Slashdot Mirror

I may buy/build a workstation soon that will need the capability of holding a large amount of RAM (say 32GB - calculations requiring a lot of memory but not a huge amount of CPU) at some point down the road (I can probably put off fully populating the memory for a while). I came across this post, which seems to say that motherboards for DDR2 will allow more DIMMs (16 2GB sticks is a lot cheaper than 8 4GB sticks right now, at least for DDR). It is talking about DDR2 with Opterons. Is there a launch date for DDR2 on Opteron? Is the capacity actually greater with DDR2? Is DDR likely to become scarce down the road, causing DDR2 to be a cheaper option for future expansion? Any opinions are appreciated (I haven't had an excuse to buy hardware in a long time, so I haven't kept up on such things).

Sidenote: Yes, I am aware of the iWill DK88 (16 DIMMs DDR) - anybody have any experience with it (especially with Linux)?

this is SO lastyear for /.

http://www.tomshardware.com/2005/09/07/can_gigabyt e/page8.html
look at 2 cards in raid0

the I-RAM2 is gonna come out Q1 2006, 300 G/ps sata, stick in a 5.25 drive bay,perhaps with external power, 8 slots and ddr2.

http://vr-zone.com/?i=3052
(the good stuff so we do not crash the vr-zone server)
Gigabyte for the first time has unveiled the specifications of their upcoming i-RAM 2 during HKEPC Tech-day. The i-RAM 2 will be using DDR2 memories instead of DDR1 and the data transfer interface is SATA 3Gb/s double of the current SATA 1.5 Gb/s. It will be externally housed with by means of an external case or fitted into the 5.25" bay with eSATA interfaces. The number of memory slots are expected to double to 8 and you can have up to 16GB max memory. The expected announcement of i-RAM 2 is targeted for February next year.

http://www.cluboverclocker.com/reviews/hard_drives /RAM_Drive/Gigabyte/I-RAM/index.htm
review of I-RAM1

so assuming that I-RAM2 scales as well as I-RAM1,2 of these should get 500 G/sec or so.not bad for a 64gb drive.

now WHY are people doing this, and not just adding memory?well, assuming i have a nifty MB such as
http://www.iwill.net/product_2.asp?p_id=102

i can have up to 32 gb of memory on the MB if we look at the o/s selection guide for windows...

32-bit 64-bit
XP Pro 4 GB / 1-2 CPUs 128 GB / 1-2 CPUs

2003, Standard 4 GB / 1-4 CPUs 32 GB / 1-4 CPUs
2003, Enterprise 32 GB / 1-8 CPUs 1 terabyte / 1-8 CPUs
2003 SP1, Ent 64 GB / 1-8 CPUs 1 terabyte / 1-8 CPUs

2003, Datacenter 64 GB / 1-32 CPUs 1 terabyte / 1-64 CPUs
2003 SP1, Data 128 GB/ 1-32 CPU 1 terabyte / 1-64 CPUs

we see that in the 32 bit land, 2003E is the only thing that starts to come close to using that amount of memory. and its 1-2k (USD)

so assuming i'm made of money, and have stuffed my box with $5k of memory and i want to get a little more boost out of it, i can go 15k rpm SAS (with extra cost, and heatloads) or i can go 6x I-RAM2, raid 0. 96 gig of space at 1200 G/s throughput.

other programs like, mail servers, proxies, web servers, and Photoshop requires lots of fast swap space, and most people are unable to use more than 4 gigs of memory, (those running 32 bit windows).

Its a nice product, IF you have maxed out the 4gb limit for your o/s and/or you have to have ubber fast swap/cache.

back when i was at a CDN this would have been a no brainier to raid up in an external chassis (I-RAM1) and connect to a 16/24 port sata controller. We would not have cared that it was only as big as the large sata drive and cost as much as a new car,it got the job done.

There is a place for silicon drives that do not cost as much as a car, if you do not see the need for yourself, great. I would rather wait for I-RAM2 to come out and use it where it makes sense,wherever i need a huge honken bit of fast space.

