[SOLVED] B-Die vs "Regular" RAM ?

[Regev]

Hi

I need a pair of 16GB (each) memory sticks for my i9-9900K + Z390 ITX Aorus combo. I have a Noctua DH-15, and will try overclock the i9.

Am I better off buying a "pre-overclocked" memory (with an XMP of 3600 at 18 or 3200 at 16), or B-Die sticks that I can manually overclock? And if the latter, how much can I realistically achieve - going from 3200 CL14 to 3800 CL18 or what? And how about the price - are the BDie memory sticks worth over the normal sticks, which I assume are much cheaper?

Thanks,
Reg

 

[Newtonius]

You will see little performance difference going higher than 3200MHz. Unless you're talking about applications that are memory clock dependent, you will ultimately see little gains in gaming. Here's a calculator to see the latency differences between the two if you were to get a b-die and overclock. Stick with 3200 unless you want the best money can buy, but it's honestly not worth it.

 

[dorsai]

For what it's worth on my Ryzen 5600x system I am running four 8gig sticks of FlareX 3200 c14 B-die @ 3600 c16 using 1.375v with no issues at all.

 

[Regev]

@dorsai But what's the point of the B-Die? Correct me if I'm wrong but a regular 3600C16 pair of sticks costs the same or less than a pair of 3200C14 B-Dies. Why bother with OC when you can just buy normal, faster ones for the same price?

 

[Regev]

@Newtonius What am I looking at? 3600 at CL18 gives me 10 nanoseconds, whereas my current 2666 CL19 gives 14 nanoseconds. How could anybody even tell the difference?! Even a pair of DDR3 1600 CL19 gives 24 nanoseconds. How could ten or fifteen nanoseconds difference be noticeable ?

P.s I don't even game. It's for productivity purposes 100%, thought it would make the computer a bit snappier

 

[Newtonius]

@Newtonius What am I looking at? 3600 at CL18 gives me 10 nanoseconds, whereas my current 2666 CL19 gives 14 nanoseconds. How could anybody even tell the difference?! Even a pair of DDR3 1600 CL19 gives 24 nanoseconds. How could ten or fifteen nanoseconds difference be noticeable ?

P.s I don't even game. It's for productivity purposes 100%, thought it would make the computer a bit snappier

Unless your productivity software relies on super fast RAM, 3200MHz is plenty snappy for everyday use. Fast RAM is only beneficial for gaming or memory clock sensitive applications. If all you do is browse the web and file excel sheets, then 3200MHz is the best option for you, anything more and you're paying extra for little to no difference in performance. Your CPU is what matters most for productivity (that's not graphical based) and your 9900K is strong enough,

 

[Regev]

@Newtonius - but will there even be a difference going below 3200 - say, to 2666? According to your calculator, it's only 4 nanoseconds difference between 2666 CL19 to 3600 CL18. Heck, it's only 14 nanoseconds difference between 1600 CL19 (DDR3?) to 3600 CL18. How could anybody possibly spot a 14 nanoseconds difference?

 

[Newtonius]

@Newtonius - but will there even be a difference going below 3200 - say, to 2666? According to your calculator, it's only 4 nanoseconds difference between 2666 CL19 to 3600 CL18. Heck, it's only 14 nanoseconds difference between 1600 CL19 (DDR3?) to 3600 CL18. How could anybody possibly spot a 14 nanoseconds difference?

Not really, especially since you don't game. 2666 is an okay speed for productivity use.

 

[Karadjgne]

Only 4 nanoseconds? That's a huge amount to ram. You have to figure that 4 nanoseconds applies to every single transaction, and you are talking about stuff that operates at 3200MHz.
Commander Data said it best.

0.68 seconds sir. For an android, that is nearly an eternity.

The big deal about Samsung B-die was its early Ryzen release compatibility, which has lasted throughout subsequent releases, not really requiring Agesa updates to modify or fix anything. That doesn't apply to Intel cpu's.

As previously stated, there's little (to no) gains to be had from manually OC ram on Intels. It's easier just to move up a notch as the ram is already rated for such. No point in buying 3200/16 and trying to squeeze it to 3600/16 unless the programs benefit from faster ram. Far better to have perfectly stable ram than push an out of spec OC and possibly induce data errors. There's also very little difference between 3000/14, 3200/16 and 3800/18 for most programs. Sure it works faster, but it goes in/out slower. And now you are getting into 0.x nanosecond differences. Not 4 nanoseconds differences.

If you can get 3200/16 to 3200/14 stable, bonus. But that has little to do with whether it's B-die or not, that's totally reliant on silicon lottery.

 

[alexa ben]

i personally perfer the regular ram than b.die

 

[Regev]

Only 4 nanoseconds? That's a huge amount to ram. You have to figure that 4 nanoseconds applies to every single transaction, and you are talking about stuff that operates at 3200MHz.
Commander Data said it best.

