Friday, October 20th 2017

G.SKILL Releases DDR4-3800MHz 32GB (4x8GB) SO-DIMM Kit for Mini-ITX HEDT Boards

G.SKILL International Enterprise Co., Ltd., the world's leading manufacturer of extreme performance memory and gaming peripherals, is thrilled to announce the release of a new high performance DDR4 Ripjaws SO-DIMM series for small-form factor (SFF) platforms, at DDR4-3800 MHz CL18-18-18-38 32 GB (4x 8 GB) at 1.35V, making this the world's fastest SO-DIMM memory kit. In addition, 3 new specifications will be added to the Ripjaws DDR4 SO-DIMM family, including DDR4-3600 MHz CL16-16-16-36 32 GB (4x 8 GB), DDR4-3200 MHz CL16-16-16-36 32 GB (4x 8 GB), and DDR4-3200 MHz CL16-16-16-36 64 GB (4x 16 GB). Such extremely high speed on SO-DIMM modules are achieved with the use of the ultra-high performance Samsung B-die DDR4 ICs.

At Computex 2017 in June, G.SKILL demonstrated high-speed Ripjaws SO-DIMM series at DDR4-3466 MHz, and we are now once again raising the speed of the fastest SO-DIMM memory kit to the intense mind-bending speed of DDR4-3800 MHz. Validated on the latest ASRock X299E-ITX/ac motherboard and the Intel Core i9-7900X X-series Processor, extreme memory performance is now a reality with small-form factor workstations.
The following stress test screenshot shows the new DDR4-3800 MHz memory kit in action with a total run time of over 9 hours:
These new powerful SO-DIMM memory kit will be available through authorized G.SKILL distribution partners in early December 2017.
Add your own comment

19 Comments on G.SKILL Releases DDR4-3800MHz 32GB (4x8GB) SO-DIMM Kit for Mini-ITX HEDT Boards

#1
dj-electric
They see the potential and use it. Nice one
Posted on Reply
#2
CheapMeat
SODIMMs should be the standard for all consumer boards. We lose out on nothing but gain more space on the board. ECC is so artificially segmented, so we don't get it and we don't need Registered/Buffered ICs either. There's not even one 3200MHz ECC kit out there even though that's the JEDEC spec currently. So the pins and the PCB go to waste on full DIMMs for us. We're getting more and more SODIMM options. There's really no reason to keep using full DIMMs even on ATX consumer (Ryzen & Core series).

DDR4 SODIMMs still have more pins than DDR3.
Posted on Reply
#3
Mirkoskji
Yeah indeed can someone explain why consumer memory hasn't switched all to sodimm?
Posted on Reply
#4
TheLostSwede
Because, standards?

Can someone tell me why we continue to use so many old computer interfaces? Same problem, standards.

In this case it's clearly also about availability, due to, you guessed it, standards. Very few companies makes high-performance SO-DIMMs and it's not something that could be switched over night.

Now, there are a fair few mini-ITX boards that utilise SO-DIMM memory, but they're one out of ever 25 motherboards at the most.

If there's no volume production, costs will be high and manufacturers will shy away from making it. It's a chicken and egg effect to an extent as well.

The better standard doesn't always win...
Posted on Reply
#5
lexluthermiester
"Mirkoskji said:
Yeah indeed can someone explain why consumer memory hasn't switched all to sodimm?
Yes, because SoDimm's have fewer data lines and thus lower data through-put which of course means lower performance.
"TheLostSwede said:
Because, standards? Can someone tell me why we continue to use so many old computer interfaces? Same problem, standards.
Standards are not, as you put it, a problem. Standards exist so we all can have the upgrade potential we enjoy. Without standards the world would be the way it once was, a convoluted mess of proprietary garbage. Perhaps you're too young to remember such a time and appreciate how much better it is.
"TheLostSwede said:
In this case it's clearly also about availability, due to, you guessed it, standards. Very few companies makes high-performance SO-DIMMs and it's not something that could be switched over night.
That's an interesting theory.
"TheLostSwede said:
Now, there are a fair few mini-ITX boards that utilise SO-DIMM memory, but they're one out of ever 25 motherboards at the most.
You really need to recount..
"TheLostSwede said:
The better standard doesn't always win...
That's true. Example; the 2.88MB floppy was an excellent and inexpensive data device platform. Double the storage space and faster access times of standard floppy's of the time. But for whatever reason no one adopted it. And it wasn't IBM's fault as they made it an inexpensive option.

