16gb SODIMMS, When will they be here?

Thanks for saying it: we can use an 'Not Liked' button here... lol...

 

DDR4 is still atleast a year away on laptops on paper. in reality, we are looking at atleast 1.5 years. till then these should do just fine.
Cant wait to see Sager NP9570 | Ivy -E with 64GB ram

 

And? How is that relevant to what we talking about today.

 

They're almost here:

See:
AnandTech | I'M Intelligent Memory to release 16GB Unregistered DDR3 Modules

 

But not compatible with current Intel chipsets... Nice one Intel.

" However I am told that currently there is a fundamental non-fixable issue on all Intel processors (except Avoton and Rangeley, other Silvermont (BayTrail) is affected) that means that these dies are not recognised. In their specifications for Ivy Bridge-E, Intel do state that 8Gb packages are supported (link, page 10), however this apparently has not been the case so far and I'M is working with motherboard manufacturers to further pin down this issue."

 

As expected, I'm tired of this behavior from Intel. Walking the fine line of not looking like **** and "not supporting things" so we have to "upgrade" generation after generation so they can keep milking us like cash cows.

~Aeny

 

I like this behavior from Intel.

Having to upgrade to a platform that the components were designed to work together instead of hoping your customers don't notice the glitches and setbacks trying to keep old/dead technology alive is what makes me an Intel supporter.

Nobody forces anyone to upgrade. If you need the new features/performance; a new platform is the least of my concern (I would have bought a complete system anyway: no point having a frankenbuild when you can have a working/known system and it's replacement ready to takes its place as soon as it is proven faster and as stable.

 

Back to the built-in obsolescence of the '80. But don't you love capitalism?

Where have we heard this old backward reasoning before? Consume! Consume! Consume! (They Live? )

It is burned into our brains but the manufacturers then extricate themselves of culpability by claiming no one is forcing you to consume. Really? Is that what ythey say to convince themselfs at night so tjhey can sleep?

Its relavent in that it creates an artificial limitation. The need needn't come before the invention. 150 years ago we didn't need to fly. But somebody invented a way and look where we are today.

Computers are the epidomy of technology; and technology moves forward not backwards. Nor does it stagnate.

 

Krane,

See:
https://www.google.ca/#q=epitome


As you state; moving forward means leaving old junk behind. Not carrying it with us until we finally realize it has weighed us down so much we're bent over (like, AMD).

 

So you like it when Intel doesn't follow the spec? That doesn't even make sense. And based on that quote, their own spec says it supports the 8Gb packages but doesn't. So are you embracing the fact that Intel faltered on their on specification or the JEDEC spec? Or both?

 

I like them following the spec, but reality reigns supreme.



...

 

And yet they work fine on AMD hardware!

Sent from my XT1049 using Tapatalk

 

With AMD hardware; sure they'll work (just taking your word for it, atm) - but will still be below what an Intel platform can output, imo.

 

How can it be below when it doesn't even work with Intel? The article stated it works with AMD, that's all we have to go by at the moment. You're the one that posted the article and not even sure you read it!

" I’M have verified the modules on AMD FX processors (FX-6300 and FX-8320 on 990FX/AM3+) as well as AMD's 760G and A75 (FM2 socket) chipsets. I was forwarded the following screenshot of two of these modules in an MSI FM2-A75MA-P33 motherboard, which is a dual DRAM module motherboard using the FM2 socket for Trinity APUs:"

 

This is a pretty crappy move on Intel's part. I have a feeling this was just an invention to provide incentive to upgrade to their latest platform since each new generation now seems to offer less and less reasons to upgrade.

 

I'm talking about productivity: real, actual work done.

The Intel CPU+RAM combo is still much productive than the AMD CPU+RAM combo - even if the AMD platforms can take more RAM, that doesn't mean it will be utilized as effectively as an Intel platform would.


All you guys asking for backwards compatibility is kind of funny: that's not how we move forward.

 

"Lacking backwards compatibility" is a funny way of saying "doesn't conform to specifications."

 

Not really, pci express does not work with pci because it is it's own specification to replace pci.