The big deal about Samsung B-die was its early Ryzen release compatibility, which has lasted throughout subsequent releases, not really requiring Agesa updates to modify or fix anything. That doesn't apply to Intel cpu's.

As previously stated, there's little (to no) gains to be had from manually OC ram on Intels. It's easier just to move up a notch as the ram is already rated for such. No point in buying 3200/16 and trying to squeeze it to 3600/16 unless the programs benefit from faster ram. Far better to have perfectly stable ram than push an out of spec OC and possibly induce data errors. There's also very little difference between 3000/14, 3200/16 and 3800/18 for most programs. Sure it works faster, but it goes in/out slower. And now you are getting into 0.x nanosecond differences. Not 4 nanoseconds differences.

If you can get 3200/16 to 3200/14 stable, bonus. But that has little to do with whether it's B-die or not, that's totally reliant on silicon lottery.

Thanks for the reply man.
I don't understand something tho - how could anybody spot a 4, heck, even a 500 billionth of a second delay ?

 

[Karadjgne]

You cannot. Because a human measures time in seconds. A pc running at 4.4GHz is running at 4,400,000,000 instructions per second so 1 instruction is happening every 2.27E-10 seconds.

We can also divide time up into tenths of a second etc. How far can the pc further divide time up...

 

[Regev]

@Karadjgne I see, then how does 4133, 3600, or 3200 memory have any noticeable benefit over 2666 ? It's a 4 nanosecond difference between 2666 CL19 to 3600 CL18. Cause everyone says anything above 3200 and you don't see a difference. Well, why not 2666 and above?

 

[dorsai]

@dorsai But what's the point of the B-Die? Correct me if I'm wrong but a regular 3600C16 pair of sticks costs the same or less than a pair of 3200C14 B-Dies. Why bother with OC when you can just buy normal, faster ones for the same price?

For an Intel rig, which is not as sensitive to latency issues as Ryzen processors, you can get away with loose timings. On previous Ryzen processors like 2700x or 3600 the B-die memory was necessary to get good tight memory timings to keep the chips Infinity Fabric happy and providing good performance. The latest 5000 Ryzen chips are a bit less sensitive to latency issues than earlier Ryzen so instead of 3200 @ C14 I run 3600 @ C16 which my 5600x is happy at since it pushes the Infinity Fabric close to max speeds.

I provided my results in the "for what it's worth" context simply to show what B-die is capable of. If I was running your rig I would likely save some money and do what you suggested by picking up some cheaper ram.

 

[jtk2515]

B die is pretty good Ram. 3200mhz cl14 is usually in the 80-90$ for 2x8gb. I have gotten multiple sets and just put 1.5volts and gone to 3600mhz and usually get in the cl14-15 range. Just make sure you get airflow over the Dimms. B-Die is really good at accepting voltage and heat.

Is it Worth it? Always depends.buying 3600@cl 16 vs overclocking b-die to 3600@cl14 makes almost no difference and could take a few hours if your trying to get the lowest possible voltage. If that extra 1-5% gets your gears turning then do it.

 

[Regev]

What's the cheapest 3200@CL14 kit on the market right now? (2x16)

 

[Endre]

Hi

I need a pair of 16GB (each) memory sticks for my i9-9900K + Z390 ITX Aorus combo. I have a Noctua DH-15, and will try overclock the i9.

Am I better off buying a "pre-overclocked" memory (with an XMP of 3600 at 18 or 3200 at 16), or B-Die sticks that I can manually overclock? And if the latter, how much can I realistically achieve - going from 3200 CL14 to 3800 CL18 or what? And how about the price - are the BDie memory sticks worth over the normal sticks, which I assume are much cheaper?

Thanks,
Reg

Just my opinion here:
Intel Z390 platform’s RAM spec is DDR4-2666.
That’s what Intel recommends.
Which means that for most users, going higher is pointless.

Since I’m on a Z390 platform myself, I’ll make a RAM recommendation:
Kingston HyperX Fury HX426C16FB3/16
https://www.kingston.com/dataSheets/HX426C16FB3_16.pdf

Pro’s:

1. It’s very stable because it runs at 2666MT/s CL16 @1.2V by default (JEDEC standard. No XMP needed).
2. It’s pretty cheap.
3. It’s on the QVL of most Z390 motherboards.

 

[Regev]

I thought Intel limits the memory to 2666 so that they can sell XMPable motherboards - like the Z390s. Won't the board run better with 3600 than 2666 ?

 

[Karadjgne]

With Intel cpu's, that spec isn't a recommendation. It's a spec situated around the working speeds of the memory controller. If the cpu has a 2666MHz capable mc, it gives a 1:1 ratio, so is most in sync, and doesn't produce any extra heat or possibly require extra voltage used, demanded or applied. All that has ties directly and indirectly going back as far as Energy Star compliance ratings, EU power regulations etc.

For non-Z boards, that ram spec is part of the chipset limits on the board, which realistically is tied to the components used. They can be lower grade silicon, not require extra or extensive circuitry etc.

And that ties back to the AIB partners and profits. You don't need to make a super-heavy duty power capable B rated motherboard that'll handle the 250w of an OC 10900k, the traces used only need be capable of the 95w TDP plus a little more.

Amd gets around all that with other chipset limitations, like full gen4 pcie on x570 and not on X470, limited on B550 etc.

Open ended Z boards have to have the capacity to deal with anything Intel can socket, but even then AIB partners split that up, reserving the elitist and most expensive boards for those who can afford to pay for the expensive cpus. But you can run 5.0GHz ram on the lowest rated Z mobo.

Whether a cpu runs better or not with faster ram isn't dependent on the cpu or mobo or any hardware, it's entirely dependent upon the software. B boards were originally slated for Bussiness use, they had different storage software included, more parallel ports, a slightly different approach. H boards were for general homeowners, had more bells and whistles and tailored more for what the average homeowner had access to. Z rated boards were for the extremists, who pushed boundaries with OC in an attempt to get more than factory performance etc. Q boards were the industrial side, production etc. Each had its niche. But as software progressed, that changed, some software makes better use of higher speed ram, some makes use of higher amount ram, some needs single core performance, some multi core performance, and very few just use 1 program. Mixing AutoCad with Arnold you need everything.

So a 10900k can do higher than 2933MHz, can do high speed single or multi core, but software demands and power demands make that cpu require a higher rated Z board, not a low ranked budget B.

And everything has ties to the past. Back in the day you had the SX-33, DX2-66 etc. That was buss speeds essentially. That's continued today with the use of multipliers. It's exactly why 3000MHz ram does not work, XMP not withstanding, on some motherboards. 2933MHz is 266MHz buss x11 multiplier. 3200MHz is 266MHz buss x12 multiplier. 3000 is an oddball, it's a 200MHz buss x15. Even 2666MHz is actually a 266MHz buss x10 since 266MHz used at decimal points is 266.67... An x8 multiplier of the original 33MHz.

1080, 1333, 1600, 1866, 2133, 2400, 2666, 2933, 3200, 3466, etc all multipliers of 33MHz buss.

Also why programs like cpu-z never read a perfect 100.00 BCLK (buss clock), it's always 99.9x etc as it is reading the 33MHz x multiplier and rounding to a decimal.

 

[Regev]

@Karadjgne DUDE! Thanks for all that golden wisdom. "If the cpu has a 2666MHz capable mc, it gives a 1:1 ratio, so is most in sync" - does that mean 2666 will be most optimal, or will I benefit from going up to 3600 ?

 

[Karadjgne]

That depends entirely on the software. Intel gives you a cpu that's fully stable and good for its default settings. Some software gets better computational speeds out of higher speed ram than Intel stock, so a 3200MHz kit will get higher fps or more compile work done for a given amount vs lesser speed ram. Some software doesn't care about ram speeds, it takes so long to push a code string through the cpu that the ram is left waiting on demands anyways. Then core count or cpu core speeds or IPC can make a big difference.

Since the early days, Intel has specialized in speeds and IPC, Amd went the route of cores with FX. AMD had better number crunching, Intel had better gaming, so software specialized even further to take advantage of that.

But software expanded, it had no choice, to get the graphics, more cores were needed by Intel, to get the fps more speed was needed by AMD. But both still have uses for higher speed ram possibly.

It's also why Xeons have disappeared from mainstream, without OC capability and with such high core use, and relatively low speeds to maintain some semblance of voltage/heat control, they end up niched out of mainstream use.

Game files are very small, mostly just a couple of Kb in size. So they pass through a cpu fast. Which means to cut down on lag, they must also pass through ram fast. If you can get ram kicking data to the cpu faster than the cpu can process, you've maximized fps. If the cpu has to wait on ram...

So there's generally a benefit to lower timings, higher speed ram. How much of one depends on the software itself.

 

[Endre]

I thought Intel limits the memory to 2666 so that they can sell XMPable motherboards - like the Z390s. Won't the board run better with 3600 than 2666 ?

RAM specs are being established by JEDEC.
For DDR4 the voltage is 1.2V.
Any voltage higher than that is an overclock, which is not approved by JEDEC, and it’s not supported natively by Intel or AMD.

Motherboard manufacturers create high-end boards that allow you to crank the voltage higher: 1.35V or even higher (for DDR4).
XMP is a “safer” overclock!
But still an overclock!
There are no guarantees that the memory DIMMs will be able to sustain those XMP speeds for long periods of time.

The highest DDR4 speed permitted by JEDEC (that I know of) is:
DDR4-3466 CL19 @1.2V like this one:
https://www.kingston.com/dataSheets/HX434C19FB_16.pdf