Does any think there are any reasons this couldn't be used in high-end laptops?
Posted on Reply
#6
Mirkoskji
i didn't knew there was a limitation in total throughput of so-dimm, good to know.. as of the discussion about expensivness of the solution, they are expensive only because they aren't extensively adopted. it costs virtually nothing to put higher frequency dram chips on a so-dimm module and extend the production. but i would think that for low end, itx or not-overclockable motherboads, so-dimm could be a good solution.
Plus they already are a standard. Memory producers put out way more so-dimm modules every year for the consumer market than normal dimm's.
Posted on Reply
#7
TheLostSwede
"lexluthermiester said:
Standards are not, as you put it, a problem. Standards exist so we all can have the upgrade potential we enjoy. Without standards the world would be the way it once was, a convoluted mess of proprietary garbage. Perhaps you're too young to remember such a time and appreciate how much better it is.

That's an interesting theory.

You really need to recount..

That's true. Example; the 2.88MB floppy was an excellent and inexpensive data device platform. Double the storage space and faster access times of standard floppy's of the time. But for whatever reason no one adopted it. And it wasn't IBM's fault as they made it an inexpensive option.

Does any think there are any reasons this couldn't be used in high-end laptops?
Uhm, buddy, don't draw conclusions from thin air. I've been around long enough to remember it all. Yes, standards are there for a reason, but many won't die a graceful death either. Why does the latest generation of motherboards have DVI ports? There's really no need for it, yet there are so many brand new boards that have DVI ports. There are even brand new boards with 9-pin serial ports on them ffs... It seems like some motherboard makers have run into a conundrum of not having enough things to stick on the rear I/O, so they just throw something else in, because...

I don't have an issue with standards, but as you agreed, we're not always using the best standard and something it seems like certain standards only exists because there wasn't even a discussion about it and everyone followed Company X (which often is Intel).

I don't think I need to re-count the mini-ITX board ratio with SO-DIMMs compared to regular boards, maybe you misunderstood what I tried to say. As in, there's one mini-ITX boards with SO-DIMMs to every 25 of any other type of motherboards.

The LS-120 was also much better than regular floppy drives, but way too expensive, just as ZIP drives, at least compared to regular floppy drives. Then again, back in those days computers were a lot more expensive than they are now as well...
Posted on Reply
#8
CheapMeat
"Mirkoskji said:
i didn't knew there was a limitation in total throughput of so-dimm, good to know.. as of the discussion about expensivness of the solution, they are expensive only because they aren't extensively adopted. it costs virtually nothing to put higher frequency dram chips on a so-dimm module and extend the production. but i would think that for low end, itx or not-overclockable motherboads, so-dimm could be a good solution.
Plus they already are a standard. Memory producers put out way more so-dimm modules every year for the consumer market than normal dimm's.
I've honestly found nothing that says there's a throughput issue with SODIMMs for consumers. I can't find anything pointing this out. Often, so SODIMMs are also double stacked, so density is still there. Again, you need the pins for ECC & Registered/Buffered ICs and their SMDs, but we don't get that anyway. I can't find anything about a performance penalty against SODIMM vs DIMM non-server.

DDR4 DIMMs have 288 pins, DDR4 SODIMMs have 260 pins, and DDR3 DIMMs have 240 pins, DDR3 SODIMMs have 204 pins.



It's probably the same kind of myth that went around that ECC has a performance hit; it does not. Or probably people conflating ECC with Registered/Buffered; two separate things that don't need to be together (or each other) but often are.

