Dual i7's?

E

Esseti

n00b
Joined
Jan 2, 2008
Messages
8
Is there a mobo (apart from a server mobo) that can run dual i7's? I'm asking, because in the next few months I'll have the opportiunity to buy a PC for about $5k, and I would like a lot of horse power under one hood for particle simiulation. I know that PhysX uses the GPU, so I'll probably buy 2x 295's.

Thanks.
 
Not currently... but Intel might put out a Skulltrail 2.0...

Let's hope they don't use proprietary $1500 processors again though... eh?

And yea, I know you could use Xeon procs.. but at the time it was released I think the fastest one was the QX9775

I don't think it should be an issue though, because server X58's / I7 are based on 1366 anyways... Skulltrail was based on a server platform... who knows
 
Blazestorm said:
Not currently... but Intel might put out a Skulltrail 2.0...

Let's hope they don't use proprietary $1500 processors again though... eh?

And yea, I know you could use Xeon procs.. but at the time it was released I think the fastest one was the QX9775

I don't think it should be an issue though, because server X58's / I7 are based on 1366 anyways... Skulltrail was based on a server platform... who knows
Click to expand...

Blazestorm, you HAVE to use server procs since they're the only ones that have more than 1 QPI link. Retail i7s (920, 940, 950, 965, 975) have only a single QPI link. To have dual processors you would need to have procs that have dual QPI (same goes for quad). So yeah, you're stuck buying crazy expensive procs again if you want to have top of the line equipment.

Esseti, if you're particle simulation is CUDA compatible, it may be way more feasible just to get the ASUS workstation x58 mobo with 4x PCIe, an i7 920, OC the thing to around 3.5GHz and pair it with quad 295s. As long as you have cool ambient temps and a well ventilated case, that should out perform even a hypothetical quad i7 system. FWIW, using Folding@Home a 295 yields around 14,000 points per day while an i7 around 4GHz will do around 10,000 points per day. I know that CPU and GPU points are calculated differently, but even Stanford says that their GPU client seriously outperforms their CPU client, and your work if ported to CUDA should result accordingly.
 
Why not use a server platform? You can get a duel 1366 board from Supermicro with 2 x16 PCIe 2.0 slots, which should fit your graphics cards (this board).

But as mentioned above, if your simulation uses CUDA then forget a big processor and get 4 GTX-295s instead. You might even be better off saving money & going s775 and spending the cash you save on a better storage system (either SAS disks or a couple of SSDs depending upon the type of data involved).
 
if your simulation uses CUDA then forget a big processor and get 4 GTX-295s instead.
Click to expand...

This. If you’re massively parallel, then 4x 295s is going to seriously outperform any CPU based setup. Honestly, a single 260 could outperform a stock i7 in some cases.
 
Well, PhysX is used just in a handfull of 3D applications now, but it's catching on. While symulating particles with GPU's is more cost effective, the downside of that is I won't have much power to render out my scenes. I think I'll have to go 50/50, 1x i7 and mayby 2x 295.

One more question, is it worth buying the Extreme i7? Or should I just overclock the base ones?
 
Blazestorm said:
Not currently... but Intel might put out a Skulltrail 2.0...

Let's hope they don't use proprietary $1500 processors again though... eh?

And yea, I know you could use Xeon procs.. but at the time it was released I think the fastest one was the QX9775

I don't think it should be an issue though, because server X58's / I7 are based on 1366 anyways... Skulltrail was based on a server platform... who knows
Click to expand...

They weren't proprietary processors. They were standard LGA771 server processors with an unlocked multiplier. They were Core 2 Extreme Edition CPUs in name only. They are physically closer to Xeon's of the day technically speaking. The standard Core 2 isn't SMP capable. They QX9775's work in any Intel 5400 chipset based board that supports 400MHz FSB CPUs. The Skulltrail board itself used the Intel Seaburg 5400 chipset and had dual nForce 100 MCP's onboard to multi-plex the PCI-Express lanes and give it support for SLI as NVIDIA still locked out consumer SLI on non-NVIDIA chipsets at the time.

There is no such thing as dual processor Core i7's. There are however Xeon 5500 series processors which are based on the Nehalem core (and are quite similar to the Core i7). These require the Intel 5520 or 5500 chipset and a motherboard with dual sockets to utilize them.
 
Esseti said:
Well, PhysX is used just in a handfull of 3D applications now, but it's catching on. While symulating particles with GPU's is more cost effective, the downside of that is I won't have much power to render out my scenes. I think I'll have to go 50/50, 1x i7 and mayby 2x 295.

One more question, is it worth buying the Extreme i7? Or should I just overclock the base ones?
Click to expand...

Rather than build 1 compromise rig, you might be better placed (if you can get the budget) to build 2 machines. 1 with massive gpu power & the other with a couple of Xeons.

But what software will you be using now? If it currently runs on a gpu, buy big gpus. If it currently runs on a cpu, buy big cpus. No point in buying for software that might exist in the future - buy the hardware in the future to suit the software you will be using, as it will be cheaper and more powerful then. Big software companies are notoriously slow at releasing versions for new hardware - hence the reason why there is so little CUDA software around. And CUDA may well have been superceded by then anyway.
 
the_beast said:
Why not use a server platform? You can get a duel 1366 board from Supermicro with 2 x16 PCIe 2.0 slots, which should fit your graphics cards (this board).

But as mentioned above, if your simulation uses CUDA then forget a big processor and get 4 GTX-295s instead. You might even be better off saving money & going s775 and spending the cash you save on a better storage system (either SAS disks or a couple of SSDs depending upon the type of data involved).
Click to expand...

Better off with this board: http://www.supermicro.com/products/motherboard/QPI/5500/X8DTH-6F.cfm
 
The problem is using server chipsets to get dual processors would not let you have the desktop features you want.
 
On a workstation you would be ill-advised to overclock anyway. Total stability & computational accuracy are your primary aim. Speed has to be a second priority.

Server chipsets = more thorough testing, support & reliability

Desktop chipsets = allow running the hardware out of spec.

What exactly do you lose out on?
 
I think in your current price point you are looking it is a interesting situation to be, there a multiple solutions that could work well. The real question what applications are you using and will they benefit more for gpu acceleration or cpu acceleration? Additionally do you want to have to deal witht he complex setups that can come from going with eatx ( server size motherboards chassis and all that those system entails?

I would also like to show currently most applications that do have gpu accerlation seem to currently have a bottleneck in the ability to take advantage of all the sp's on the beefier cards ( such as the 285 or 295 ) So you may be better looking at go with a high performance gpu accerlator that is low cost ( such as the 9800 series cards )

There are 2 great platforms to take a look at that also happen to be atx form factor

http://www.asus.com/product.aspx?P_ID=JoOxbkv8HgVqHLrz

http://www.asus.com/product.aspx?P_ID=9ca8hJfGz483noLk&templete=2
( just imagine an app taking advantage of 7 gpus plus the power of an i7 );):D:)


Lastly i would strongly agree witht the beasts statements that you do not want to overclock on a workstation based platform
 
You must log in or register to reply here.
Back
Top