If you want an enthusiast board for OC'ing the CPU's, I'm afraid you'll have to wait awhile.
Perhaps the prices will have dropped slightly by then, but I wouldn't expect by much.I'll see what I can post later tonight.
ASUS P6T6 WS Revolution
Core i7 920, OC'd to 4.11GHz
6GB OCZ Platinum DDR3 @ 1600MHz (it didn't do as well with Crucial UDIMM ECC, due to slower CAS=7, so it's stored for later)
ATI HD Radeon 4870 1GB
300GB Velociraptor (boot)
8*WD 320GB RE3 RAID 5 (ARC-1231ML)
Lian Li PC-V2010 Full Tower (Silver)
2x Lian Li EX-H34 Hot Swap bays
Corsair HX1000W PSU
Noctua NH-U12P SE1366 cooler
Noctua NF-P12 fans (4)
Yate Loon 140mm (2)
Vista 64 Ultimate
Linux
Basically, the i7-965 with ECC.
I'll see what I can post later tonight.
ASUS P6T6 WS Revolution
Core i7 920, OC'd to 4.11GHz
6GB OCZ Platinum DDR3 @ 1600MHz (it didn't do as well with Crucial UDIMM ECC, due to slower CAS=7, so it's stored for later)
ATI HD Radeon 4870 1GB
300GB Velociraptor (boot)
8*WD 320GB RE3 RAID 5 (ARC-1231ML)
Lian Li PC-V2010 Full Tower (Silver)
2x Lian Li EX-H34 Hot Swap bays
Corsair HX1000W PSU
Noctua NH-U12P SE1366 cooler
Noctua NF-P12 fans (4)
Yate Loon 140mm (2)
Vista 64 Ultimate
Linux
Basically, the i7-965 with ECC.
It gets as high as 1.2GB/s.
Cache of course, but 575MB/s on avg. NO, AFIAK. It doesn't have the second QPI channel used in DP boards, and I don't recall seeing that you could. You can however, use a DP Xeon 55xx in an SP board, as it can shut its second QPI off. Sort of a waste though, as they're more expensive. Might be worth considering if you plan to get a DP board and second CPU at a later time however. You only loose the SP board's cost, minus anything you get if you sell it.
It gets as high as 1.2GB/s.
NO, AFIAK. It doesn't have the second QPI channel used in DP boards, and I don't recall seeing that you could. You can however, use a DP Xeon 55xx in an SP board, as it can shut its second QPI off. Sort of a waste though, as they're more expensive. Might be worth considering if you plan to get a DP board and second CPU at a later time however. You only loose the SP board's cost, minus anything you get if you sell it.
If you're using an SP board, use an i7 if you don't need/want ECC, or a W35xx if you do. You're the only one who can answer that one.
It gets as high as 1.2GB/s.
NO, AFIAK. It doesn't have the second QPI channel used in DP boards, and I don't recall seeing that you could. You can however, use a DP Xeon 55xx in an SP board, as it can shut its second QPI off. Sort of a waste though, as they're more expensive. Might be worth considering if you plan to get a DP board and second CPU at a later time however. You only loose the SP board's cost, minus anything you get if you sell it.
I went with the Green Drives because I wanted quiet for my storage space and don't really need the speed for writing images or mp3's to disk. May I ask what kinda of software you run that requires that kind of transfer speed?EDAThat is an insane transfer rate. It eats SSD
If you're using an SP board, use an i7 if you don't need/want ECC, or a W35xx if you do. You're the only one who can answer that one.
For DP boards, you'd be better off sticking with 55xx parts. You can upgrade it to a DP system at a later time if you choose, assuming you can't from the beginning. It's an option, if you're on a tight budget ATM.
It's a memory type that has the ability to detect and correct errors, and results in a more stable system.
You can get a higher clock out of it, but you have to be careful, as either too much heat or voltage can kill it. You also would run into stability issues, and have to drop it back anyway.
You'll have a lot fewer headaches if you do. I dropped mine back to 4.12GHz (i7 920 on air) for stability, and haven't had any issues.
I wish the math geek that propagated this rather disingenuous proposition had never done so. It's caused billions of needless questions and spawned billions of wrong answers. This math also means that in a 100 or 1000 drive array we will have one drive failing every month or something crazy like that. Doesn't happen. In a 1000 drive array ya get a lemon or two right away and then kept properly the system operates for years on end - no problem.
Well, there's really only two things to consider and on average they scale (or we can for most purposes, pretend that they scale) linearly.
There's clock speed. Some have called this "pep" which works for me.
And then there's bandwidth. Number_of_Cores X Clock_Speed = bandwidth.
That's it. We can benchmark particulars but generally speaking, that's all there is. Pick the balance between the two (plus price) which you think you want and go for it. I mean if you're set on buying a new system.
Anything else is just an order of complexity added in that probably won't matter much if at all, in practice.
Another consideration is to get a $200 used system and run it as your server. And trust me a $200 used system these days ROCKS! as a server!
EDIT: something like this:
Price $250
Clock 3.8GHz with HT
3GB Ram
73GB (SCSI/15000rpm)
COMBO Optical drive
Quadro FX3400
Windows XP Pro
KB/Mouse
etc. http://page19.auctions.yahoo.co.jp/jp/auction/x68836645
Sorry for the Japanese but that's where I live. Anyway systems like this are all over the place in almost every country - in great abundance. They make great personal servers! I ran one off of a DEC Alpha 233MHz running Digital Unix up until just a little while ago.
And the one I linked to will just cream a mac-mini!!! Plus all the cool crap it comes with... FireWire, USB, UltraSCSI 360, ATA, etc. It can get confusing. MTBF/MTBR doesn't change, but the UBE does. For example, a 10 drive stripe will change the UBE (entire set) by an order of magnitude , so a consumer set would drop to 1E13, and an enterprise set to 1E14.No. I see that false math all over the place. The failure rate MTBR/MTBF for the drives are exactly the same. That math is technically correct but it's basically 99.9% bullshate. It would be like saying that if you have a rock in each hand the odds that you'll drop one of them doubles. No, sorry, I'm not going to drop either till I'm good and ready to set it down.
Failures during a rebuild would drive me nutz. But say $50/drive per year (3 yrs), it's cheap enough that it wouldn't warrant any complaints.
It can get confusing. MTBF/MTBR doesn't change, but the UBE does. For example, a 10 drive stripe will change the UBE (entire set) by an order of magnitude , so a consumer set would drop to 1E13, and an enterprise set to 1E14.
I certainly wouldn't want to have to deal with a 1000 drive array of any type.
If it's managed by software RAID, it's likely no big deal, as it has the ability to spin the drives down, and likely not being run 24/7. In a small array, it shouldn't be noticed until the drives reach "old age". So at least 3 years, and possibly 5. (Given some of the recent issues, and warranty changes, I'd figure on the low side).
Hehehehe... I was exaggerating to make a point but I dunno... it might actually be fun - I mean if we got paid for it of course.
It would have to be a really BIG check.
With No Guarantee on data security, let alone availability.
I know what you mean.Yeah, true, but this is different than "double the chances of [catastrophic] drive failure". Most people (I guess) won't know what 1E14, 1E13, etc. actually means in practice. I think the math is confused (every time I read posts) to mean the drive has twice the chance of breaking - boom - no more worky worky. Most people don't know or don't want to know about internal (and typically silent) error handling - they just know it works or not. Ya know.
Too much to make an accurate model I guess, or they'd be available.
I dunno! But I can tell ya it's usually NOT a good idea to run dedicated server facilities on a machine you use as a workstation! And $200 for a system like that - what's the hard part? That's about the price of one HDD or something.
Here's the manual BTW.. ftp://ftp.software.ibm.com/systems/support/system_x_pdf/88p9138.pdf
).
Well someone who has actually administered high volume commercial servers would know more but I ran a 500 member FTP, hosted HTTP that got about 1000~2000 unique IP hits per month, and served up an IRC network with, umm, I think there were about 20 channels at the end. The FTP served out about 90 gigs a month and received around 30 gigs a month, the HTTP was around 30 gigs a month outbound, and I don't remember how the IRC profiled. The system stress was so light that I also use it for my email and some web surfing. Even with all of that I don't recall the CPU ever hitting 30% or higher for more than a second or two. the HDDs were not stressed either and the traffic (when I capped FTP at 75k per connection) was well within the ISP line speeds which of course was not a problem AT ALL for the link speeds - TBase100 at the time.
All this was not even a challenge for the little digital Multi with it's VX42 Alpha 21066A 233MHz processor and it's 128 megs or memory. So IMHO no, you do not need two Mac Pros to serve up files over a web connection or two (PS: yes, you can use multiple lines.