My backup server is a good example. I have a few boxes dumping deltas as fast as they can across a couple gigE ports to a backup server, so that server can toss them to LTO.

my backups are only as fast as the slowest device, having a I-RAM2 raid 0, works for me. the other boxes are waiting on their local/network issues, not on my backup servers slow large raid. (the x-fer to the slow raid can happen whenever, as long as i have enough space in the I-RAM2 array for the next dump.

right now disk is the slow part in the local box,this addresses the issue

No, you might as well go all the way: http://www.iwill.net/product_2.asp?p_id=90&sp=Y

The difference is no longer an order of magnitude or two even for the SMP boxes that you are talking about. Googling a bit, I have found these reasonably priced Opteron boxes. These are variations of the same opteron v40z boxes that Sun sells. Quad Opteron boxes w/ hot swap power supplies and hot swap hard drives w/ ECC memory for reasonably cheap for the chassis.

I've never seen a terabyte of ram on a single domain but the
very high end sun boxes will allow you to use either 576 or 288 gigs of ram on a single box, so I will use that as the base. The maximum in SMP mode for the opteron boxes you can homebrew is probably one based on the iwill motherboard which supports 64 gigs, which is now grabbing the midrange line of the sun server series.

In fact HP makes the ProLiant DL585 with opteron chips. They will sell you this machine w/ up to 128 gigabytes of ram, redundant roms, redundant power supplies, hot swappable hard drives, ECC memory.

Its not quite at the high end server sun range but the opteron chip and its use in machines built by traditional "big iron" builders is not really an order or magnitude off. Its really about a factor of 5 from the really high end (aka $1m+ sunfire installations).

And as the IWILL MB points out its not even that far away from home brew or at the very least a $100k officebrew given the amount of ram one has to buy.

Also its worth pointing out the processor supports up to a terabyte, the issue is the memory sticks/mb's are not available yet.

And here we are only talking about the SMP/redundancy style computing . For distributed style computing platforms (aka everything on the top 500 super computers), the 10th and 11th fastest computers in the world are Cray machines based on opterons currently running at Sandia National Labs and Oakridge National Labs respectively.

Now perhaps in your world the top 500 super computers are not considered big iron because they are distributed but i'm w/ the joe slashdotter crowd that considers "big iron" to include the top 500 fastest computers in the world.

Dunno about server, but I was wondering about a new desktop...
http://www.orionmulti.com/products/specs_ds96
- Performance 230 GFlop peak, 110 GFlop sustained (Linpack) -

What else ??? ah, yes, an ugly as hell server I wouldn't mind this winter : http://www.iwill.net/product_2.asp?p_id=90
Dunno if many people are into rack server modding, but you cannot make it more ugly than it is on the outside....

The inside can host 16 Opteron 800 cores and 128Gig of Ram, which make it pretty hot - both senses.....

For the Network War room, what about a nice visual representation of network attacks/activity, using "The Spinning Cube of Imminent Doom" which is both impressive and easy to explain...:
http://www.nersc.gov/nusers/security/TheSpinningCu be.php
(maintenance of servers going on, use the cache, luke...

Personnaly I like people that forego LCDs and such and directly use a nice and silencious video projector for general informations (say a Sanyo Z3).
Even more if you show your skills at system management using "Doom, the Aftermath"http://www.cs.unm.edu/~dlchao/flake/doom /after.html

Yeah, follow my words, I can garantee your customers will be impressed 8)

On a tangent, where have you seen AMD64 MBs that support 64GB of RAM ?

I have not seen any Opteron system which would require less than four processors to go to 64GB.

Hey curri,

Okay scratch that. I've just come across this. 64GB with only 2 Opterons.

You dork. I said high end x86 boxen have lights out, and yes, I'm aware the HP makes some of said high end x86.

As for 8 ways and such, give me a break. Go to HP's website (http://www.hp.com/ select an 8 way system, and increase its RAM and drive configurations to the standard in the US-IV system. What's it come to? Here's a hint, it's pretty close to $150,000 depending on exactly what you consider to be precisely equivalent to the Sun boxen.

Also, the US-IV systems are all dual core, so we're talking 16 cores. If you could buy a 16 core athlon machine with the sort of ram and Disks that the suns have, it wouldn't cost 39k, if you can get it at all, that is.

