Fast build on the bench

[ddrueding]

i7-980
Gigabyte G1.Assassin
Kingston HyperX 24GB DDR3-1600
Gigabyte GTX580
OCZ RevoDrive 240GB
Corsair AX750 PSU

I have it very slightly OC'ed (3.4Ghz) and significantly undervolted (1.225v CPU, 1.54v RAM) at the moment, undergoing stress testing overnight.

Anything anyone would like to see in particular?

 

[CougTek]

An FAH big unit's time per frame.

 

[Stereodude]

What did you build this machine for?

 

[CougTek]

My guess would be just because he could.

Another possibility is the i7 980 is necessary for fast multithreaded performances while switching from MS Word to Live Messenger and iTunes and the GTX580 is paramount to play games on Facebook.

 

[Mercutio]

I've got an i7 940 in my office right now that I'm trying to use explain to my co-worker why she doesn't need to specify full-blown SQL Server for this application she wrote and wants to re-sell. I basically put in the most ridiculous collection of parts I had on hand, including a 120GB OCZ Vertex2 to host guest images, and 48GB RAM.

One of the images is Server 2008 running on one CPU with 2GB RAM and SQL Server Express. The other is Server 2008 with 4 CPUs, 16GB RAM and full blown SQL Server. Performance for her application is absolutely identical to the best of my and her ability to test it.

Her response: "You just can't see the differences because this is a virtual machine and that slows everything down."

Headdesk. Facepalm. Headdesk.

 

[ddrueding]

A customer. I've only kept the rich ones who want something faster than Dell can sell them.

 

[Handruin]

I've got an i7 940 in my office right now that I'm trying to use explain to my co-worker why she doesn't need to specify full-blown SQL Server for this application she wrote and wants to re-sell. I basically put in the most ridiculous collection of parts I had on hand, including a 120GB OCZ Vertex2 to host guest images, and 48GB RAM.

One of the images is Server 2008 running on one CPU with 2GB RAM and SQL Server Express. The other is Server 2008 with 4 CPUs, 16GB RAM and full blown SQL Server. Performance for her application is absolutely identical to the best of my and her ability to test it.

Her response: "You just can't see the differences because this is a virtual machine and that slows everything down."

Headdesk. Facepalm. Headdesk.

What's the virtualization software you're using for these VMs?

 

[Mercutio]

What's the virtualization software you're using for these VMs?

ESX.

The reason there's no performance difference is that her application is basically just a web based time card system. Even if 200 people all submitted their forms at once, it's still just maybe 50kB per transaction.

 

[Handruin]

A customer. I've only kept the rich ones who want something faster than Dell can sell them.

Sounds like a nice and fun machine. There is nothing specific I'd want to see from it, but would be curious on the FAH time for a big-unit like Coug requested.

 

[ddrueding]

At the moment I don't have an OS on it...I'll get the F@H time once I finish the OC/voltage testing.

 

[Handruin]

ESX.

The reason there's no performance difference is that her application is basically just a web based time card system. Even if 200 people all submitted their forms at once, it's still just maybe 50kB per transaction.

Sounds like the application is in no way CPU-bound, but likely something else. In most cases multiple vCPUs don't make that big of a difference unless if you're doing something highly optimized for multithreading. I've often times seen performance go down with more vCPUs. The problem her application could face is running into the SQL Server express storage limit of 10GB, but that might not be for some time at 50KB per transaction. Are both of your VMs using the VMXNET3 NIC vs the E1000?

 

[Mercutio]

Yes, but I'm not testing them at the same time, either.

 

[Mercutio]

I think she's basically working from the assumption that expensive, not-free SQL Server is better than free SQL Express, and no matter what I show her, she isn't going to be convinced.

But that also means that the cost for her product is going to be massively higher for the database and CALs. Which is ridiculous given the nature of the application.

 

[Handruin]

I think she's basically working from the assumption that expensive, not-free SQL Server is better than free SQL Express, and no matter what I show her, she isn't going to be convinced.

But that also means that the cost for her product is going to be massively higher for the database and CALs. Which is ridiculous given the nature of the application.

That does sound crazy. Can she decouple her product from the database and let customers use their own SQL server instance if they have one? Shipping the product with SQL server sounds like a pain. Is she managing their backups, maintenance, etc...?

Unless she's doing something unique in her coding to require SQL server, she may be better-served with a free database, something like PostgreSQL. We've moved many of our products off Oracle and onto PostrgreSQL and have had positive results.

 

[CougTek]

What about MySQL? Is it still free?

 

[Mercutio]

What about MySQL? Is it still free?

She really only knows how to deal with Microsoft products. I've been beating over the head for years that she needs to get her head out of Access and VB and SQL Server, and she did finally break down and build an ASP.NET application, but every time I bring up working with Python or PHP with a MySQL backend so that her work can actually be widely portable, she tells me she's too busy to learn any of that stuff.

 

[Handruin]

What about MySQL? Is it still free?

Yes, it's licensed as OpenSource under the GPL, but it is now under the hands of Oracle.

 

[timwhit]

Yes, it's licensed as OpenSource under the GPL, but it is now under the hands of Oracle.

If Oracle does anything really stupid, the community will move to one of the forks en masse. See LibreOffice as an example.

 

[LunarMist]

Meh. I've had one for going on a year now. Six cores are nice for LR, PS, PTGUI, etc., but the CPU really pumps out heat under load when the voltage is increased to reach 4.3 or 4.4GHz. If you are going to run that night and day maybe 4GHz is about right.

 

[ddrueding]

Well this one is at 3.5Ghz and 1.1v stable. Still testing the rest, but heat production is way down.

 

[LunarMist]

I think mine is about 12.5V or higher.

 

[ddrueding]

Yeah, that is a lot of electricity.

 

[LunarMist]

It is higher, 1.325V.

 

[ddrueding]

IIRC, wattage increases at the square of the voltage? I know it isn't linear...

 

[time]

Yes, voltage is the killer.

For the same architecture and configuration:

(4.3GHz / 3.5GHz) * (1.325V / 1.1V)² = 78% more power

 

[LunarMist]

That would be fine for resistive loads, yet I suspect the actual power drain is not such a simple equation.

 

[ddrueding]

I'm fairly certain that voltage and clock speed scale with load, but for a certain voltage and clock speed, I suspect it is just a resistive load.

 

[LiamC]

I've been working with Oracle and DB2 for years (and I've also done SQL Server). Microsoft advocates as a general rule (there are exceptions) are dogmatic about Microsoft because they usually have no experience working on anything else. It scares them and they are afraid to admit that they don't know it all.

The latest version of SQL Server has come a long way I understand. I've not played with it myself, this just comes from people whose opinions I trust. A long way in the sense where it's still not up to Oracle or DB2, but "good enough" for the vast majority of cases. And it is simpler to administer.