Originally posted by: Gothgar
Originally posted by: larciel
Can I just watch while you build that machine and test it?
me too
lmao were all in aw
I don't think the Quattro series cards need a compatible chipset for SLi, unlike the GeForce cards. I read somewhere about ordering SLi Quattro's on a Dell workstation, even though it wasn't on an nVidia chipset.
nerp
Diamond Member
- Dec 31, 2005
- 9,865
- 105
- 106
Don't get too excited guys. Custom linux and dedicated apps on such a machine is pretty much a yawnfest.
Now, when that machine is sold for scrap, that's when the real fun begins. . . but then again, we'll probably have octo or 16 core CPUs and 16GB standard by then. 2x8 or 1x16GB stick? hmm. decisions decisions.
Custom linux and dedicated apps on such a machine is pretty much a yawnfest. Now, when that machine is sold for scrap, that's when the real fun begins. . . but then again, we'll probably have octo or 16 core CPUs and 16GB standard by then. 2x8 or 1x16GB stick? hmm. decisions decisions.
What you say is correct, SLI Quadro's are possible on non-SLI chipset mobos but you are still required to have two x16 (electrical, not just physical) PCIe slots or else you may as well not bother putting the Quadro's in the box as they will be crippled for bandwidth.
Couldn't you use this board, Tyan Tempest i5400XT (S5396)?? If you were using Quadros for CAD, can't you SLI those in any board that supports dual lane x16 PCI-E? If so, I'd try to pick up that board or a similar one.
I don't really see where you are going to beat a 4p (or 8p) AMD in a Thunder s4985.
DDR2 800 ECC
triple gigaLAN
4xPCIe x16 (2 w/x16 signals)
Your 16gb DDR2 and video cards would most likely move to the new Tyan.
4 CPUs and the mobo would be less than $6k
DDR2 800 ECC
triple gigaLAN
4xPCIe x16 (2 w/x16 signals)
Your 16gb DDR2 and video cards would most likely move to the new Tyan.
4 CPUs and the mobo would be less than $6k
He doesn't need 4 processors or 8 processors, he wants 8 cores. Two Quad Xeons will be beastily and I think that's what he was aiming for especially since the software that he will be using was written with the Core architecture in mind.
He's already stated that his software was written with the Core architecture in mind and it's optimized for it. Going to an AMD platform would hinder his performance. I'm not trying to push Intel on anyone, if AMD was the better option I'm sure 99% of us would tell him to go that direction, but from what he wrote that won't be the case.
GuitarDaddy
Lifer
- Nov 9, 2004
- 11,465
- 1
- 0
I disagree
The S4985 is $838 + 4x 2.8 8series optys is $4800 for a total of $5638 and it would seriously lose to either config I mentioned (especially considering the app is optimized for core architecture)
D5400XS $614 + 2x QX9775 3.2ghz $3220 total $3834
D5400XS $614 + 2x E5450 3.0ghz $1940 total $2554
The 8p version is not in the same price range
S4985 $838
M4985 daughter card $ 617
8x 2.8 8series optys $9600
Total $11,055
GuitarDaddy
Lifer
- Nov 9, 2004
- 11,465
- 1
- 0
E5450 should be good, according to benchmarks. Just make sure to get the 80w one, not the 120w one. 1333FSB vs. 1600FSB doesn't make much of a difference, so I don't see much reason to get the QX9770 other than that it is 200MHz faster. The reason to get the QX9770 would be for the unlocked multiplier, otherwise, if you want 1600FSB the E5472 is practically identical--200MHz slower, but only 80w.
well cool, I had be agonizing over whether or not corssfire would work on a 5400 board but seeing as I have quattro now I'm sure they didn't write out support with the latest version of the software(whole reason I have to switch to intel within the next few months). I'll check to be sure though.
custom linux based on the latest kernel. It's kidna sad but, I don't even know what the latest kernel revision is. I haven't played with my own linux box since 2.6 had just come out and 2.4 was still the norm.
am i going to see any real world performance gains with 1600FSB? Why the 80W one specifically. I'm sure I have room for the extra $90 per cpu but, is there any real reason to go with the 80W besides a little savings on the electical bill?
A mac? I could probalby make myself buy a mac if it's honestly the best route but, a mac? really?
And it's not my money so I'm not worried about saving a few buck anywhere.
Originally posted by: nerp
Don't get too excited guys.
Now, when that machine is sold for scrap, that's when the real fun begins. . . but then again, we'll probably have octo or 16 core CPUs and 16GB standard by then. 2x8 or 1x16GB stick? hmm. decisions decisions.
It's sad but true. I enjoy my job, a lot but, it's not like I get to try and run Crysis on it or anything.
I have a dual opty worksation now, and love it, I wouldn't be upgrading if I didn't have to. While I'm upgrading though, I have a few goals in mind and would like to see a good performance gain. I'm an AMD fanboy here, I still run my San Diego 3700+ at home and I'm throwing my own money at a chilled water loop to get it run in the negative degrees to stretch it to last me a while longer. The system has been re-designed on a newer linux kernel and in the re-design they optimized for what the majority of the company is already running, intel.
Well its your software, you'd have to tell us if you want 8 cores trying to access FBDIMM ram on a FSB that is 1333MHz or 1600MHz. It changes the RAM speed too.
Skulltrail gives your 8 cores access to dual-FSB at 1600MHz AND let's you feed them cores with zippy DDR2-800 FBDMM with CAS3 timings.
