"Rate my Rig" thread

[starslayer]

All that looks perfectly fine. If you decide on the more powerful video card option with your remaining budget, you could get a GTX 460 for about £160 (a much better card for only a little more money), or a 5850 for about £250. You may want to get the best of both worlds and nab the 460, along with the Wii + Wii Fit if you can manage it.

[Xon]

Downsize the PSU, that rig isn't going to pull even 350-370 watts even under max load, so a 400watt PSU is the most you should consider. Phenom II X4 955 has a 125watt TDP, ~220 for the HD 5770. But it is rare to push your GFX card and CPU to the max even with intensive games, ie one or the other.

4gb DIMMS are horrifically expensive. Is there any reason you are going for them? The difference between 8gb and 16gb is a massive price difference for very little if any performance change unless you are hitting some niche computing topics.

[Lief]

Downsize the PSU, that rig isn't going to pull even 350-370 watts even under max load, so a 400watt PSU is the most you should consider. Phenom II X4 955 has a 125watt TDP, ~220 for the HD 5770. But it is rare to push your GFX card and CPU to the max even with intensive games, ie one or the other.

4gb DIMMS are horrifically expensive. Is there any reason you are going for them? The difference between 8gb and 16gb is a massive price difference for very little if any performance change unless you are hitting some niche computing topics.

I already have a corsair 450w psu, the vga I chose recommends 450w, due to the 5 year warranty and likely future upgrades, I'd rather just get the 550w now and forget about it (replacing a psu is more annoying than replacing a vga card, though I did strip, clean and rebuild this machine the other night, so it's not a massive issue).

I want to keep this machine running (I can scavenge HDD / dvd from another pc), and the 550w is only a tenner more than the 450w anyway.

Sorry, that is 2 x 2gb DIMMS, copy and pasted from notepad

Thanks for the tip about the gtx460, it does seem like a lot more bang for the buck.

[Uraniun235]

I already have a corsair 450w psu, the vga I chose recommends 450w, due to the 5 year warranty and likely future upgrades, I'd rather just get the 550w now and forget about it (replacing a psu is more annoying than replacing a vga card, though I did strip, clean and rebuild this machine the other night, so it's not a massive issue).

The graphics card manufacturer recommends a minimum of a 550W PSU because they know that there are a lot of people running shitty no-name power supplies that can't actually deliver as much power as they claim to.

[Starglider]

Motherboard for my new workstation turned up. Here are the parts I have now;

Code: Select all

Motherboard  : EVGA Classified SR-2 (dual socket 1366)
Water Loops  : 2 x Zalman Reserator 1 V2 (with 6 x Y connectors)
Storage      : 2 x Athena Power BP-SATA1842B 5.25" to 4 x 2.5" adapter
Power Supply : 2 x Zalman ZM1000-HP 1000 watt water-cooled power supplies
               (one for motherboard / CPUs /drives, one for GPUs)
Monitors     : 1 x Dell 3008 WFP 30" 2560 x 1600 (16:10)
               2 x HP LP2065 24" 1600 x 1200 (4:3 portrait mode)
Case         : Lian Li PC-777 black
               (modified to take HPTX and two PSUs, 2.5" drive bays removed)

Here are the parts on my plan;

Code: Select all

Processors   : 2 x Xeon 5680 (3.33 GHz 6 core, should overclock to 4 GHz easily)
CPU Coolers  : 2 x EK-Supreme HF waterblocks
Memory       : 12 x G.Skill RipJaw 4 GB (48 GB total)
Graphics     : 3 x Sapphire 5970 Toxic 4 GB (should overclock to 1 GHz core ok)
GPU Coolers  : 3 x EK FC 5970 Plexi Full Cover VGA Water Block
Storage      : 4 x G.SKILL Phoenix Pro Series FM-25S2S (120 GB SSD) in RAID 0
               4 x SAMSUNG SpinPoint MT2 HM100UI 1TB 5400 RPM 8MB in RAID 1:0
Optical      : Blu-ray writer

Although in practice I am probably going to buy one 5870 for now and get the decent GPUs when the next generation comes out (Radeon 6970 currently rumored for next spring). There is of course no point having three 5970s for gaming, I am getting them for GP-GPU use (AI research). The reasons I am not getting a rack full of cheap cluster nodes is that (a) the bisection bandwidth and latency requirements are such that I would need a rack of very expensive cluster nodes with Infiniband cards and switches, (b) clusters are just more hassle during development, (c) I already have access to that at the office, and (d) keeping the noise down would be even harder.

