IO die as possible.
2. All of the IF logic to communicate with the chiplets needs to be functional. One assumes that Ryzen 7 and Ryzen 5 models may be making use of IO dies where defects in one of the IF chiplet interconnects is defective, but just deactivated.

In production terms, that means it's far more likely to be the purer silicon towards the middle of the wafer - so whilst it's still a lottery - you are getting better odds than you would if you were using a single-chiplet model like a Ryzen7 or Ryzen 5.

I'm speculating, obviously - but this is the exact sort of thing both AMD and Intel have done before to maximise their profits - and that is their #1 priority, after all!

All you can do is try. Even if it won't run all 64GB at 3600MHz, you have the ability to tweak the timings around a bit.

It's worth noting that 1usmus just updated the Ryzen DRAM Calculator today.