ssd's as ram??

i was contemplating getting some new ram for my sager, but now im thinking about saving up for a g3 ssd and using the pagefile as ram.
i know most ssd users turn off page filing to decrease ssd degradation and that ssd's although flash chips are limited by the sata2 interface.
im just thinking if anyone would know the difference in performance for say a ram hungry program like adobe premiere pro cs5 and after effects cs5.

 

The type of flash memory (NAND) used in SSDs has a completely different i/o interface and it doesn't have access times anywhere near as fast as RAM but this type of memory is non-volatile which means it can retain data even when it is unpowered. RAM is volatile so if you cut power to RAM you lose the data temporarily stored in it.

I wouldn't turn off the system pagefile. Intel for instance qualify their non enterprise drives (10,000 cycle write endurance) as being able to tolerate 100gb of writes every day for 5 years. A single home user is never going to write that much with that frequency. Some of the more recent drives where the cells have lower write tolerances like Sandforce ones are 4,000 or 5,000 I think but they have proprietary technologies like DuraWrite and large dynamic spare areas to minimise wear.

Add to the fact that you can get ssds these days with a 3 to 5 year warranty as standard and you have a good deal of certainty that you can use the drive as you normally would. I had my last computer for 5 years (an Athlon 64, 1gb of RAM, 160gb hdd and a Radeon X1600 pro).

There are somethings you can and should turn off like disk defrag because they are pointless but other than that just use your drive. Its what you paid for right?

 

On a sidenote, Tony from OCZ said that their Vertex 2 uses the same flash memory as Intel G2. I'd think there would be no difference in write tolerance.

 

+1 to Hayte. SSD's are fast, but nowhere near as fast as true RAM.

To give you an example, my relatively "old" desktop system (Intel Core 2 Quad Q6600 CPU, 4GB DDR2 RAM @ 800Mhz) has RAM read speeds of about 3500 MB/s. The fastest SSD out there can max out at ~280MB/s read speeds... an entire order of magnitude slower than RAM.

So, RAM is always better than page file. My general rule is that you should always try to have so much RAM that your computer never needs to use the pagefile.

 

yes i understand this but have you actually tried measuring the difference in realworld terms? for example how much slower is a premiere pro cs5 1080p simple encode with a 840qm(or whatever cpu) with 4gb ram/4gb fast ssd pagefile vs 8gb ram?

 

Ahhh I wasn't sure. I vaguely remember someone saying (possibly Anand) that because of things like DuraWrite it was possible to use lower quality NAND and still get the same lifespan. Such things are predicated on usage patterns of the end user however.

I've never stop watched anything I do day to day so I honestly couldn't say. I find it unlikely that you would really use that much RAM though. I don't do video/image processing but I do audio design and work with massive, high resolution .wavs and ROMplers and I can't say I've ever used much more than 4gb of RAM.

I guess the first question is, are you currently running out of system memory with Adobe Premiere right now with your normal work routine? If not then is this really an issue you need to entertain? If you are running out of RAM then stick to the simplest solution - which is to either buy another stick of RAM or tidy up your workflow so that you are using the RAM you have more efficiently.

 

there is much disk activity during some of my ppro encodes which could indicate accessing of the pagefile. however with my current setup i still feel the cpu is more of the bottleneck. i asked the original question because im kinda on a budget since i also plan to get a new cpu and a g3 or equivalrnt ssd, and i would like to know how much of a performance hit id get by using a 4gig ssd vs 4gig additional ram. as in many benchmarks, 'performance magnitudes' dont really apply in some realworld uses hence the query.

 

True Anandtech did say it was possible to use lower quality NAND. Maybe some of the other manufacturers actually do, I don't know.

On another note, even though the NANDs in Intel G2 may have been rated at '100gb of writes every day for 5 years', I wouldn't be surprised if significant performance degradation happens way before that.

 

Sorry, I haven't ever measured that. I'm not a benchmarker, so I don't really run measurements. The only reason I know that my RAM reads at 3500MB/s is because I recently ran Memtest86+ to test for bad RAM, which also tells you reads speeds during its tests.

I'd imagine that RAM wouldn't really have much of an effect for 1080p encodes. Encoding is a CPU-bound activity. The way to influence encode times is to get a more powerful processor.