Does anyone know of a sane single-slot cooling solution for the 5870? Six of those might make more sense than three 5970s, but I'm not aware of any non-homebrew solution.

[phongn]

Although in practice I am probably going to buy one 5870 for now and get the decent GPUs when the next generation comes out (Radeon 6970 currently rumored for next spring). There is of course no point having three 5970s for gaming, I am getting them for GP-GPU use (AI research). The reasons I am not getting a rack full of cheap cluster nodes is that (a) the bisection bandwidth and latency requirements are such that I would need a rack of very expensive cluster nodes with Infiniband cards and switches, (b) clusters are just more hassle during development, (c) I already have access to that at the office, and (d) keeping the noise down would be even harder.

No ECC? Also, why are you getting 5400RPM drives and then putting them in RAID 10?

Does anyone know of a sane single-slot cooling solution for the 5870? Six of those might make more sense than three 5970s, but I'm not aware of any non-homebrew solution.

Watercooling your GPUs is the only thing that comes to mind. Modern high-end GPUs just produce too much eat for single-slot cooling

[Ace Pace]

I'm not really sure why you're getting AMD GPUs for GPGPU, from my understanding the CUDA SDK is far more powerful and in better shape in terms of update and support than Brook or whatever AMD is currently marketing.

[phongn]

I'm not really sure why you're getting AMD GPUs for GPGPU, from my understanding the CUDA SDK is far more powerful and in better shape in terms of update and support than Brook or whatever AMD is currently marketing.

AMD's FireStream may work better than their consumer-grade GPGPUs but Nvidia's Fermi looks to be more mature and robust right now (not the least since it has ECC)

[Starglider]

No ECC?

No. I have ECC in my servers, also using ECC memory with overclocked components would make no sense. This is a development workstation, the odd bit error when churning massive data sets is tolerable.

Also, why are you getting 5400RPM drives and then putting them in RAID 10?

Because I'd like 2 TB of raid capacity and there are only three 5.25" slots left (and zero 3.5" slots) after fitting two PSUs in. Technically I only need a terabyte or so for ghosting the SSD array (current and previous images), but it'd be handy having extra local TB for DB logs, pushing them back and forth from the file server is kinda slow even on gigabit ethernet. Next solution is of course a ten gigabit home LAN and a new file server, but my budget doesn't go that far.

Watercooling your GPUs is the only thing that comes to mind.

Obviously I'm quite happy to watercool but the usual 5870 blocks are too tall to stack them one to a slot, plus the cards usually have a third output in the upper half of the bracket.

Modern high-end GPUs just produce too much eat for single-slot cooling

Well, you can get single slot 5870 equivalents for compute servers, but they need really loud forced air cooling.

I'm not really sure why you're getting AMD GPUs for GPGPU,

AMD GPUs have about twice the FLOPS/dollar compared to Nvidia GPUs. Also I don't like the GF100 shader design and PTX (CUDA assembly) in general. For probability distribution matrix computations, the Evergreen VLIW shaders work quite well.

my understanding the CUDA SDK is far more powerful and in better shape in terms of update and support than Brook or whatever AMD is currently marketing.

CUDA has been obsoleted by OpenCL for non-gaming applications and DirectCompute for games. I'm using OpenCL, for which AMD's drivers are significantly better than Nvidia's. Excepting critical sections that benefit from assembler, there is no reason to develop for a nasty proprietary solution like CUDA.

Nvidia's Fermi looks to be more mature and robust right now (not the least since it has ECC)

That is useful for a lot of applications but it isn't really worth it for a development workstation. I am prepared to tolerate 1 in 1000 test runs spuriously failing if the time taken to run a test sequence can be cut from 2 minutes to 40 seconds.

[phongn]

No. I have ECC in my servers, also using ECC memory with overclocked components would make no sense. This is a development workstation, the odd bit error when churning massive data sets is tolerable.

Well, the CPUs do dynamically overclock so it might make some sense. ECC should protect against some errors during overclocking, too?

Also, why are you getting 5400RPM drives and then putting them in RAID 10?

Because I'd 2 gb of raid capacity and there are only three 5.25" slots left (and zero 3.5" slots) after fitting two PSUs in. Technically I only need a gig or so for ghosting the SSD array (current and previous), but it's handy having extra space for DB logs, reading them from the file server is kinda slow even on gigabit ethernet.

