QPI 5.8GT/s vs 6.4GTs/ ... noticeable difference?

Maverick2002

Diamond Member
Jul 22, 2000
4,694
0
0
We're pricing out some Xeons for a new server and are looking at the following two processors:

E5540 (2.53GHz, 5.8GT, 80w) - $825
X5550 (2.66GHz, 6.4GT, 95w) - $1050

Besides the speed (GHz and FSB), the only other performance difference looks to be a faster QPI. Since we'll be running these in dual socket motherboards, will the extra QPI bandwidth make that much of a difference? I can't seem to find much (any) performance data comparing the two. I just want to know if the $200+ price difference is worth the slight (?) bump.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
It's entirely going to depend on the specific application and usage patterns you intend to task the 2S system with.

The extra QPI bandwidth between sockets is going to matter if you have rather high demands on the interprocessor communications bandwidth in your specific usage patterns.

Whoever makes your software app should have a sales team that wants to be able to answer this kind of question for you, and they are the best place to start since the only people who will know your usage demands better than yourself is the people who made the software.

My suspicion is that the extra clockspeed of X5550 will provide you more a performance boost in the vast majority of applications versus the extra bit of QPI bandwidth that comes with the X5550.
 

DominionSeraph

Diamond Member
Jul 22, 2009
8,386
32
91
From a cost-of-hardware standpoint alone, the E5530 or 5520 would be give better price/performance.

E5530 (2.4GHz 5.8GT 80w) - $540
E5520 (2.26GHz 5.8GT 80w) - $380

(I can say this with confidence because it's such a large drop in price over the E5540 for only a 5/10% clockspeed drop)

But the cost of hardware in a server environment is often negligible compared to other costs. I mean, saving $400 on a $5k server that's running $$$$ in software which is heavily used by an employees with a payroll of $15 million wouldn't make much sense. A 1% increase in worker productivity would be worth $150k.

As far as the 10% increase in QPI between E5540 and X5550, as Idontcare pointed out, it depends. Same goes for the increase in supported DDR3 speeds from 1066 to 1333MHz. If you're burning up those CPU's, the increase in QPI could give a nice little boost of a few percent. If you're bottlenecked by memory, the increase in RAM speed would probably be worthwhile. But if you're storage-system bottlenecked, even a single E5520 could be massive overkill.
 

Maverick2002

Diamond Member
Jul 22, 2000
4,694
0
0
Thanks, I guess there really is no clear cut answer. I think we're going for the E5540 chips. You'd be surprised but a few mhz speed difference actually ends up being noticeable in our simulations and this will be a 16cpu/32 thread box (2 boards connected via quad infiniband).
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
You're throwing quad infiniband $ at this thing but shying away from the higher clocked cpu's?
 

Maverick2002

Diamond Member
Jul 22, 2000
4,694
0
0
In our case we can't justify a $1200 bump (4 cpus) for a 100mhz increase. The quad infiniband doesn't end up costing us much more than a normal rack would. We're ending up with a $7000 box that is 1U, 2 node, 4x E5540, QDB Inifiniband, 4TB total storage, 24GB total RAM. It actually came in at less than what we were thinking so we're probably going to order 2. Keep in mind that our QDB is rated at max 40Gbps which still makes it the bottleneck (we do have to write out result files, but it's like 500mb every 10 minutes so it's nothing a normal drive can't handle).
 

DominionSeraph

Diamond Member
Jul 22, 2009
8,386
32
91
Originally posted by: Maverick2002
Thanks, I guess there really is no clear cut answer.

Well, not when we have to make wild assumptions as to the environment.

X5550 has more MHz, faster QPI, DDR3-1333 memory support, and better Turbo Boost (2/2/3/3 instead of E5540's 1/1/1/2 [for 4/3/2/1 cores active]). How much of this comes into play? Can't say with what you've told us.

You've said the MHz makes a difference, so we can check that off for X5550.
That brings us up to the Turbo Boost, so now there's the question of the thermal environment: Do you have the cooling to get either the 2 or 3 bin increase? Is it likely the E5540's 2 bin single-core increase ever come into play? (latter is doubtful just on the surface)

Faster QPI? Well, what are you running? Do the processors have to share large quantities of data?

DDR3-1333 support? Would you install it or be using DDR3-800/1066 instead? Just how memory intensive is your app?