That disk activity is most likely writing the output of the encodes to disk, and not page file access. Even if it is pagefile access, you are still heavily CPU-bound, and not storage-system bound. An encode can be as fast as single-digit or low double-digit MB/s. So any bottlenecking would occur at the CPU.

Have you ever checked to see how much RAM you are using during an Adobe Premier Pro encode?

Again, I think that you are right - the bottlenecking for this particular activity is CPU-bound. I don't think that going with 4GB SSD + 4GB RAM, vs 8GB RAM would really make a difference in pure encode speeds.

 

i agree with all your observations. as i mentioned in a certain pm to a certain someone, ram seems to be more of a bottleneck in desktop cpu's where you have more horsepower to work with.
but in a laptop environment with our relatively weaker cpu's it seems that the ram will be waiting on the cpu to crunch data instead of being able to constantly feed it with new data.
i am not so sure but i believe audio processing could be the same too (cpu bound).
my choices for my upgrade was fast ssd/more ram/mid cpu vs fast ssd/hi cpu. it seems ill stick to the later even if it may turn out more expensive.

 

Well, I think that bottlenecking doesn't really depend on desktop vs. laptop CPU's. The difference between a desktop and laptop CPU isn't really all that great, as long as you are comparing CPU's that are relatively within the same class.

The larger factor that will influence where the bottleneck is the kind of activity that you are doing. If you are encoding video, then the bottleneck is almost always going to be at the CPU, regardless of how fast that CPU is. If you are playing games, then the bottleneck is generally at the GPU. If you are loading applications, then the bottleneck is generally at the disk system.

I don't know what you consider to be "more" or "less" RAM... but in general, 4GB of RAM is sufficient for most peoples' use of their laptops. The only situations where you would use more than 4GB of RAM on a laptop is if they do a lot of photo/video work, if they run a lot of virtual machines, or if they run very specific server applications that need lots of RAM (e.g. Oracle DB). But for 99% of the people out there, they will never even come close to using all 4GB of RAM, so a 4GB config will be enough.

 

but i do believe the performance levels between desktops and laptops are quite great, especially if you are considering top tier performance levels.
the fastest laptop cpu right now is the 940sm which i believe is a far cry from the fastest performance desktop cpu the hexacore 980x.nhowever i believe the 940xm is even underpowered compared to its desktop i7 counterparts.
in my current setup it definitely the cpu bottlenecking since its always at 100% when video editing/encoding. ram seems near maxed out too but it really seems like the cpu holding things back.
i just get confused when everyone on the adoe forums always suggest more ram first

 

trvelbug,

Just want to jump in and state what makes the biggest difference in actual throughput performance in descending order (biggest bang/buck first):

1) Platform - High clock speeds don't mean anything - instructions/clock is what makes one platform peform better over another.

2) Assuming the same (highest end) platform, then the CPU.

3) Again assuming the two above are a 'given', then the OS makes the biggest difference (x64 vs. x86, but also Win7 vs. XP/Vista too).

4) With all of the above 'optimized', then RAM makes the biggest difference. My advice here? Always max out your system's RAM as soon as you possibly can: not only will you be more productive, but your next 'upgrade' can be put off for considerably longer because your currrent system is operating at its peak (and it would take a very big platform/system 'jump' to get appreciably above where it would be at).

5) Get the fastest HD possible - even if it means you have to upgrade 3 times in two years or even more often (because new/better models keep coming out).

6) Get/Use the latest programs that take full advantage of all the resources you have at your disposal (if you've optimised everything as above).


7) Finally, maintain your system(s) so that they do operate at their peak for as long as possible. I highly recommend ccleaner, MSE and PerfectDisk for those tasks.


With a notebook, the first two items are a given (or, at least not economically feasible to 'upgrade'). The things that are left are: getting a good/current 64bit OS like Win7x64, maximizing your RAM, and ensuring your HD is current tech (no, it does not need to be an SSD). Assuming that the programs you are running are also a 'given', ensure that your system is maintained with proper defragging (online and offline) and it is as free of junk files as possible.

Notice that the order of recommendations does not change (desktop/notebook makes no difference).