I never said the price for performance was equivalent, but spec out a few machines and see how close it is. It's the same in the 4-8 core realm (with the V40z), and it's about the same in the 16+ core realm due the difficulty in getting good 8+ core x86 systems. 48 cores of US-IV will be expensive, but there aren't a lot of x86 boxes out there that can touch it's whole-system performance. And yes, I too know what I'm talking about.

It's worth comparing this box with some of the Pentium M machines/barebone systems available based on miniITX motherboards with integrated video and no AGP. If you're not doing 3-D gaming, then these guys are very attractive.

In particular, look at some of the designs coming out of Soldam, such as the Alphia, Lepty and Rhapsody.

On the other hand, if you're looking for 3-D gaming with the best performance in the smallest possible package, it's hard to go past something like Iwill's ZPC64.

Drool over this: http://www.iwill.net/product_2.asp?p_id=36 ZMAXdp

As you were saying...

You may want to look into the IWILL Dual Opteron SFF PC It's in a small form factor design like a Shuttle XPC, but with support for Dual AMD Opterons.

Even if you don't have the money for both CPU's right now...it's a good start and you could add the 2nd CPU later. This would be the most powerful small form factor number cruncher.

According to the manufacturer's page, it looks like the other color is silver. I'm guessing the $499 is the bare machine w/o an processors.

After reading the rest of the comments:
I agree, SFF is a better route to go for "mobile gaming only" machines.

OK...
One of these, $499
Two of these, $250.98 x2
And so on...

Come on, be the ultimate performance whore. Take the performance hit of registered RAM. Spend the $3500-4000 for one bitchin' portable system. Just wait until the dual-cored chips; you'll have four effective processors. Don't listen to those who say SMP doesn't isn't worth it. If even the sound mixes in a different thread, even if the other processor only handles OS tasks, you'll gain some performance.

Or maybe I'm just obsessed with dual procs and power. I mean, what do you think I'm saving up for right now?

Yes, there is: http://www.iwill.net/zmax/zmaxdp_1.asp

So I guess those tiny dual Opteron systems were just a figment of our collective imagination?

Get off it, Apple certianly is a company that can claim many innovations but the MOTHERBOARD isn't one of them. They don't even build them (Foxconn does) and have very little to say in their design. Most of it is simply dictated by what they want on it. You want 8 memory slots and 64-bit PCI slots? Ok, that is going to take a certian amount of traces and a certian amount of space to do.

Apple doesn't need any innovation to stick a G5 in a smaller case, they could do it now. They just put less on the board. iMacs do not have all the expansion slots towers do and that makes the board much smaller both because the slots take up space, and because the traces they need do as well.

The real issue with G5s in lower end systems is cost, heat and competition. They don't want their lower lines competing with their towers, and the G5 DOES generate a good bit more heat than a G4, which is something you have to deal with. It's more expensive too.

So no, actually, Apple's proprietary system is doing the OPPOSITE of what you suggest. If it was commodity hardware, smaller boards would exist, as the existance of mini-itx P4, Athlon and Opteron boards demonstrate. Given that there is competition, companies usually roll out hardware as soon as they can. As is, Apple decides when it is in their intrests to roll out the G5 in their lesser lines.

i didn't see it mentioned here by anyone else, but as reported on a couple of (sites for one)
there's a neat new box coming out from IWILL that crams two(2) Opterons in a SFF case.
Unfortunately, if you need something now, this one will be coming too late for you unless you're a
developer/partner/etc:

"IWILL ZMAX based on nVIDIA nForce3 Pro 250Gb chipset will sample in July.
Volume production is planned in September, with a suggested price of $499.
IWILL plans to get attention in workstation market. ZMAXdp will include proprietary
form factor motherboard, 300W power supply, up to 2x3.5" HDD bay, and 1xAGP;
PCI and SI can offer various configurations for workstation market demand."

it sounds like it could be a nice little box...

other pre-built systems include:

Pre-built
Caliber

there are others, but I've lost my wish-list ;-)
You could also build one yourself, but I'd look for the nForce3 pro 250 or 250Gb, the NF3-150 didn't
exactly get extraordinary reviews.
Good Luck!

1 2 3

Introduction

Memory performance is becoming a key concern of computing system designers across many market segments. From PC applications surrounding multimedia (gaming, streaming media and video/photo editing) to networking systems routing hundreds of gigabytes per second of packets, high-performance memory systems are a must. Most people would know that the memory bandwidth on modern video cards is crucial to performance levels in Direct3D and OpenGL, and overclocking the memory will guarantee an overall boost 100% of the time. Essentially, main system memory goes by the same rule, and is often overlooked in most cases. SDRAM and DDR SDRAM are the most common types of memory used in personal computers across the globe, and offers fairly good performance as well as being very cost effective. RDRAM is another option, originally brought to power with the original Pentium 4, but dumped for DDR to satisfy the low to midrange market.

RDRAM is considerably more expensive that DDR SDRAM, and up until now the 100MHz PC800 specification didn't do well in comparison. Just recently 133MHz PC1066 was launched, and is now officially supported by the new Intel P4 and the Intel 850E core logic chipset, but this time promises to bring memory performance to the next level. Lets find out...

SDRAM and DDR

SDRAM and DDR SDRAM, being close cousins in architecture, share many of the same features and characteristics. An SDRAM or DDR SDRAM memory system is constructed out of multiple devices working in parallel on a wide data bus to answer a memory request. Lines are conserved on the command (address) bus by multiplexing or sharing the lines to transmit row and column addresses, or the information that locates a specific memory item in the 2-dimensional memory core array. DDR attempts to theoretically double performance of the memory system by applying double data rate technology to send data on each edge of the clock signal in theory, doubling the data output of a device while not increasing it's clock frequency in MHz. Currently DDR memory modules run on a data bus of 133MHz (PC2100 or DDR266) and 166MHz (PC2700 or DDR333). PC2100 and PC2700 refers to the 2.1GB/s and 2.7GB/s of memory bandwidth that each memory channel is theoretically capable of, although current AMD and Intel processors only run on a 133MHz data bus.

What is RDRAM?

RDRAM architecture takes a different approach from SDRAM. Rather than using a matrix topology with wide buses for data and shared buses for command information, narrow high-speed buses are used with an emphasis on making the electrical subsystem balanced and resilient, enabling high frequency operation by design. A single channel of RDRAM memory can fully service a data request out of a single device. A 16-bit data bus runs at a 400/533MHz clock frequency and sends data on both clock edges to deliver 1.6GB/s (PC800) and 2.1GB/s (PC1066) of memory bandwidth. Since lines are conserved on the data bus, the command bus can use separate lines for row and column address information to ensure the command bus can keep up with the abilities of the data bus. Up to 32 divides can be connected together in one RDRAM Channel to allow memory footprint expansion. Devices can be attached to the same PCB as the memory controller or put on modules very similar to SDRAM DIMMs.

Currently RDRAM memory modules run on a data bus of 100MHz (PC800) and 133MHz (PC1066). Up to 1.6GB/s of memory bandwidth per channel can be delivered by PC800 RDRAM on a 100MHz data bus, while the new breed of P4s run on a 133MHz (133MHz*4=533MHz) data bus, delivering up to 2.1GB/s per channel with PC1066 RDRAM. The P4 can make use of the full 2.1GB/s because it runs on the same data bus PC1066 is designed for, using a far superior memory technology in contrast to DDR SDRAM.

Testing

I used the Iwill P4R533-N motherboard based on the Intel 850E chipset to benchmark both the PC1066 and PC800 RDRAM, and the SOYO P4I Fire Dragon based on the Intel 845D chipset for benchmarking the DDR memory. SiSoft Sandra was used to measure CPU, CPU/multimedia and memory bandwidth, MemTach 0.93a was also used to measure memory bandwidth, and MadOnion's 3DMark 2001 SE was used to measure real-world multimedia performance. Direct 3D/OpenGL vertical sync and audio was turned off in all multimedia tests, and assume 'higher is better' for all benchmarks unless stated otherwise.