 

It would depend on the workload. If it's CPU-heavy and RAM-moderate, the Intel option probably would be better. But if you have a workload that's been heavily RAM-limited, or you could speed it up significantly by putting it all in RAM instead of on-disk with these memory modules, it's quite possible that the AMD CPU + additional RAM combo would lead to better performance, even if the CPU itself is slower in single-threaded tasks. It all depends on where your bottleneck is. TBH, I don't follow your logic in your posts - there's no benefit for Intel customers in that their CPUs don't support this amount of memory and they'd have to upgrade (or switch to AMD) to be able to use it. I really don't see how this can be spun as an advantage for Intel.

As for myself, I'm happy that AMD has support for this and thus has a competitive advantage. Those have been far too rare for AMD in recent years. So if this leads to a few extra AMD sales, good for them. Sure, it'd be better if AMD, Intel, and VIA processors all supported this. But if only one of them can support it, it's probably for the better that it's AMD. Even if that means I can't upgrade to 64 GB in my desktop.

 

What I thought I was conveying was that an Intel i7-4xxx platform with 64GB RAM will still outperform an AMD platform with 96 or even 128GB RAM installed.

Sure the workload will decide who 'wins' in the above hypothetical comparison, but I'm talking about my real world workloads (of course) not benchmarks, or gaming or other synthetic numbers that show theoretical advantages.


When Intel has processors that are over 20% to 60% faster (yeah; synthetics) than what AMD can do with almost double the TDP ratings, it doesn't matter how much RAM that lower end system can support.

The platform/cpu/OS/RAM balance will still favor Intel. That is their advantage.

When AMD seems to be catching up to them... Intel will introduce DDR4, 16GB SoDimm support and other advantages (including lower power draw) and still put AMD back in it's place again.

And while everyone complains that Intel isn't making things compatible or even 'to specs' - everyone that uses an Intel platform will still be getting the best performance possible (for that time period).


For servers, AMD may have a chance (except for power draw, again) - for workstations and my workflows; AMD hasn't been viable for decades.

 

Processors don't load up each bit of data they need from memory. Most of today's processors hit in L1 over 90% of the time. So increasing memory will only help in software that's written so poorly that it suffers frequent cache misses and is so large that it needs to fit in more than 32GB (because for less than that, you can get away with 4x8GB modules).

These modules will only be put to use in systems where you use software that needs more than 32GB of data. You aren't going to solder such modules onto Ultrabook motherboards. So if we consider only such systems, then software written for them is going to be professional software. And we all know that Intel processors will outperform AMD on every level on such systems. There's no software that you can "speed up" simply by using a ton of system memory.

 

That's exactly what they do: load each bit of data from memory, which loads it from the storage subsystem.

If you have a workflow that has such a small data set that 90% can fit in the L1 cache; consider yourself lucky.


My workflow is in the TB's per session - having an affordable system with 2TB+ of RAM would be what I'm wishing for (for a few years now).

 

Present -- and I'm not alone. Though these requirements may not be mainstream, those that require workstation computing can certainly put to immediate use upgrades such as these. My only limitation to RAM is the hardware's technology, and my pocketbook.

Not necessarily "speed up" but you can increase its overall efficiency. RAM is the backbone of video editing as Vram is to 3D modeling; so the more you can utilize, the more efficient your editing experience will be.

 

The entire program doesn't need to fit in L1 because the entire program isn't what is being operated upon at any given instance. Also, cores do NOT have direct access to main memory. ALL accesses to the pipeline are through their respective first level caches.

If a CPU is doing image processing which usually comprises of MAC operations, it will load up a certain number of pixels ("certain number" depends upon the width of the superscalar architecture) from the cache and issue them in the pipeline. Then it will write the result back. Usually, multiple operations are performed on the same set of data. A 1080p 32 bit image is about 8MB in size which means half of it can fit in most L3 caches. You can have a hundred hour long movie that is ten terabytes in size. It doesn't matter. Are you going to work on it? No. You work on one image at a time. You load up half the image into L3 and ten rows into L1 (for example). Then you process them. Each time you load one of those ten rows, you hit in L1. Then on the 11th time, you miss and load up the next ten (or how many ever fit in L1). When you are done with the current image, you miss in all three caches. Then you load up the second half, rinse and repeat.

That is how all processors behave for large datasets. Larger number of cache hits are ESPECIALLY true in large datasets. If you have a 2TB dataset that you are frequently suffering cache misses on, you are doing something very, very wrong. Main memory also acts like a cache. Also, with all due respect, I call BS on a TB dataset that needs to be loaded into RAM at runtime. ANY software you purchase, will be written such that small amounts of data will be loaded into cache. The rest will be stored to swap. For some reason, people have the notion that swap is always bad. That isn't true. Swap is bad if you frequently need 5GB of data and you only have 8GB of memory, leaving just about enough for the OS to do its work. If you have a huge dataset you want to work on, there is no way you need terabytes all at the same time. No program works like that. Unless you wrote yours, to which I'll strongly suggest you read a good programming book. If you have datasets numbering in the terabytes, you need a server because memory isn't your bottleneck anymore.

 

I'm curious, how much did you pay for that laptop of yours? The M6700? You add a $400 memory module (three rather, to go over 32GB) and you are looking at a very nice server that will be much faster than your laptop. You could buy a nice ultrabook and carry it around and purchase a 4G dongle that lets you access the server from wherever you wish.

Like I said above, even for GPUs, data is loaded up as batches. You don't need all that data at once because the GPU can't process all of it at once. If you load up LESS than the CPU/GPU whatever oyu use can process, then you are inefficient. If you load up 10GB and only 2MB cn be processed each time, you have a huge amount of data sitting unused. In servers, you have huge amounts of memory (>128GB) or on supercomputers because on large SMPs, you have a much larger processing elements to GB ratio than on a laptop (where it is usually 1 or lower). On large servers, it can be around over 2 if you have 2 4 socket mother boards with 20 cores per socket and 128GB of memory.

 

32gb SODIMMS, When will they be here?

 

Your point was lost if you were trying to make one.

 

OK, the bolded parts, clarifying that you're talking about your workloads, is what was missing before. Intel may well still be better for your workloads, and may (and probably will) be better in most cases. But there will be some cases where the extra RAM will make a difference. It's not such a far-fetched theoretical advantage that it will never happen in practice. An FX-8350 with 64 GB of RAM and a well-multithreaded workload could conceivably be a better choice than a Core i7 with 32 GB of RAM. It just depends on the situation.

And while I can't argue that for your workloads AMD hasn't been viable for decades, that hasn't been the case for the industry at large. AMD was by and large the superior option during most of the Pentium 4 and Pentium D timeframe. They certainly were a viable option for most people. Yes, there probably were workloads where the Pentium 4 was better - IIRC there's at least one instruction on the later Pentium 4's that can execute twice as many times per cycle as on the Core architecture, so there could be workloads that are quicker on the Pentium 4 than both the Core, and possibly AMD64 CPUs as well (it's been a few years, so I don't remember what the particular instruction was). But for many workloads, AMD was certainly a viable option during the circa 2001 - 2006 timeframe, and in many cases superior.

As for myself, I've used Intel CPUs in all the primary computers I've used since 1999, including a Northwood Pentium 4 (in retrospect, it probably would've made sense to go AMD then, but I didn't know that at the time). So I haven't been much of an AMD customer, and wouldn't have been at all had they not bought ATI. But the impression of your posts in this thread has been unrealistically pro-Intel.

 

I am Pro 'anything that works for me' and that has been Intel for the last few decades.

AMD processors have shown to not work with all the programs I use. Not even on a brand new $2K notebook (circa 2000/2001).

Of course I'm going to be biased to my workloads: would make no sense to be biased to anything else.

I have the same bias against anything but Intel chipsets too (NVidia; I'm looking at you). When the drivers for a M/B can't be made to work with a new version of Windows; the platform is dead (NForce 'junk'). Do I have to say 'for me' again?

I need (and test - before I buy/upgrade all my workstations) products that are stable, upgradeable in terms of O/S and currently available drivers, cost effective and above all; offer me real and tangible performance/productivity improvements for my varied workflows. Over the last 4+ decades; the only platform consistently providing that has been Intel.

So; call me biased. So what? I know what works for me. Because I found out the hard way (and with my own $$$$$$$).

Unrealistic is thinking that I am Pro Intel because I'm a fanboy.

No, I'm Pro Intel because their products keep me and a few dozen employees in business - and have for a very long time.

 

Right, do remember though L1 consists of instruction cache and data cache. Depends also on your data set too. For data that is only ever used once it is a waste of cache space and just causes pollution to have it cached in the CPU. If you ever want to get a real appreciation for what CPU cache does though try running with L1,L2 and L3 disabled. I think you only ever need to try that once.