https://www.techspot.com/article/845-ddr3-ram-vs-ecc-memory/


Like others mentioned, it's ALREADY a standard. This is purely one of those things that no one questions, so no one bothers and if done, it would be like "hmm, why wasn't this the norm before".
Posted on Reply
#9
lexluthermiester
"TheLostSwede said:
Uhm, buddy, don't draw conclusions from thin air.
Thin air? Ok. Your whole post shows your depth of thought, reasoning and mild ignorance. I'll explain..
"TheLostSwede said:
I've been around long enough to remember it all. Yes, standards are there for a reason, but many won't die a graceful death either.
Ok, Let's observe your reasoning.
"TheLostSwede said:
Why does the latest generation of motherboards have DVI ports? There's really no need for it, yet there are so many brand new boards that have DVI ports.
That's easy to understand. DVI is still around as a standard because it's still relevant. The industrial/professional world upgrades it's motherboard more often than the displays they power. Additionally, as a general rule, two displays can be connected to a single DVI connector which is very useful. So where there is a demand, there will be a need for supply, more on that point in a moment.
"TheLostSwede said:
There are even brand new boards with 9-pin serial ports on them ffs... It seems like some motherboard makers have run into a conundrum of not having enough things to stick on the rear I/O, so they just throw something else in, because...
Oh yes, there still are 9pin serial ports. There is a great deal of industrial systems and machines which were built with that data interface. Why? Because it's a standard and easy to use port. And unlike USB, a properly built 9pin serial cable can be screwed into it's port and will not easily come loose or fall out due to vibrations. It is a durable, well known and easily built standard port still in wide-spread use today. That limited, but significant, market sector will not be going away anytime soon either. It is FAR more cost effective to continue to support an older standard than to replace machines and hardware which continue to be useful and relevant 20, 30 and even 40 years after the computers they were originally designed to interface with have become obsolete. For example, the lab I work at has a machine that is older than anyone who works with it. But it does it's job and as swiftly and efficiently as it did when it was brand new. Replacing it would literally cost multiple millions and for what, so we can upgrade a $1 data port? Can you see how laughable that is? And this is a VERY common example of situations that exist all over the world. Motherboard makers aren't "throwing" in a serial port willy-nilly.
"TheLostSwede said:
I don't have an issue with standards, but as you agreed, we're not always using the best standard and something it seems like certain standards only exists because there wasn't even a discussion about it and everyone followed Company X (which often is Intel).
This has little to nothing to do with Intel or any other big company. Industrial Standardization groups exist to make sure that big companies DON'T control everything, making sure that standards which are adopted have some logic and reason to them. And opinions about which standards are the best or most useful are as varied as the standards themselves.
"TheLostSwede said:
I don't think I need to re-count the mini-ITX board ratio with SO-DIMMs compared to regular boards, maybe you misunderstood what I tried to say. As in, there's one mini-ITX boards with SO-DIMMs to every 25 of any other type of motherboards.
Oh, it really seems you do. That ratio of 25/1 was WAY too high. There are far fewer boards with SODIMM's than that 25 to 1 ratio would suggest.
"TheLostSwede said:
The LS-120 was also much better than regular floppy drives, but way too expensive, just as ZIP drives, at least compared to regular floppy drives. Then again, back in those days computers were a lot more expensive than they are now as well...
Actually, I had both. The Zip drives, when connected by IDE, SCSI or even USB were MUCH faster than the LS-120. The JAZZ and Bernoulli drives before that, were also much faster than competing products from other companies. For a few years I stopped buying HDD's because JAZZ 1GB and then 2GB discs were fast[SCSI] and big enough for a complete OS installation, including software. Had a bunch of discs that had various OS installations on them. It was an amazing product. Ahh, good times.
"CheapMeat said:
I've honestly found nothing that says there's a throughput issue with SODIMMs for consumers. I can't find anything pointing this out.
How hard did you look? Try again..
"CheapMeat said:
Often, so SODIMMs are also double stacked, so density is still there. Again, you need the pins for ECC & Registered/Buffered ICs and their SMDs, but we don't get that anyway. I can't find anything about a performance penalty against SODIMM vs DIMM non-server.
It seems clear that you need to do a little more research on the subject. Everyone who understands how technology works knows that more data pathways=more data through-put at any given speed. This isn't just electronics principles at work, it's the laws of physics. But you are about to illustrate the point for me.
"CheapMeat said:
DDR4 DIMMs have 288 pins, DDR4 SODIMMs have 260 pins, and DDR3 DIMMs have 240 pins, DDR3 SODIMMs have 204 pins.
Yes, so DDR4's 288-260=28. That is about 10%. And that's about the performance difference between standard DIMM's and the SODIMM variant running at the same speed and timings. DDR3's 240-204=36. That's about 10% also. Shocking how those numbers correlate rather conveniently. Science is fun, isn't it?
"CheapMeat said:
Like others mentioned, it's ALREADY a standard.
Yes, a standard intended for mobile computing. It just so happens to fit compact micro-computing desktop needs as well, but at a performance loss.
"CheapMeat said:
This is purely one of those things that no one questions, so no one bothers and if done, it would be like "hmm, why wasn't this the norm before".
That's not true either, as evidenced by this discussion. The reason why is the aforementioned performance drop-off, so to anyone who actually understands the way the technology works, the question you posed answers itself.
Posted on Reply
#10
Mirkoskji
I was thinking that it isn't possible to have a reduced throughput, simply because so-dimm are 64bit modules and work at a certain speed with a certain latency. thoughput of memory is mathematical. frequency x data per clock x channel width. so that's it.
3800 mhz is 1900x2x64= 24,3gb/s theoretical per channel
Posted on Reply
#11
lexluthermiester
"Mirkoskji said:
I was thinking that it isn't possible to have a reduced throughput, simply because so-dimm are 64bit modules and work at a certain speed with a certain latency. throughput of memory is mathematical. frequency x data per clock x channel width. so that's it.
3800 mhz is 1900x2x64= 24,3gb/s theoretical per channel
It would be nice if it were that simple. Very nice. Unfortunately, it's a lot more complicated. And to be clear, I'm not expecting anyone to take my word for it. Everyone is encouraged to do research on this subject. Knowledge is power.
Posted on Reply
#12
CheapMeat
What a joke. You are just as much spectulating. Show me an actual report & test. Show me what we'd lose out on. Show me what we currently gain. You're making assumptions the same way and trying to appeal to authority. You're going to tell me, it's 100% fact, the pins, whether no-ECC/REgistered ICs are on the PCB or nothing, get ALL used exactly the same? So consumer DIMMs, without ECC & Registered ECCs are getting some magical throughput benefit because we have those extra fins freed up? You don't know JACK either, stop pretending. "Knowledge is power". Yep, being vague & broad definitely makes you smart.