Test System

Processor: - Intel Pentium 4 (533) 2.4GHz
Motherboard: - Iwill P4R533-N (i850E)
- SOYO P4I Fire Dragon (i845D)
Memory: - Samsung 2x128MB PC1066 RDRAM
- Samsung 2x128MB PC800 RDRAM
- Apacer 256MB PC2100 DDR266 DIMM
Storage: - Maxtor 20GB ATA133 7200RPM
Graphics: - PixelView GeForce4 Ti4600 128MB
Sound: - Disabled
Software: - Windows XP Professional
- Intel Chipset INF Driver
- NVIDIA 28.32 Detonator XP Drivers
- SiSoft Sandra 2002
- MemTach 0.93a
- 3DMark 2001SE


Effective Bandwidth

Effective bandwidth is just what the name suggests - it looks beyond subtle specs and gets a handle on what real world performance expectations should derive from a memory technology. As you will see from the graphs below, the PC1066 based system thrashes the DDR and PC800 by an exceptional margin in the memory bandwidth tests, but is pretty much on par with the others throughout the SiSoft CPU and Multimedia tests, although that is not really what we're really looking at here. The PC2100 DDR also proved to be strong up against the PC800, which came in dead last overall.

Multimedia Performance

Bandwidth is one thing, but multimedia performance is another, and is a direct indication if all that extra bandwidth is going to good use. I decided that MadOnion's 3DMark 2001 SE would be sufficient for the evaluation, since it offers an all round performance benchmark of the graphics subsystem - the most demanding device in the average PC. As you can see from the results, the PC1066 based system excels with nearly 500 additional 3D Marks over both the PC2100 DDR and PC800 RDRAM platforms.

Conclusion

It is quite evident from the benchmarks shown in this article that PC1066 RDRAM teamed with the Intel 850E chipset is currently the fastest, most superior and effective combination available for achieving the highest possible memory bandwidth. Intel started off in the right direction with the initial RDRAM support for their original line of Pentium 4s, but would have suffered in the long run if they didn't make the switch to DDR just in time to keep face with the growing demand, pressures, and competition from AMD. Until AMD or Intel produces a CPU that runs on a 166MHz or 200MHz data bus, the benefits from anything faster than PC2100 are going to be sorely missed on any DDR based platform. Intel has once again secured their piece of the pie in the high-performance arena with their 850E chipset, the new variety of Pentium 4, and their true support for PC1066 RDRAM. If your tossing up between DDR and RDRAM for your next upgrade, then PC1066 is definitely worth considering if you seek nothing more than pure muscle alongside the '533' Pentium 4.

I agree -- 64C sounds too high. Here is a good article on the subject from Via Hardware.

I recommend the following three steps to cool it down:

1. Run H.ODA's WPCREDIT/WCPRSET and set the ACPI HALT cooling on, if your processor is running at under 1.33 GHz or if you're not running Win2K. This will keep your idle temperature down. See the end of the VIA Hardware article for the admonition about CPU speed and Win2K stability.
2. Use Arctic Silver II thermal paste. I bought some at Fry's and it's pretty cheap. It brought the temperature down 2-3C under load.
3. Try the NoiseControl Silverado fan, if it fits in your face. North Americans no longer have to buy it from Germany, as Plycon sells it in the US now.

I have a 1.2GHz Athlon which I run at about 1.35GHz by upping the FSB. My IWill KK266 board claims that it idles at 26C, and it gets up around 41C during heavy use, and 49C in a tight loop.

I have a shutoff at 50C, which it last reached when Outlook went into a tight loop overnight. I ran a program called MBM to check on it, and it recommended a program called Shutdown Now to shut down and power off in case of alarm. Unfortunately, I hadn't noticed that Shutdown Now was nagware, and my system was up all night at 50C, sending me pages every 5 minutes. When I got to work in the morning, there was a pop-up dialog saying to please send in $15 to them before it would shut off my computer. Talk about lame! It would have been fine to nag a boot time, not not at shutdown time! I'm just glad the program didn't fry my CPU. Anyway, I replaced it with the NT Resource Kit program called shutdown.exe that took a little bit of mousing around to get into MBM's configuration, but no way was I going to give money to the guy who almost fried my computer.