Hope this clears things for you a little more.

 

thanks tiller

however the choice i have is fast ssd/more ram/mid cpu vs fast ssd/hi cpu.
in a laptop setting it would seem that the latter is advisable for my purposes.
in a desktop setting the former may be more reasonable because the midrange cpu maybe powerful enough to handle my tasks.
thats the point i was trying to drive at.

 

I'd say the former is better for both desktop and laptop for the majority of Adobe products. Regardless of CPU speed, no matter how far overclocked the CPU may be, RAM is the bottleneck for these absurdly large programs with 8GB requirements.

In my mindset, the program itself is loaded in memory, but there's not enough extra space to hold the file/data you're working with. Due to the lack of space, the program constantly retrieves from the hard drive, making you think the hard drive is the bottleneck or that the processor isn't fast enough to keep up (because the processor is wasting cycles writing RAM to disk, clearing the RAM, grabbing new data, placing it in RAM, then working on it.. However, if there had been enough RAM to hold the data to begin with, the hard drive wouldn't get hit nearly as often.

 

Yes but that wont have anything to do with the frequency of writes. Thats more like a garbage collection/TRIM/Sandforce dynamic spare area thing?

Yes. I get disk activity too when recording (memory to disk) and when loading .wavs, plugins etc. into memory etc and its all perfectly normal for data that is destined to end up as a file on a disk drive. When you run out of physical memory you will totally know it. Your computer will literally grind to a halt, you won't be able to stream playback etc. I normally bust the cpu load limit much sooner than running out of RAM and that results in jerky mouse cursor, stuttering, glitchy playback and you just can't do anything.

 

woofer00,

I find that this is not strictly an Adobe issue: all programs benefit first with more RAM then with a faster HD.

I am in total agreement with you though... especially in the case with Adobe apps, the RAM makes a bigger difference than anything faster in the storage side (other than another 48 - 96 GB of RAM used as a RAM Drive for the Adobe Scratch disks...)

 

actually this is what i see most of the time in the adobe forums.
the thing with cs5 programs is they are both cpu and ram bound. inkow its one of the reasons why professionals will use 8 raid setups and up to 24gig of ram. data is constantly read from the array (seperate for target and destination), cached in ram to be fed to the fast cpu.
i was dead set on getting the former, fast ssd/more ram/mid gpu because of this.
but i just cant get over the cpu being more of a bottleneck in a laptop environment. but i do see the logic of both sides of the argument.
i think its hard for people who have not used these programs to see how ram intensive they are and that old notions of performance per part may not necessarily hold true for these apps.
but then again, im back to square one

 

True, but Adobe is the only brand that makes my brain snap to going RAM RAM RAM at me.
Every other brand is product-specific.

That's a ridiculously overbuilt system. If the data is fed to the RAM so fast and the RAM is just holding it and waiting for the CPU to munch through the data, the system is incredibly inefficient for all the money that's been thrown at it. That's an example of a severe and ridiculous CPU bottleneck. Such an excess of RAM. I can't imagine that more than 12GB is ever actually touched or that much of that hard drive bandwidth is used to shave more than a few seconds off a workload.

However that problem isn't the one you've got. From your sig spec it looks like you're lacking on RAM and disk bandwidth, not CPU speed. Laptop CPUs are slower, but not necessarily the cause of bottlenecks. Frankly, CS5 is the biggest bottleneck for multicore systems. The suite is not optimized to use available cores across the board. A faster pipe is and has always been more important than more pipes for all but a handful of features.

In any case, you should check CPU, RAM, and HDD activity when you're doing a typical run. If your RAM is at 100% and there's a lot of I/O off the HDD while your CPU looks bored, RAM capacity is an issue. An SSD would help as well, but wouldn't give as much band for the buck in purely calculation. The SSD wouldn't get hit much for small workloads. If you're having issues reading/writing data for workloads fast enough, that's a different story.

 

actually its not (overbuilt) and is acutally the norm for pros and semi pros who work with resolutions much much higher than 1080p such as 4k, which is 6 x the resolution of 1080p. check out some of their comments here:
Adobe Creative Suite Forum at DVinfo.net
a lot of these people do commercials and tv shows.
of course i dont do that kind of thing, but you do get the resources needed by these programs to run at the higher levels of use.

also cs5 is optimized to use all cores , id say its one of the most optimized multi threaded programs i know of and you can see some examples of how 12 cores benefit from this in the link i posted .
when im working on an edit cpu is almost near 100% and ram is up there too but not maxed out. when encoding cpu is always at 100% on all 8 cores and ram usage is high but not 100%. hdd is also very active but have not noted its level of activity

 

The DVInfo people builds are even beefier than what you cited, included dual hexacores, so the bottleneck I pointed to disappears. I forgot that ad agencies tend to go dual-hexacore-xeon-only (btw, DVInfo builds are always a little over the top, almost always pointed at ad agency requirements and trying to shave off every last bit of time. just my opinion of em). If it weren't dual-xeon, I'd stand by saying it was overbuilt.

I hesitate to accept that CS5 as a suite is optimized for multi-threading. Products within the suite like AE and Premiere are, and have gotten substantially better than previous editions, but the suite as a whole is not (Nor should it, waste of time to multi-thread many of em, imo). I think a big part of the improvement is the jump to 64-bit memory access.
Pieces of Photoshop are, but not as a whole (note: the only piece of cs5 I really use is PS, so it's a sticking point for me)

 

trvelbug

I know you have been exploring your options pretty vigorously in the forums, but at this point your options are pretty clear.

1. Live with the your present performance, it is what it is, be happy with it’s present performance.
2. Upgrade slowly, Start with 8 gigs of ram, then when money permits move up to an SSD. I would not spend on an upgraded CPU unless you can get a really really good deal on a 920 XM or better yet a 940XM. If you are comfortable with a ES type of CPU that is one way to cut the cost in half of an OEM. As is the price to move up to a top of the line CPU is quite high, and with SB around the corner not sure it makes sense to upgrade that part of your system.
3. Wait till SB and upgrade your laptop to a new system.

Your system is not that old, so the upgrades to ram and SSD makes the most sense. Those upgrades could be carried over to an upgraded SB laptop as well. If you wait a bit, I would think by the end of the year you may be able to get 8 gigs of 1600 DDR3 so that any benefit from upgrading to that type of ram could be used if you decide on a SB laptop later on. (Even though 1333 would most likely work on SB as well)

This is how I would be thinking if I was in your situation.

 

just a fyi, i think the biggest update in the cs5 suite is premiere pro performance wise. making it a 64bit only application and adding the new mercury engine with nvidia gpu acceleratrion really makes it much better from the slower, less stable cs4 version. the only downside to cs5 seems is that most plugins that used to work for cs4 have to be rewritten for the new version.
in terms of features ppro remains almost the same but after effects and photoshop get notable additions like rotobrush and content aware fill.

 

thanks for the suggestion. i will definitely get a fast ssd, in fact if the g3's were available now at a price im comfortable with id snap a 400gb one in a heartbeat.
at present however, i have modified my workflow to use less demanding applications like power director on video edits when time is of importance. i am actually quite happy with my system but just like most other enthusiast, we always want more out of our machines. but since this is more of a hobby than a business theres a certain leeway to be had.
common prudence would indeed dictate that a ram upgrade, being much less would be the next step. however, i dont see the additional ram being of much use if the bottleneck is indeed the processor. i also dont see any benefit to other tasks such as gaming, normal office work, surfing, multitasking with an additional 4gig. of course the best thing for me is to testbuy the ram and see if i do get a boost, but that is not possible where i live.
my optimal config would be a fast g3 ssd, my present momentus xt for storage, my bd on external, and a fast 920/940 cpu (when prices drop) and 8gig ram.
if money becomes a little tight my choices would be limited to the 840qm with ram (and the ssd of course) or the 920/940 without the ram and use the pagefile of the ssd.
i really dont think i wanna go sb yet as battery has never been an issue for me, and upgrading to a 920 by that time would still prove cheaper than buying a new system all together.