This stuff should matter to you when you want performance from massively multi-threaded applications...right?Regarding an extra 200MHz for an additional $1200...you MUST normalize this analysis to a system-level performance/cost.
3,200/3,000 = 1.0667....i.e. the computer will be 6.7% faster with the 3.2GHz processors.
$10,000/8,800 = 1.136...i.e. the computer will be 13.6% more expensive with the 3.2GHz processors.
So you got to ask yourself (or your boss should be asking of you) if you are already going to plunk down $8,800 on a computer, would it be worth it to spend an extra 14% to extract an extra 7% more performance?
To be honest, if you answer the question "no" or if your boss answers the question "no" then you have actually concluded YOUR time is worth very very little to either you or your boss. 6.7% faster means 6.7% mor getting done, day after day after day. The only time a boss ignores a chance to extract 6.7% more productivity from their human assets is when those human assets are considered quite inexpensive.
Originally posted by: Idontcare
Well its your software, you'd have to tell us if you want 8 cores trying to access FBDIMM ram on a FSB that is 1333MHz or 1600MHz. It changes the RAM speed too.
Skulltrail gives your 8 cores access to dual-FSB at 1600MHz AND let's you feed them cores with zippy DDR2-800 FBDMM with CAS3 timings.
Regarding an extra 200MHz for an additional $1200...you MUST normalize this analysis to a system-level performance/cost.
3,200/3,000 = 1.0667....i.e. the computer will be 6.7% faster with the 3.2GHz processors.
$10,000/8,800 = 1.136...i.e. the computer will be 13.6% more expensive with the 3.2GHz processors.
So you got to ask yourself (or your boss should be asking of you) if you are already going to plunk down $8,800 on a computer, would it be worth it to spend an extra 14% to extract an extra 7% more performance?
To be honest, if you answer the question "no" or if your boss answers the question "no" then you have actually concluded YOUR time is worth very very little to either you or your boss. 6.7% faster means 6.7% mor getting done, day after day after day. The only time a boss ignores a chance to extract 6.7% more productivity from their human assets is when those human assets are considered quite inexpensive.
...which is probably why heyheybooboo was suggesting a multi-socket HyperTransport-based system. DerwenArtos12, do you have any way of determining what's limiting the performance of your application? Is it L2 cache size? DRAM bandwidth? SSE throughput? x87?
It's a good question, scratch that, it's a GREAT question.
Would you see higher performance with dual-socket quad-core (MCM'ed, so really its a 2x2x2 from a fine-grained software perspective) or from a quad-socket dual-core (again, 2x2x2 from software perspective) with this particular software?
Is the IPC for the specific software memory speed limited?
Given that he has already stated the code has been optimized for Core µ-arch by the inhouse programmers I am inclined to believe this is proof-in-the-pudding that they already know which scenario is best for their codes as surely they optimized it for Core chips for good reasons.
Originally posted by: CTho9305
Originally posted by: Idontcare
Well its your software, you'd have to tell us if you want 8 cores trying to access FBDIMM ram on a FSB that is 1333MHz or 1600MHz. It changes the RAM speed too.
Skulltrail gives your 8 cores access to dual-FSB at 1600MHz AND let's you feed them cores with zippy DDR2-800 FBDMM with CAS3 timings.
Regarding an extra 200MHz for an additional $1200...you MUST normalize this analysis to a system-level performance/cost.
3,200/3,000 = 1.0667....i.e. the computer will be 6.7% faster with the 3.2GHz processors.
$10,000/8,800 = 1.136...i.e. the computer will be 13.6% more expensive with the 3.2GHz processors.
So you got to ask yourself (or your boss should be asking of you) if you are already going to plunk down $8,800 on a computer, would it be worth it to spend an extra 14% to extract an extra 7% more performance?
To be honest, if you answer the question "no" or if your boss answers the question "no" then you have actually concluded YOUR time is worth very very little to either you or your boss. 6.7% faster means 6.7% mor getting done, day after day after day. The only time a boss ignores a chance to extract 6.7% more productivity from their human assets is when those human assets are considered quite inexpensive.
...which is probably why heyheybooboo was suggesting a multi-socket HyperTransport-based system. DerwenArtos12, do you have any way of determining what's limiting the performance of your application? Is it L2 cache size? DRAM bandwidth? SSE throughput? x87?
Honestly all I have to go on is what the guys who wrote it are telling me and it's not enough. Just from sitting at the workstation I would have to say that it's probably going to run like a cross between CAD and something like 3D studio max.
They didn't specify ram speed but, I know other workstation running the program are running DDR2-667 FBDIMMS with general success. I'd like to get DDR2-800 just for the sake of my sanity and to make sure I'm not going to be getting annoyed in a year or so when they re-compile and re-optimize again. Lengevity is really important here as I don't expect to get another system before my contract ends in about 23 months.
And I'm on contract and the $10K is gaurinteed to me specifically for a new build so I figure I've got it, why not use it. I'm working on multi-billion dollar projects here that will probably profit the company in the high hundred of billions to low trillions so, I really doubt they're going to quibble about $2K.
Skulltrail dude, 3.2GHz QX9775 FTW (even if only at stock clocks). You won't be sorry with the performance and you'll have e-penis like no tomorrow until Nehalem comes to town.
Honestly I'd be more comfortable with xeons but, this is a 771 system so it'll run xeons right?
Well, wait, I'm starting to do some research on skull trail and it's just a workstation board marketed to enthusiasts? Why don't I just get something like this supermicro workstaion board that uses the same chipset and has, as far as I can tell, every feature that skull trail sports and for like $200 less?????
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter