How to build fastest pc with no moving parts (except cooling fans)

[beyondcurrent]

This post is to generate ideas on building a super pc.

When computing power of that which is available on the market is not enough or just cannot be accepted, most users consider building or upgrading PCs choosing quality motherboards, CPUs, RAM, and over-clocking, etc. which should be a given. As far as OCing, I am an advocate (with proper extreme cooling of course) because I prefer to have things not being limited in their capabilites. With speed increasing everday in motherboards, CPUs, and ram, users are still being limited by the hard drive (device drives in general) that stores the operating system which is why alterantive device drives is important as well when it comes to building a SUPER PC beyond the conventional.

My main focus is on I/O's for the device drives, basically on flash memory with the best performance speed/efficiency ratio to replace the current movable-parts hard drives. Up to date there are ways to boot operaring systems from flash memory. Microsoft however is currently working with manufacturers to implement windows to be bootable from usb on a universal level without having to fiddle around. This still would not meet the criteria because of the USB technocolgy itself being limited in several ways, bandwidth, CPU overhead etc. Current hard drives on the market, even the fastest, use IDE/ATA, USB 2.0, FIREWIRE, and the under-rated SCSI. I am a fan of ULTRA SCSI. The SCSI transfer rate may not be as fast as USB 2.0 or the new FIREWIRE 800, but when dealing with a handful of devices connected into a single PC, SCSI outperforms the others with no overhead. SCSI doesn't degrade CPU performance as much as USB and FIREWIRE (which makes my wonder why Mac is thinking about abandoning SCSI). So why not use an UTRA SCSI hard drive currently available on the market?

-Well basically, the HDs available have moving parts. In my opinion current PC devices (CD/DVD ROM, SUPER FLOPPY ZIP, HDs, etc.) are prehistoric, given the fact that their technology is based on wear and tear movable parts. It's not that I am looking for a quiet, anti-shock system with increase life usage or anything (noises can be tolerated, devices can be replaced if damaged by shock or wear and tear,) but the way I see it is that even with performance improvements on these hard drives with movable parts they still are limited to the speed in which they can rotate, which is an analog mechanism downside. So no matter how fast the transfer rates are improved (by connection, increase RPM speed, etc.) the moving mechanism would be the limiting factor. This drawback can be compared to reasons why Over Clockers do their thing. It's not that (in some cases) OCers can't afford a better CPU or motherboard because when they do actually upgrade to the best ones, they still OC the new hardware even knowing that those upgrades outperform their previous hardware even if left untouched. It's just the fact that something is capable of doing more that makes them do this.

As for my project, I am developing ideas to build a super PC with only solid state used for computing, reading, writing, and transferring (which is why the cooling fan(s) is is not counted, even though non-fan cooling exists but are not good enough for OCers). I am looking to have flash replace all devices with moving parts. Of course I am still going to have DVD Rom, maybe floppy, connected to the PC because software on the market is based on those devices.

So here are my questions:
1) Why is MAC considering to abandon SCSI?
2) Would an Ultra SCSI flash memory hard drive perform faster than a USB 2.0 or Firewall 800 flash hard drive?
(I know that the tranfer rates on USB 2.0 and firewire 800 is faster, but taking in account the efficiency of SCSI's command mapping capability)

Any other insights and ideas to add? Thanks.

 

[felixc]

SCSI has no advantages over SATA in a desktop PC application. With the Sata interface now being handled before the PCI bridges, the speeds limits will increase by another magnitude. The closer the drive is to the CPU and memory system, the quicker it will work. USB and Firewire are at the end of the food chain, with some extra software layers to slow them dowm even further.

 

[beyondcurrent]

Thanks Felixc. Now what if you had multiple SATA connected to the PC, when imformation be handled less efficiently than SCSI?

 

[beyondcurrent]

Thanks Felixc. Now what if you had multiple SATA devices connected to the PC, would imformation be handled less efficiently than SCSI?

 

[ceh4702]

They make IDE to flash adapters. However, they usually want to take over both IDE Channels.

Also I have heard talk about the speed of PCI-Express being good enough for a memory access device if enough Channels are used. If you had enough RAM, you might be able to load an operating system into RAM. Theoretically you could put an OS on a DOC (Disk-On-Chip).

Some embedded motherboards use this. However, you still need some swappable space. Windows XP needs like 350 megs of swappable Hard Drive Space for virtual RAM. Most DOC Chips are limited to 1 or 2 gig and sometimes only 512 Meg.

Some motherboard made by Via in the Mini-ITX form factor have flash memory and card bus interfaces right on the motherboard. Mini-ITX isn what I would call a powerhouse of computing power, but that shows that some other ideas are possible.

If you do not like my post feel free to point out your opinion or my errors.

