backtop

Print E-mail del.icio.us 78 comment(s) - last by hstewarth.. on Apr 12 at 5:31 PM

Peak growths for DRAM memory have occured with major Windows launches - Courtesy SEC Marketing
How Vista will affect your next memory purchase...

With Windows Vista’s anticipated launch later this year, a concern on everyone’s mind is how Vista will tax existing PC platforms.  Although the new graphical user interface will require DirectX 9 support, and Intel G965 (or better) graphic accelerators, the real question mark in everyone’s minds is where DRAM requirements will head for Windows Vista.  Baseline Vista offerings will require 512MB of DRAM just to install, with a 1GB recommendation -- but is there more to this story?
 
Integrated graphics from ATI, Intel, and NVIDIA all use shared memory architectures. This means that even though the graphics core is on the motherboard Northbridge, the graphics controller accesses memory from the system main memory.  Low end, PCIe 3D accelerations from ATI, and NVIDIA also use shared memory support, using in excess of 256MB of system DRAM in exchange for a dirt cheap graphics accelerator.  On these systems the Vista recommendation for 512MB is not acceptable as a significant amount of main memory is consumed by the graphics accelerators.
 
Furthermore, Windows Vista will come with a new feature called Superfetch.  With Windows XP, Microsoft included a feature called Prefetch: a dynamic service that preemptively loads files into the pagefile in order to speed up application load time.  Superfetch advances further in two steps.  Step one is to build profiles of frequently used applications and store those profiles into the pagefile, and system memory.  Step two is to pool NAND and all other available memory to move as much of the pagefile as possible off the hard drive and onto the solid state memory.  As a result, anyone with a heavy usage profile will have a significant portion of their system memory dedicated to application data.  
 
At IDF we recently had the opportunity to talk to Tom Trill, Samsung Semiconductor's Director of DRAM Marketing.  An interesting point Trill mentioned to us is that system integrators generally spend 6-8% of the system cost on memory. Retail DDR2-667 crossed over into the $80 USD per gigabyte range a few months ago with the price for system integrators hovering around $60.  AMD and Intel both have new processors expected to utilize DDR2-800 before the Q4 launch of Windows Vista. By conservative estimates, we can expect to see the average system integrator bundle new computers with 1GB of DDR2-667 by the end of this year.
 
Samsung’s internal research recently published a figure claiming that the average PC system (including SI, OEM and home built computers) averages 871MB of DRAM in 2005, up from 620MB the year before.  The DRAM industry has traditionally seen large growth around the launches of Windows operating system such as Windows 95, Windows 98, and Windows XP.  With large growth come large economies of scale, and ultimately lower prices for DRAM are on the horizon.  Furthermore, with cheaper DRAM prices, system integrators are free to integrate more memory into the magic 6-8% budget. With such favorable trends, seeing 2GB of memory as a standard in every PC by the end of this year would be of no surprise to us at all.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
.
By hans007 on 3/20/2006 7:12:12 AM , Rating: 2
64bit doesnt mean it will take twice as much memory.

in general all that means is a dword is 64bit and pointers are 64bit. the size of say an interger in a piece of code will likely still be up to the programmer (though , on most systems and "int" is 32bits , it may still be 32bits on a 64bit system , it all depends really).

that said, sure memory use will go up.

i also would have to say that it probably all depends onwhat vista features you turn on. with aero glass probably more memory will need to be used than now. with it off, probably will make no difference compared to xp.

i have run builds of vista on a p4 2.0 512mb gf4mx420 machine at my work (software development company) and well, it seems to be as fast , if not faster than winxp just in general. IE seemed to load faster, the gui was "snappier" etc. and a p4 2.0 with 512mb of ram is not even remotely modern.




RE: .
By TomZ on 3/20/2006 10:33:55 AM , Rating: 2
quote:
in general all that means is a dword is 64bit and pointers are 64bit.

No - pointers are 64-bit, but integers/dwords are still 32-bit.

I haven't seen any reasons yet that 64-bit anything will require any more memory than 32-bit. To be sure, 64-bit software has the ability to address more memory, but that doesn't mean that it necessarily will.


RE: .
By masher2 (blog) on 3/20/2006 12:30:27 PM , Rating: 2
> "No - pointers are 64-bit, but integers/dwords are still 32-bit. "

No, integers are 64 bit, and dwords are 32-bit. This is the ANSI default-- bytes and words are fixed size units, whereas integers depend on the base allocation unit. In 16-bit programming an "int" was only 16 bits.

Still, I believe most if not all development tools will have an option to force ints to 32-bit.

> "I haven't seen any reasons yet that 64-bit anything will require any more memory than 32-bit."

64 bit code *will* require more memory. In the worst case, twice as much...in the best case, a negligible increase. But it will always be larger.


RE: .
By stephenbrooks on 3/20/2006 2:46:45 PM , Rating: 2
Yep, a friend and I compared the size of various default windows utilities between XP and XP 64-bit. There was something like a 30% increase in size going to 64-bit executables.

And now Vista is going to load all those larger executables into RAM. Great. Maybe I might actually want to use my RAM for something other than trying to hide the bloat of such programs as Acrobat Reader or Office, which would load much faster if they were programmed with a "load modules only when needed" approach.


RE: .
By stephenbrooks on 3/20/2006 2:57:22 PM , Rating: 2
RE: .
By TomZ on 3/20/2006 3:48:40 PM , Rating: 2
Are we still talking about Windows?

This page from Microsoft on the topic of 64-bit Windows pretty clearly states "only pointers expand to 64 bits; all other basic data types (integer and long) remain 32 bits in length."

http://msdn.microsoft.com/library/ default.asp?url=...


RE: .
By masher2 (blog) on 3/20/2006 4:26:42 PM , Rating: 2
> "This page from Microsoft on the topic of 64-bit Windows pretty clearly states "only pointers expand to 64 bits"

As already said, this is true only in the LLP64 programming model. Use ILP or LP and its no longer true. And even when LLP is chosen, you still must remember that pointers are also stored as data. So if pointers expands, data storage requirements do as well.

So its more than just code that increases under Win64; data does as well.


RE: .
By TomZ on 3/20/2006 4:34:13 PM , Rating: 2
Microsoft only talks about supporting LLP64. Are you saying that people will be using other models with Win64, or that it is even possible?


RE: .
By masher2 (blog) on 3/20/2006 4:55:54 PM , Rating: 2
Sure it is-- its up to the compiler, not the OS. And just think about it...if you couldn't use 64 bit data values when you needed them, there wouldn't be much purpose in 64-bit computing. Sure you get more address space, but if that was the only reason for it, we'd just keep using PAE forever.

Obviously, when interfacing to the Win64 API, you need to adhere to their model. So if you're not using LLP64, you'll need to do some casting or otherwise ensure alignment of data types.


RE: .
By drebo on 3/20/2006 7:31:17 PM , Rating: 2
If 64-bit memory addresses are to be used, pointers MUST be 64-bits long.

Why? Because pointers store memory addresses. You cannot fit a 64-bit memory address into a 32-bit pointer.

Unless, of course, you're talking about relative memory addresses, similar to the way MIPS handles jumps. But, we're not.


RE: .
By masher2 (blog) on 3/20/2006 11:39:08 PM , Rating: 2
> "If 64-bit memory addresses are to be used, pointers MUST be 64-bits long."

There's no confusion here; we've already stated just this several times. The issue is over the size of *other* fundamental data types. In LLP64, ints and longs remain at 32-bits.


RE: .
By hstewarth on 4/12/2006 5:31:11 PM , Rating: 2
Well with 64 bit points, this also means that the cache elements are like 64 bit.

Also I was at .Net development course and Microsoft staff and stated that .Net internal uses 64 bit for intergers. Vista is largely .Net so this means that a lot of code will 64 #s also. Also if there are any arrays these, values will be 64 bit.

I would be that because of 64 bit, there will be allow more fatter code and more importantly, the data access will also be more. So Vista is likely to required more memory based on fact that it mostly designed for 64bit. There are other elmenets that can explained it. IE over time software gets bigger and bigger. Same for OS's - just look at Windows.


Cause and Effect
By TomZ on 3/19/2006 9:23:52 PM , Rating: 2
This article asserts a correlation between higher average memory per system expected in 2006 and the release of Vista later in the year. But how have they demonstrated any cause and effect? Maybe they are just happening at the same time?

My opinion is that the reason that memory amounts are increasing is that the memory costs are decreasing. Therefore, when you are shopping for computers now, you see typical configurations of 512MB or 1GB offered, mainly because memory this size doesn't cost much.

The designers of Vista are well aware of this trend, and plan the memory requirements of the OS accordingly. But I really don't think that end users, even corporate buyers, are really proactively specifying today's computer purchases with Vista in mind. Many companies are just now transitioning to WinXP.




RE: Cause and Effect
By joust on 3/19/2006 10:43:00 PM , Rating: 2
TomZ, you imply there is no causation between the releases of windows and an increase in memory sold. To prove yourself correct, you would need to show some other factors. You did not do that; I find that argument unconvincing. Consumers want the new version windows to run well, so they buy computers with more memory to compensate for the new OS's demands.

I agree there are certainly other factors that affect memory increases, but I find it very hard to believe windows has little to no impact on the memory market. Did you mean to say that memory prices fell, coincidentally, at the same time as windows was released? Even if that were correct, it may have been in response to windows being released, hence windows would still be in the chain of causality.

I agree with your second point -- people buy more ram because it's cheaper, however, I don't believe it to be the exclusive reason; windows releases have a roll to play.

I think people and corporations don't proactively buy for the next OS because they know they will buy another computer in three or four years; Why pay the added expense now when you'll buy a new vista-ready computer later down the line?


RE: Cause and Effect
By TomZ on 3/19/2006 11:04:46 PM , Rating: 2
quote:
To prove yourself correct, you would need to show some other factors.

I already mentioned that I think the main reason is that memory prices continually drop. I also suggested, and you also stated, that most computer purchases don't anticipate the requirements of the next OS they'll load in 2-4 years. Therefore, I don't buy the argument that OS releases are driving memory increases.


RE: Cause and Effect
By joust on 3/20/2006 12:22:47 AM , Rating: 2
I never said that OS releases are "driving" memory increases. I said they have "a roll" or "have an impact" in them. Your post seemed to support the notion that windows has absolutely no roll in memory amounts.

The article stating, the "industry has traditionally seen large growth around the launches of Windows..." is not the same as saying, "growth in anticipation of the launches of Windows..."

I think in this context (where we're discussing a ~10 year timeperiod) , "around" doesn't mean 2-4 years before a launch.

There is no doubt that memory prices continually dropping causes more people to buy memory. But what causes the memory prices to drop? Mass quantities. What industry to produce mass quantities? lots of customers. What gets lots of customers? A new release of windows. This is the "chain of causality" I mentioned.

To be clear, ram prices drop FASTER (and, more memory is sold) when an OS is out. (without windows I don't know what it would do). Ultimately, we're BOTH correct; I'm just explaining the spikes in the graph, you're explaining the intermediate logic (which I mostly accepted).


RE: Cause and Effect
By mircea on 3/20/2006 12:08:53 PM , Rating: 2
I still think that the release of an OS does influence and in a big way the price and sales of RAM. It's like the theory of relativity. The price of RAM will always go down at a steady rate but the release of an OS creates a disturbance, a valley in the smooth slope of the RAM pricing and production so things precipitate around the OS release. Yeah it would have gotten there at some point even without a OS release but the release rushes this a bit. Not much but it hapens non-the-less (???)


RE: Cause and Effect
By Egglick on 3/19/2006 11:14:11 PM , Rating: 2
I agree with Tom actually. The article is implying that buyers and manufacturers are preemptively increaseing memory amounts to prepare for Vista, and Tom is saying that he believes otherwise. Because of that, he doesn't have to prove himself right, but rather (if you're supporting the stance of the article), you have to prove him wrong.

The majority of Joe Schmoe users out there don't know that another version of Windows is coming out, nor do they have any idea what it's memory requirements will be.

I think that the drop in DRAM pricing is the main catalyst. Manufacturers are willing to spend a certain amount on RAM, and other manufacturers follow suit because they don't want their systems to look underpowered by comparison.


RE: Cause and Effect
By joust on 3/20/2006 12:36:32 AM , Rating: 2
quote:
Because of that, he doesn't have to prove himself right, but rather (if you're supporting the stance of the article), you have to prove him wrong.


There is no way for me to prove he's wrong, (and I'm not trying to, I actually agree with him) because he hadn't presented any other factors. I can't prove that Windows is the ONLY factor (and I wasn't making that assertion). I was just asking for more facts.

The one factor he did seem to support was that Ram amounts increase because of lower prices; the article even accepts that, saying, "With large growth come large economies of scale, and ultimately lower prices for DRAM are on the horizon."

Here's the argument:
windows vista --> more memory demanded --> mass production --> more memory, same price --> more memory sold