Couldn't you do a 2TB RAID-1 (or even a RAID 5) array? It just seems weird that you're going for RAID 10 with slow drives (RAID 10 usually is done to get more IOPS)

That is useful for a lot of applications but it isn't really worth it for a development workstation. I am prepared to tolerate 1 in 1000 test runs spuriously failing if the time taken to run a test sequence can be cut from 2 minutes to 40 seconds.

What about using AMD's FireStream cards? They've got some better features for compute that were left out in their consumer cousins.

[Starglider]

Well, the CPUs do dynamically overclock so it might make some sense. ECC should protect against some errors during overclocking, too?

ECC won't help prevent errors in CPU overclocking. Theoretically it might with memory overclocking, but I doubt it helps much otherwise all the gold-plated super-enthusiast memory (and people going for overclocking records) would use ECC. I'm not a hardware engineer, but my guess is that timing skew errors are the primary problem with memory overclocking and ECC doesn't really help with that (because it tends to cause command/address corruption and multi-bit errors). In any case all the available ECC DDR3 memory is considerably slower than good non-ECC memory, so I won't be using it for this machine.

Couldn't you do a 2TB RAID-1

Simple RAID 1 is almost never done with more than two disks, because the capacity of the array is only the capacity of the smallest disk in the array. Given four disks, the usual solution is to aggregate them into two large volumes with RAID 0 and then RAID 1 mirror that (or vice versa, it doesn't make a lot of difference).

(or even a RAID 5) array?

I'd need a controller card or software RAID for that, the motherboard doesn't support it. RAID 5 is neat, I'm running it on my file server, but I'd rather go for the simpler solution here.

It just seems weird that you're going for RAID 10 with slow drives

It's a simple way to get redundant storage. It's a little less cost-efficient than software RAID 5 of 3 drives but really the cost is trivial compared to the rest of the machine. Now that I think about it, perhaps I should ignore RAID for the HDs and leave them as separate volumes, then set the backup script to mirror the SSD array to different drives in sequence. Or I could use 3.5" drives in external cases (the motherboard has a couple of ESATA ports), but that would be kinda messy. To be honest this is the least important bit of the machine.

What about using AMD's FireStream cards?

Not quite as hideously overpriced as the Nvidia Tesla cards, but still, lots more money for only one real advantage; 4 gigabytes of memory per GPU instead of 2. Which would be nice, but I'd rather have GPUs I can overclock by 20%. Remember that unlike Nvidia, AMD don't cripple down the dual precision performance of their consumer cards by 75%. Also I don't think standard waterblocks would fit FireStream cards.

They've got some better features for compute that were left out in their consumer cousins.

Well, currently I am running the application of interest on a 1 GB 5850 with no problems (albeit on relatively small document sets). How do you think a FireStream (or Tesla for that matter) would help?

[Starglider]

Got the memory (48 GB of RipJaws), waterblocks, pump and SSDs. I went with 4 x 100 GB Vertex 2s (in RAID 0) instead of 4 x 120 GB Phoenix Pros, because the Vertex 2s have the best sustained performance after heavy write without TRIM, and TRIM /still/ isn't supported on RAID arrays. Processors are on order. I decided to take Phongn's advice on the HD's, I'm waiting for the 750 Gb Seagate 7200RPM drives to come back into stock. Turns out the motherboard will do RAID 5, but I'm not sure if it can span an array across the SATA 3 and SATA 2 ports, so I may end up using Windows 7 software RAID on the HDs. Which should be tolerable as the HDs are for bulk data storage only, they'll be spun down when the machine is doing compute runs.

The motherboard is rated for 7 GPUs and apparently the 6970 is due out in December, so my current plan is 1 x AMD 6870 (slot 1, might have to be the Eyefinity 6 edition if that's the only way to get a one-slot version), 3 x AMD 6970 (slots 3, 5 and 7), 1 x Dual 10 Gigabit NIC (slot 2) with two slots unused. That's assuming the 6970 fixes the 'can't turn off crossfire' flaw, otherwise it'll be six 6870s. AquaComputer make good single slot waterblocks so that shouldn't be a problem. I'm a bit concerned with power though, 6-core Xeons at 4.2 GHz take about 180W each, the motherboard takes 80W, 50W for the drives, 60W for pumps and fans, 250W for an overclocked 6980, 400W each for overclocked 6970s and 30W for the NIC (10GBASE-T is a power hog) = 2030 watts. Practically you'd never be able to max out everything at once, but I may substitute a 1200 watt power supply for one of the existing 1000 watt PSUs just to be on the safe side.

The current cooling plan is eight Reserator V2s in three loops, a CPU loop with one per CPU (shared pump), and two GPU loops, each with a pump feeding two cards via a Y-splitter, then two parallel 'hot' reserators, feeding back into one 'cool' reserator, which will feed back into the pump. That should stay within the rated TDP dissipation at heavy compute load. Alas I won't be able to claim the machine is completely fanless, as no matter how much water you have 2000 watts demands a couple of case fans, but it should be pretty quiet, especially for a 2-CPU 7-GPU machine.

[Kuroneko]

Starglider, if you ever get a Radeon watercooling solution that would make it analogous to the GTX 480 Hydro Copper), please let me know. Since I'm unlikely to get a motherboard monstrous as yours, not blocking the adjacent PCI slot would be a nice bonus for me.

The current cooling plan is eight Reserator V2s in three loops, a CPU loop with one per CPU (shared pump), and two GPU loops, each with a pump feeding two cards via a Y-splitter, then two parallel 'hot' reserators, feeding back into one 'cool' reserator, which will feed back into the pump.

What's the heat dissipation of the Reserators?

You might consider the Lian-Li V2120 with stackable 3x120mm radiators on top. Each one dissipates 400-500W depending on flow rate, though I'm unable to find stats on how their performance scales. Obviously, that by itself is inadequate for your needs, but it might be better--and much cheaper--than an entire Reserator farm.

I was looking into that case for an E-ATX system, though it will fit an SR-2. I'm hoping that this slot cooler can reverse the orientation of the PCI bracket and mount a 2x120mm radiator, so I'd have a very substantial amount of watercooling without case modding or going external. And look better than some Mountain Mods monstrosity (props to having a case that can fit 12 4x120mm radiators, but... no).

[Starglider]

Starglider, if you ever get a Radeon watercooling solution that would make it analogous to the GTX 480 Hydro Copper), please let me know.

If you mean cards with waterblocks pre-installed, PowerColor made a bunch (5870 and 5970), but (a) from what I hear the heatsink was slightly too thick to allow use of the adjacent slot and (b) they're hard to find now as production terminated in anticipation of the 6 series. If I was going to go Nvidia I would use the Hydro Copper, several people have successfully installed six of them on the SR-2, but after all the Nvidia underfill-related reliability problems there's no way I'm buying from them.

Incidentally, according to my calculations the SR-2 should support 10 GPUs based on BIOS memory allocation alone, but EVGA only claim support for seven. I intend to investigate this using old surplus video cards, to see if it is in fact possible to fit five 6970s. Bandwith would suck as most of the slots would be knocked down to PCIe x8, but I think it would still be a significant performance gain for the code I want to run. Of course I would need a 1600 watt PSU just for the GPUs...

What's the heat dissipation of the Reserators?

Nominally 300 watts, although some people report running 400 watts through them with good temperatures (<50 celcius processor under load); it probably depends on the airflow around the unit. I'm putting mine on raised mesh panels for better convection. Even intensive GPU compute tasks usually don't load the GPUs above 60% TDP or so, a lot of the hardware is idle (rasterisation, tesellation, texture sampling), but then again I will probably do a mild GPU overclock which will raise TDP again.

You might consider the Lian-Li V2120 with stackable 3x120mm radiators on top. Each one dissipates 400-500W depending on flow rate, though I'm unable to find stats on how their performance scales. Obviously, that by itself is inadequate for your needs, but it might be better--and much cheaper--than an entire Reserator farm.

To be honest the Lian-Li Legacy case and Reserator array is not an entirely practical choice, there is a certain aesthetic/coolness (no pun intended) factor involved. In fact the Legacy case is the smallest case that anyone on the EVGA forums has tried to fit an SR-2 into. In any case since I have them now I will not be switching, but certainly this solution would not make sense to anyone who prefers their computers to hide under the desk.

I was looking into that case for an E-ATX system, though it will fit an SR-2.

I've seen a few people posting (on the EVGA forums) that they're happy with that case, it looks decent to me, and using the 2 spare PCI slots for an extra radiator is a good idea.

[Starglider]

System is now operational on test bench;



Running with two X5680s @ 3.33 GHz (no overclocking yet), 48 GB, 4 x SSDs in RAID-0 etc. The two reserators are keeping the CPU temps at about 48 degrees C under stress test, which is pretty good for silent cooling. As yet there is absolutely no cable management;



In fact the cooling hoses have an unintentional resemblence to entrails; red, coiling, slightly sticky. They give it a vague feel of biomechanical horror, some hideous hybrid of computer and synthetic life form. I find this entirely appropriate for its intended purpose.