 

another option im looking at is to get an extra xt instead of the g3. this would allow me to get both the ram upgrade and hi cpu.
im thinking that a combination of both xt's plus the 8 gig ram (with superfetch) could be as good a performance boost rather than an xt plus ssd plus hi cpu with 4gig ram.
ive been extremely satisified with the performance of my xt and i can see that it could be a good data drive especially if you will be accessing the same data over and over again as is the case in video editing.

 

The XT doesn't cache data.

I would recommend the two XT's though over an SSD.

But be sure to short stroke/partition it properly to take full advantage of the drives.

If you do opt for an SSD and an XT, I would set up the SSD as the 'scratch disk' for Adobe products and the XT as the boot/OS/app/data drive. This will give you a performance boost (instead of just a 'snappiness' upgrade).

To partition a HD to make it perform as consistently (fast) as possible even while doing work:

See:
http://forum.notebookreview.com/har...-hitachi-7k500-benchmark-setup-specifics.html

 

are you sure about this? a link would be useful.
i just ram previewed and rendered an old sequence in after effects and i was surprised to see how much faster it was now. i was particularly surprised with ram preview.
however eventhe render, which used to take 20 minutes or less was done in less than 10 minutes.
now i should state that a few weeks ago, adobe did update its opengl engine and this could be a direct effect of that. however in ram preview i had open gl turned off, and from the looks of it there was minimal hardrive thrashing as compared to before the xt.
i use 1gig pagefile and im thinking that maybe the xt cached the pagefile into the NAND effectively giving me an additional 1gig of ram for ae ram preview to work with.
on a side note, i was utterly surprised by the render speed increase. i use to hate using opengl render acceleration because it was actually slower than cpu rendering. not anymore it seems. this, coupled by the theoretical caching of the pagefile plus much lower cpu usage and more effective gpu usage has really breathed life into my laptop with regards to AE use.

 

See I don't get the 920/940XM because they are gimped desktop cpus. They are high TDP and absolute battery life killers plus they cost so much that you could easily put together a much faster desktop PC for much less money.

The cooling you need to cram into a notebook chasis just to keep the thing from melting down increases size and weight so everything about these processors is defeating the purpose of a notebook which is about portability and good power consumption.

I own a notebook with a Clarksfield i7 740M but I have to admit these things aren't great mobile cpus. I'm a user that can and frequently does utilize all 4 cores because I use music production software on a day to day basis. Even still, making music on my old PC was never an impediment because I had a dual core Athlon X2 3800+. And before that it wasn't an impediment when I had a single core 1.6ghz Pentium 4. I was just far far more efficient back then because I had to be and the software was more efficient too because it had to be. Nowadays you can get a single softsynth that could completely max out a modern dual core cpu with a 3 note chord because the output can be oversampled to such a ridiculous, impractical degree i.e. uhe ACE. All modulator signals can have sampling rates up to 96khz which is insane.

I never used to have multiple windows open because I couldn't get away with it. Now I can so I have tonnes of windows open at the same time. I never used to use convolution reverb in realtime because its very hard hitting on cpu load. Now I can get away with it so I run it in realtime. My songs aren't really any more complicated or better because of the technology nor am I that much more productive so I have to say that for creative types - work habits are just as important if not more important than pc specs. Its crazy the kind of stuff we try to do in realtime now that we used to do offline.

To me, Intel's real mobile cpus are Arrandale because those things look like they were designed for notebooks. Clarksfield looks like it was designed for desktops and then miniaturized for notebooks.

 

with the travelling, lifestyle and work i have i cannot be tied down to a desktop system.
the reason i got my sager is precisely to have a workstation level laptop that would double also as my entertainment center when need be. im always on power so battery never was an issue for me.
i also live in the Philippines, a very warm, tropical country. but i can count in one hand the number of times my cpu has hit 80*c and this was not in an aircon room and with the cpu at 100% and no cooler.
imho there are very few 15inchers that can handle a 920/940 and the sager is one of them. in fact some guys the in the sager forum have o/ced the 920 to 3.6hz quad with a cooler and ht disabled.
just a fyi post

 

I'm not saying that these computers have no use but they are extremely niche. I don't need long battery life or extreme portability either unlike some people but I do need limited range in the studio whilst plugged in, kind of like you I suppose. i.e. I want to be able to move my PC onto my keyboard stands, set up guitar in a nice sounding room and be able to have my PC right next to me so I can play and track simultaneously without having to run back and forth to hit record/stop all the time.