 

Dufus, I don't doubt for a second that L1, L2 and L3 are important (can't remember which cpu/platform I had tried that on; but you're right - it was slow like an 8086 processor used to be).

When you're converting TB's worth of RAW images (with presets) though; there is mostly nothing to store in the caches (relative to the size of the task session). This is where RAM is king.

 

Not sure about that. Intel didn't support it earlier because it was not available. BIOS seems buggy from some manufacturers and other than Asus fixing their own X79 BIOSes, they don't seem interested in doing much about.

IIRC Micron are producing monolithic 8Gb DDR3 chips so more 16GB modules to come perhaps.

 

has anyone run 2x 16gb on any laptop right now?

 

I've not even seen any on sale anywhere... They're almost impossible to find..

 

Only ones available are ECC So-DIMMs and no laptop is capable of running them. Mainly for niche server use.

I'M (Intelligent Memory) was supposed to make unregistered 16GB So-DIMM's and have them specc'ed at their website: DRAM Memory Modules | 16 Gigabyte DDR3 unbuffered DIMMs | Intelligent Memory

but I haven't heard of them available anywhere for purchase, and I doubt if Intel tested with 16GB So-DIMM modules so they may not even work, as a matter of fact I believe someone linked to an article showing that they wouldn't be supported by current Intel chipsets anyhow.

 

16GB DDR3 SODIMMs are proving to be a lot more elusive than 8GB DDR2 SODIMMs ever were!

 

IIRC one can email [email protected] for buying of 16GB SODIMM but a couple of things to know...

1. Probably going to have to modify your own BIOS to have a chance of them working.
2. No confirmation if even with a modified BIOS it will work with the CPU.

A lot of money spent if it does not work.

 

Aha. It's so cool to see a thread which took your attention and find out that you already posted here 2 years ago and you were completely right about everything!

So... I't, I't, I told ya! I't, I't, I told ya! So was I a pessimist or realist, ha?

 

James D,

Wait for it...













Here's your virtual High 5!



Okay, you were right, but still a pessimist.

 

Just to be clear here, the memphis link posted is for 16GB DDR3, not DDR4.

 

well memphis needs to test out those modules on a few w series laptops. I've got a few w520, a w530 and a w540 I'd like to upgrade to 64gb ram. And I know 2 people in the office who would like to do this also. Probably expense it since I'll be using testing out some work stuff too.

 

sadly it wont work for the same reason I cant do it to my M6700 and Elitebook W's and Zbooks. Intel messed up on the memory handler on the CPU's and they wont accept the 16GB single modules unless your are running a Xeon in a laptop ( of which I have tried ) meaning that for my work purposes I will have to get a lot of new kit when it is supported by intel as AMD's APU's wont work too well for me in my line of work.

 

Did try modifying a 4GB DIMM module SPD to 32GB module some time ago and it posted alright with a modified BIOS but really need to test with the real thing, well 16GB DIMM not 32GB. However as I don't need that much memory and due to cost it probably wont happen.

 

Thanks for testing it out and letting us know. I was hoping since Ark Intel has been wrong so many time on max memory, 16gb sodimm might work.
Now I need to find a laptop with a xeon socket and microde to take a xeon.

Interesting work Dufus.

 

Clevo makes a workstation that uses desktop Xeon chips. and we also use the same ECC SODIMMS in the new Mac Pro trashcan.

 

Yup although the Clevo's P570WM3 is now discontinued, you can still get them for now... Xeon chips are an option with them..

 

Some guys are working over fixing the BIOS so it can support 16GB DIMMs on this very machine, with the RAM in question.

 

I remember when testing 1x4gb+1x8gb on my signature laptop that my bios reported the right amount, it could simply not access the upper 4gb which meant that loading windows would bluescreen instantly because from my research back then it likes to fill out the upper memory with kernel stuff. I could boot Linux but ended up in a kernel panic each time I tried to utilise more than 8GB ram. It seems that reporting and using memory are two completely separate things. So be careful possibly wasting money on this.
~Aeny

 

That could be the case Aeny, if it were just a few dollars I'd try the real thing out of curiosity but $600+ for 2 modules is too much.