[Zixinus]

Planning to upgrade my computer. I only need to upgrade my CPU and thus motherboard as well. Here is what I have in mind so far:

AMD ATHLON II X3 450 (after the reccomendation of Tom's hardware guide, would it be worthwhile to get a x4 or something a bit more expensive?)
Gigabyte GA-M68M-S2P (this has DDR2 RAM)

I'm also looking for upgrading the memory and found Kingston 2GB for up to 4 gigs on my system. Does anyone know the difference between "regular" kingston DDR2 and HyperX? So far, I am a bit confused. The memory is only a hundred forints more expensive (half a dollar roughly), so I wonder whether it would be worth it.

[starslayer]

Could you post the memory specs for both modules please?

[Zixinus]

KVR800D2N6/2G 2GB 800MHz DDR2 Non-ECC CL6 DIMM < Regular

KHX6400D2B1/2G 2GB DDR2 800MHz Non-ECC CL5-5-5-15 1.8V < HyperX

I strongly suspect that they are the same, except that the tiny bit more expensive one has a fancy sticker on top.

[starslayer]

The HyperX has a heat spreader mounted on it and slightly lower latency, plus it probably has a lower failure rate (that's a guess, though). Will you see much benefit from this? Probably not, but since it's only 50 cents more, may as well get it.

[Zixinus]

Will it cause problems that my other DDR2 RAM is not the same and perhaps a different manufacturer?

[His Divine Shadow]

Rate this old ass rig

CPU: INTEL CORE 2 E6600 2.4GHZ 1066/4M S775
Graphics card. CLUB 3D GF 7300GT PCI-E 256M CRT+TV+DVI
RAM: 2GB 667MHZ DDR2 NON-ECC CL5 DIMM KIT
Motherboard: P5B DELUXE/WIFI-AP ASUS S775 IP965 DDR2 SATA2 WIFI GBL 8-CH

I actually upgraded today with a new graphics card: ASUS GTS 450 1GB DDR5. I believe it's been the weakest link in my computer for a while. It's holding up pretty well for a computer bought in early 2007.

[phongn]

Will it cause problems that my other DDR2 RAM is not the same and perhaps a different manufacturer?

Probably not but then the extra (slight) performance of the HyperX becomes sort of irrelevant.

Rate this old ass rig

CPU: INTEL CORE 2 E6600 2.4GHZ 1066/4M S775
Graphics card. CLUB 3D GF 7300GT PCI-E 256M CRT+TV+DVI
RAM: 2GB 667MHZ DDR2 NON-ECC CL5 DIMM KIT
Motherboard: P5B DELUXE/WIFI-AP ASUS S775 IP965 DDR2 SATA2 WIFI GBL 8-CH

I actually upgraded today with a new graphics card: ASUS GTS 450 1GB DDR5. I believe it's been the weakest link in my computer for a while. It's holding up pretty well for a computer bought in early 2007.

Only thought I have is adding more RAM. 8GB is rather nice on Win64.

[Starglider]

@Kuroneko; in initial testing on my new system, a simplified version of my AI code is running four times as fast on the two Radeon 5770 GPUs (160 cores / 1280 threads / 850MHz), compared to running it on the two Xeon x5680 CPUs (12 cores / 24 threads / 3.33 GHz). As the price of the former was roughly one tenth as much as the later, that's a price-performance ratio of about 40:1, which is in a nutshell why all the HPC people are moving to GPGPU for compute. This is with software emulation of double precision as well, it should go even faster with a combination of single precision FP and hardware support for DP (which the 5770 lacks) on critical paths. I am looking forward to putting four 'Antillies' Radeon 6970X2s in this system and cranking up the clocks.

[Zixinus]

I'm buying a new processor. I originally thought about getting a AMD Athlon II X4 955 Bl. ed but figured that the price is not that much worth it.

So instead, I have decided upon an II X4 640. Just how large would be the performance difference, as I'm a bit confused in regards of AMD processor performance (oh the days where you could tell the performance of a processor by looking at its clock speed).

[phongn]

So instead, I have decided upon an II X4 640. Just how large would be the performance difference, as I'm a bit confused in regards of AMD processor performance (oh the days where you could tell the performance of a processor by looking at its clock speed).

The 955 has a 6MB L3 cache and is 200MHz faster whereas the 640 has no L3 cache; the performance difference will probably depend on what you want to do.