 

[garebo]

I dont have quite the technical knowledge and expertise that some of you people have, so my question might be right out in left field, i dont know. Here it is anyway.
When i saw flash memory devices come out in the 2 to 4 mb size it make me wonder. Why wouldnt that technology move fast to replace hard drives? With no moving parts, a flash device would have an advantage right there.
So why havent they developed 100 gig flash hard drives, or flash devices that take the place of a hard drive?
A one gig compactflash card, for instance, isnt very large, and it seems as though it can be read and written to fairly quickly, so why not make a 100 or 200 gig flash drive?
What is preventing this, or is this a dumb question?


Good advice + great people = tek-tips

 

[beyondcurrent]

Thanks Ceh4702.

The problem with IDE flash adapters is that the flash performance is limited to the I/O of the IDE controller.

Ultra 320 SCSI Flash still seems to be the top choice if cost is not a problem with its sustained fast transfer rates and its separate processor. The next generation of SATA will probably beat U320 SCSI though. The current SATA is good as far as ecomony + performance. The problem with it is that it still drains the CPU and does not have command quenuing the way U320SCSI command mapping does.

As far as swappable space still being needed for large applications, this is another area that would benefit in having a solid state hard drive. Geting virtual memory to speeds closer to RAM would be possible with Flash through a high speed I/O. My ideal system would have Gbytes of RAM chips anyways so that Virtual memory wouldn't ever be needed. In case of huge applications when file swapping is needed, then having a solid state flash drive with high speed I/O would act almost like having additional RAM. See where I am trying to get to? Abandon mechanical components because of their limits.

As far as advanced technology, the US Army uses Ultra 320 SCSI Flash for their HDD because of the benefits: durability, can with stand shock, fast performance, longevity when using over and over again, etc.

As for me, I am a fan of efficiency, even if it costs more.
I am looking for the best ideas. So far I am a fan of using flash but am looking for a better I/O for the flash. Fibre Channel Flash?

Thanks for all input everyone.

 

[ceh4702]

I wonder if they make a SATA2 Flash adapter?

If you do not like my post feel free to point out your opinion or my errors.

 

[beyondcurrent]

Garebo, the thing is, manufacturers need to make functional technology affordable so they can market their products. As for me, that's why I'm working on a custom PC without limiting performance to affordablity. Anyways, with current flash on the market, you are able to replace standard mechanical HDDs. The prolem is that most of the flash devices in the market today would still be slower than the standard rotating HDD because they are limited by their I/O, ie: USB 2.0 and FW800. IDE, SATA, SCSI flash should be faster than standard movable HDD. I personally am trying to to find the fastest I/O to control flash interms of tranfer rate combined with not sacrificing CPU with mulitiple devices connected, sustained transferring, reading writing etc. Thanks for the questions and input Garebo.

 

[beyondcurrent]

ceh4702,
Even if a SATA2 Flash Adapter isn't available or hard to find I think it is poossible to have the flash connected to an a different type of adapter (ie: Flash to IDE) that can then be connected to another adapter (IDE then converted to SATA2) into the SATA2.

 

[ceh4702]

Well I suppose you could just use one of those server motherboards with 8 memory slots and just load it up with as much of the fastest RAM you could buy and load everything into the RAM and run it from memory on a dual processor board. However a lot of RAM would be expensive when you have to buy high capacity RAM with over 1 gig on each stick. Sooner or later though you have to put something in storage.

The SCSI interface and the right disk is important. This site in the UK claims to sell SCSI Flash devices with a 40MB Second Burst R/W speed. I could imagine that in a RAID Configuration with Ultra Wide SCSI.

http://www.reactivecomputers.com/flashdisks/k3mffd35ultrawidescsi.htm

Most of these devices are probably up there in price a bit.

If you do not like my post feel free to point out your opinion or my errors.

 

[beyondcurrent]

Garebo, another thing is that I don't think mechanically PC devices are going to disappear any time soon in the market. Manufacters of that sort would lose a huge sudden share of the market.
Thats why I have always been a fan of going beyond what is available when knowing something better is possible. Overclocker's syndrome.

 

[cdogg]

Main disadvantages of Flash:

A. Slower than magnetic - average transfer rates are less than 10MB/s, and you'll pay a pretty penny to get one that can go as high as 20MB/s. Flash really only shines when it's accessing data that is store sequentially. It is "dog slow" when it comes to scattered file access. And as we all know, a Windows OS drive becomes fragmented in just a short period of time. Expensive flash drives that have been placed in parallel and other configurations have been known to write up to 40MB/s, however.

B. Less reliable than magnetic - as an area is erased and overwritten, it becomes less and less capable of storing data for long periods of time. When brand new and erased/rewritten less than 100 or so times, the spec says flash can retain data for as long as 10 years. But there is a steep downcurve the more times you overwrite. Also, flash is just as sensible if not more sensible to shock than a magnetic hard drive.


Solid-State Solutions

This is still an extremely expensive option which is why it is not mainstream. Solid-state drives use a form of DRAM (usually SDRAM). They rely on a constant external power source when you turn off the system to prevent loss of data (usually through a lithium battery).

These puppies are extremely fast. They are not similar in any way to flash memory, which a lot of people tend to confuse. In terms of performance, a regular magnetic hard drive isn't even in the same ballpark - 'nuff said. If you can afford one, more power to you. I, in the meantime, will stick with magnetic until prices become reasonable (which should happen within the next 5-10 years). If you look at that tweak3d.net link I posted below, you'll see some of the other types of technologies that might make it to mainstream even sooner.

Some of the information above was pulled from the following. To learn more see:

http://www.diskonkey.net/documents/Performance_reliability.pdf

http://www.tweak3d.net/articles/solidstate/

beyondcurrent,
I don't necessarily agree with your viewpoint on overclocking necessarily, though I understand where you're coming from. Saying that a device is "limited in their capabilities" isn't exactly true if you take reliability into account. They are designed to run at their full potential without sacrificing reliability and lifespan. Overclocking is just a way of saying you don't mind stressing the other factors involved...

~cdogg
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884

 

[cdogg]

Sorry ya'll, but I spent some time on that last post and didn't mean for it to seem that I was repeating what was already said in the last couple posts...

~cdogg
[tab][navy]For general rules and guidelines to get better answers, click here:[/navy] faq219-2884

 

[ceh4702]

What I see as a good use for flash drives is for storing code that is retrieved and then run in memory like one of those Linux Firewalls. An embedded product might work well if it basically is retrieved and runs in memory.

I am still using IDE Hard drives and probably will continue doing for a while. However they keep making components run a lot quiter.

If you do not like my post feel free to point out your opinion or my errors.

 

[garebo]

I figured there had to be a good reason that flash memory didnt translate into flash hard drives. Unless, of course, some breakthrough comes along that would mitigate the problems that exist with flash memory now.
This is certainly interesting, i have to check out the links cdogg posted!


Good advice + great people = tek-tips

 

[beyondcurrent]

Thanks for the in depth information Cdogg.

By the way sorry for not being clearer when I mentioned myself looking into Flash for reasons other than anti-shock avaoidance. I meant I meant blunt shock and vibrational shock to be exact for example if something were to drop. Just like a standard incandescent light bulb vs an LED bulb, solid state devices (FLASH HDDs) could tolerate more impact. Of course solid state and magnetic HDDs can be damaged by electrostatic discharge shock. The benefits of solid state to me is just a plus as far as being quiet, consuming less energy, emitting less heat, and so on. Im not gearing towards the solid state technology for those reasons. If an item were to break then it can be replaced. If it makes noise, there are ways to make it quieter (if such low noise would even be nuisance!). I'm basically looking at solid state for its performance. Price doesn't matter. I'm not looking for an inferior technology that would be more cost effective. Not looking for economical reasoning to save money. Just a way to improve a system as a whole (performance wise) by implementing current technology available, be it expensive or not, because its possible. Just looking for the best of the best.

 

[beyondcurrent]

Thanks again Cdog.
Yes you are correct with everything you stated taken in context. Now I forgot to reply to the Overclocking view you had. You are correct for that as well.
You mentioned:
"Saying that a device is "limited in their capabilities" isn't exactly true if you take reliability into account. They are designed to run at their full potential without sacrificing reliability and lifespan. Overclocking is just a way of saying you don't mind stressing the other factors involved... "

The way I see it is that if something can be modified and any negative end result can be corrected then enhancing the object is a plus. Its true that devices in general are not limited given that they are tested to perform at their best without sacrificing reliablity and lifespan. I never did say that devices in general have limited capabilities if taking safety, reliability, and lifespan were taken into account. So that it is true that general PC hardware in the market is limited in its capabilities but are designed to perform at there best within a certain given range. So I agree with you.

 

[beyondcurrent]

As for the atatement I made:

"The way I see it is that if something can be modified and any negative end result can be corrected then enhancing the object is a plus"

An example would be overclocking. The negative would obviuosly be excess heat that would decrease system longevity, trigger crashes, and create safety hazard. The problem can be corrected if you do it seriously. Every OCer would most likely add extra cooling to have temperatures fall within a recommended range. Extreme cooling could even be added by injecting moist-free cool air to further benefit the system therefore adding another plus. The negative to that plus would be extra power consumption.consumption.

What I look for in an ideal super computer (for myself to own) is one that is efficient in performance and has no basic configurable limits. It's of no significance to me if the system would have a shorter lifespan or costs more, although I do look for stability.

 

[beyondcurrent]

Cdog, about the solid-state drives you mentioned, yes they are very fast. To me though, they seem dangerous as far as keeping important saved memory protected. I know lithium batteries have a lot of power to keep the data saved but the fact that a device needs constant power in order to retain data is very risky so I'm not too into that. It would however be an excellant primary HDD to have to store applications to enable instant access. If battery fails in any case, application programs could simply be reinstalled in no time, without the typically lag time on currently on-the-market PCs (of cousre program installing speed would be limited to its transferring source, ie: CD ROM drive. The solution to the limiting factor would be to have a copy of the primary solid- state drive backed up on another solid-state drive, enabling quick program installs). I wouldn't use solid-state HDDs to store important data. If there were flash devices that are as fast as solid-state drives then things would be simpler.