But a 940XM over an 840QM? Its double the price and you don't get anywhere near double the computer. And for what an 840QM costs are you really getting more benefit over a 740QM? And I sometimes ask the question to myself - do I really need a 740QM when theres alot I can still do to work more efficiently?

I'd still want to move over to an Arrandale if I could. For watching the odd movie on the couch I'd much rather have a quieter laptop with switchable graphics that I can unplug and sit on my lap without burning my nuts. And I feel like I could use that PC for work no problems because its still vastly quicker than what I was using only a few days ago (a 5 year old AMD Athlon X2 3800+ from the old socket 939 days with 2gb of RAM and 160gb hd).

Like I said, alot of the heavy cpu usage I have can be mitigated in full by rendering and performing offline signal processing instead of trying to crunch stuff like reverb, equalization all in realtime. For some reason, quite alot of people have an aversion to doing that.

 

they are extremely niche indeed. and video editing and compositing is one area where you would want the most processor as you can have becasue it will be utilized, usually to the max.
however if you saw my previus posts regarding after effects and open gl, adobe seems to have done a very good job in optimizing their opengl engine with the last patch.

 

trvelbug,

I don't have a link for you (have to run out soon...) but the 4GB nand is just too small to be caching data (and still have an overall positive effect on the performance of the platters).

Not only that, but data is usually much bigger than 2MB anyways. With a single nand chip (hint: also single channel), it can only deliver about 20-30MB/s sustained/sequential read speed. A normal HD can do this up to 5 times faster. If it caches data - the performance would actually decrease - not increase as has been proven for the XT. As the files get bigger (over 10MB), this discrepency gets worse for the single channel nand vs. the mechanical platters.

If we can think of our platters as a FAT32 partition (hope you're running NTFS though!), what the nand does is cache the File Allocation Table of the HD along with small *.exe, *.dll, *.sys and other 'program' files that are accessed over and over, many times per Windows session.

This is what makes the small size of the nand make sense - the small files are searched/found 'instantly' compared to locating/finding/reading them on the platters and also, the rest of the data on the platters is 'mapped' so that the normal route of asking the slow platters where a specific file resides is bypassed and instead, the R/W heads are simply told where to go directly to get the needed files - this is why the XT feels so very much like an SSD.

 

tiller,

i believe your explanation does hold for the after effect 'caching' effect i mentioned.
for one, ae does not deal with large media files like ppro. usually you edit sequences, and in my case they are only 20 seconds long. they are composed of small elements. also if i understand ae right, it is also a bunch of plugins; aka smaller programs that execute different protocols to deliver the desired output.
with this in mind it really does seem that the xt loaded these elements into the nand, or were present in the pagefile that was cached to the nand. i dont think the opengl update alone could account for the increase in ram previews and rendering speeds.
i did a test run on ppro a few hours ago and i did not see any increase in performance. again i think this is because the media that i use there is too big to fit into the NAND.

 

trvelbug,

I didn't explicitly state it before but I would be hard pressed to believe that the XT caches the (entire) pagefile - it is just too big and would drastically slow down the system as I explained before. It can cache certain locations/specifc items inside the pagefile though... I'm pretty sure.

Remember that we cannot write directly to the nand on the XT - it has it's own algorithms that picks and chooses what to write not only to increase performance as much as possible, but also to reduce the wearing out of the cells too (with a single chip, it is much harder to 'over provision' for wear-leveling). With this in mind, it would make no sense for the nand's algorithm to cache the pagefile (and be left for dead in a matter of days).

Cheers!

 

dead in a matter of days - a little dramatic tiller
but i do get your point. it does seem however that it did indeed cache some elements of the sequence and maybe a few plugins. but as you suggest probably not the whole 1gig pagefile

 

trvelbug,

It may sound dramatic, but it is very close to the truth.

With no over-provisioning, there is very little leeway to use the full 4GB chip and expect it to last in a 'normal' O/S initiated (vs. user initiated) r/w environment.

When an Intel G1 can die in as little as a few weeks (and with 10 Channel 80GB capacity) when it's used in the wrong environment (servers), then a single chip, single channel nand chip can easily die in a few days for 'normal' O/S r/w's.

This is the main reason we can't 'see' the nand - what we can't touch, we can't hurt.

 

i dunno, im no ssd expert. but i do remember reading that the 4gb nand on the xt is enterprise level.
i also remember reading some posts stating that ssd's are not as volatile as some make them out to be.

 

I gotta disagree with you here. I think you are SEVERELY exaggerating the fact that NAND chips have limited write cycles.

The situation you mention (single-chip, single-channel NAND chip) never happens. Nobody engineers a drive like that. It is a contrived scenario specifically intended to demonstrate how to kill a NAND chip as fast as possible. I could easily come up with an equally ridiculous and unrealistic scenario talking about how mechanical hard drives are horrible, because they die within minutes if you put them in an oven or repeatedly drop them onto concrete.

SSD's have been out for a while now. I challenge you to name ONE case, or find ONE thread on the internet discussing how an SSD hit its write cycle limit, and became a read-only drive... not a discussion on theoretical limits, and not a situation where the entire drive died for some unknown reason like all computer hardware has tendencies to do. Can you find an actual confirmed and documented case where the drive was used as it was intended to be used, the NAND chips hit their limit on the number of allowable writes, and reverted to a read-only mode?

 

Just as an FYI - my desktop recently got hit with a power surge, and fried 2x1GB RAM chips out of my total 4x1GB RAM sticks. I am now running on the remaining 2x1GB chips (2GB total) while I am waiting for replacements on the bad RAM.

Running a game that uses >2GB RAM now shows VERY noticeable stutter and slowdowns when loading textures, when it never had this problem before. I can check Task Manager, and confirm that I am in fact using every bit of available physical RAM, and paging out to disk. This slowdown / stutter happens, even though I am running an Intel X25-M 80GB SSD in my desktop system.

Long story short: If you page out to disk, you will experience slowdown - regardless of whether that disk is a mechanical hard disk or an SSD. Always have enough physical RAM so that your system never needs to page out to disk.

 

kent
thanks for you observations re the bad ram etc, but dont game textures load to the discrete gpu's ram and not actual system ram?

 

I don't think you're following the conversation here...

The Seagate Momentus XT Hybrid has a single chip, single channel nand device with no overprovisioning included in it's spec's.

And, search for the G1 Intel's that, when used in the wrong scenarios, died in a matter of weeks.


No, I am not exaggerating about this at all .

<!--EndFragment-->

 

Perhaps it isn't loading textures, then... perhaps it is loading game assets, or models, or somethign else.

Regardless of WHAT it is loading, I can definitively say that a pagefile running on an SSD is no replacement for true RAM.

 

thanks kent. thats the most definitive real world answer ive gotten so far, although it isnt a perfect assessment since the damaged ram could be adding as much to the sluggishness as the pagefile access itself.

 

Well, I took the damaged RAM out. The RAM I have in the system right now has passed about 7 straight hours of MemTest86+ without error. Since my desktop is just a regular desktop system, I'm using non-ECC RAM. If the RAM was bad, then it would simply freeze up or crash my system. It would not correct itself with a performance penalty.

I see that you're running a Seagate Momentus XT. That isn't all that much slower than an SSD. If you want to see how your system reacts to paging out to disk, then just remove 2GB of RAM from your system and run off of your remaining 2GB. Paging to a Seagate Momentus XT may be better than a 7200rpm mechanical hard drive, but it is still going to be painfully slow compared to having true physical RAM. Having a real SSD may be slightly better than your Momentus XT, but it's still going to be noticeably slow when paging.

Once I experienced this, I quickly remembered why I came up with a rule several years ago that I will always have enough physical RAM in my system to the point where I never page out to disk.

 

yes actually thats where the discussion with tiller started from. with my momentus xt my ram previews and renders are much faster.
my contention was that the pagefile (or at least a large part of it) was cached in the NAND, since nothing else had changed in my system except an opengl adobe update - but that should not have affected my ram previews.
the point is moot however since i caved in and ordered 8gig ram. hopefully i get to sell my 4gig to recoup some of the cost.