Like I mentioned, you're probably the same fake enthusiast type that would have claimed ECC reduces performance or that RAM speed doesn't matter or that "xxxx" low MHz was "good enuff".

Look at these "dummies" over at ASRock......SODIMMs??? WHAT WERE THEY THINKING!!! THINK OF THE THROUGHPUT LOSSSSSS. Come on.

http://www.asrock.com/mb/Intel/X299E-ITXac/index.asp

Posted on Reply
#13
lexluthermiester
"CheapMeat said:
What a joke. You are just as much spectulating. Show me an actual report & test. Show me what we'd lose out on. Show me what we currently gain. You're making assumptions the same way and trying to appeal to authority. You're going to tell me, it's 100% fact, the pins, whether no-ECC/REgistered ICs are on the PCB or nothing, get ALL used exactly the same? So consumer DIMMs, without ECC & Registered ECCs are getting some magical throughput benefit because we have those extra fins freed up? You don't know JACK either, stop pretending. "Knowledge is power". Yep, being vague & broad definitely makes you smart.
If you don't want to take my word for it, OK, don't. Not gonna lose any sleep over some whiny twaddle from a troll. As I suggested above go do your own research. You've heard of Google, right? Or better yet, find a CPU that has motherboards supporting both types of ram, buy them, benchmark them and show us the results. See, I've done that before as a test for evaluation and validation of system upgrades. I know the differences because I saw them first hand.

"CheapMeat said:
Like I mentioned, you're probably the same fake enthusiast type that would have claimed ECC reduces performance or that RAM speed doesn't matter or that "xxxx" low MHz was "good enuff".
You're the only one talking about ECC. No one else has brought it up. But what seems clear here is that my response to you seems to have struck a nerve and wounded your pride.. You should really see to that.
Posted on Reply
#14
Mirkoskji
Isn't a simple aida64 cache bandwidth test enough? My single channel 16gb ddr4 2400 laptop scores around 18, 5 gb/s. Isn't it about right for the frequency and the bandwidth?
Posted on Reply
#15
lexluthermiester
AIDA64 Cache bandwidth only tests the CPU cache. Ram is untouched, AFAIK. I prefer to use a variety of benchmarks to find actual performance.

And I wish to qualify, the last time I tested DIMM's vs SODIMM's was in the DDR2 gen Intel CPU gen. So it's been a while. However, there shouldn't really be a serious difference with DDR3 and DDR4. It would be VERY interesting to see TPU do a review testing the current differences.
Posted on Reply
#16
Mirkoskji
So the first line is not a memory bandwidth test?
Posted on Reply
#17
lexluthermiester
"Mirkoskji said:
So the first line is not a memory bandwidth test?

Oh gotcha, I see what you're doing.

Now show those same tests, DIMM vs SODIMM. Make a video and throw it up on YouTube and make sure you show the hardware being tested. You think I'm wrong? Prove up or shut up.
Posted on Reply
#18
Mirkoskji
I won't. And I don't think you're wrong... Im ignorant so I was only asking.
Posted on Reply
#19
lexluthermiester
"Mirkoskji said:
I won't. And I don't think you're wrong... Im ignorant so I was only asking.
Fair enough.
Posted on Reply
Add your own comment