Min-Maxing a Digital Imaging PC

[jtam314msa]

Forewarning: This one's a doozie, guys. Lots of information here. But, if you want to gloss over the details and skip right to the build comparison and debate, scroll down to the "Build Comparison" header and "tl;dr" notation.

Here we go...

Min-Maxing a Digital Imaging PC for Developer Specifications

Topics include:

1. CPU/GPU pairing.
2. Haswell Pentium + DDR3-1333 vs. Trinity A4 + DDR3-1866
3. Whether i3 / i5 / i7 is necessary.
4. Developers padding their system requirements.

This post is aimed at beginning a discussion about the ideal components for a specific application. Through having this discussion, I hope to open up some debate regarding current tier hardware competing in similar price points, as well as shed some light on which components are ideal for certain tasks.

The build in question is intended for a program called Schick CDR DICOM; a CT scan image rendering application for dental surgeons. Here are the minimum and recommended specs according to Schick:

Minimum

1. i3 / i5 or equivalent
2. 4GB RAM
3. 1GB "video"
   Documentation simply states "video." Obviously they mean a 1GB GPU.

Recommended

1. i7 "or equivalent"
   Equivalent to an i7? Like what, another i7? Anyway.
2. 8GB RAM
3. 1GB video

As a final note, their documentation includes the following footnote:

"To achieve maximum image transfer-to-display time for Schick 33 Sensors, we recommend quad-core processors, 8 GB RAM, Windows 8 Pro or Windows 7, and 1 GB graphics card."
​

I don't know how much of this I believe. Anyone here who has read the minimum and recommended specs for any variety of software ought to know by now how unreliable they can be. Developers want some assurance that their product will perform to client expectations, so it's reasonable to assume that some extra padding goes into any list of system requirements.

Of course, that could be completely wrong. In order to reach the widest range of clients, developers may wish to make their product compatible with a wide range of hardware. I intend to contact the company tomorrow and speak to their IT department. Hopefully I can ask some detailed questions. Until then, we have this discussion.

tl;dr - i was hoping you guys might enjoy debating the merits of specific builds geared towards squeezing the most value possible from your dollar while maintaining reliable performance
​

Build Comparison

I intend to build five or six of these things for my colleague. The goal here is to maximize performance while minimizing cost. The first two builds I have in mind are as follows:

Haswell Pentium Build - $248.94

1. Fantastic price/performance Pentium.
2. Cheap, thoroughly tested GPU.
3. A well reviewed case/PSU combo with surprisingly few complaints, mostly related to poor packaging.

Trinity A4 APU Build - $246.94

1. An APU that compares surprisingly well with the G3220. Intel still has significantly better single core performance.
2. Fast, dual-channel RAM, perhaps to make up for any slack behind the G3220.
3. Paired with the same AMD GPU.

Debate Points

1. My thinking is that the APU + DDR3-1866 will yield a significant improvement over the Pentium with DDR3-1333. This may become even more apparent if I OC the memory to 2133 with the MSI board.
2. I'm also hoping that having the AMD GPU working in tandem with the APU will yield some stronger results, but I'm pretty sure it's not Crossfire compatible.
3. AMD also has a reputation for being more suited to encoding, rendering, and performing calculations than its Intel counterpart.

I'm hoping these three points will factor in to make the AMD build the stronger competitor. Both of these builds cost about the same, so ultimately I'm trying to determine which is superior.

Still, if CDR DICOM relies heavily on single-core performance, the G3220 likely wins outright. Even in that instance, it may turn out that the application is as demanding as Schick suggests. In that instance, I've already got several builds waiting in the wings for comparison. For now, I'm interested in seeing what /g/ thinks when comparing these two builds for their intended use.

If you've made it this far, here's a delightful little video for your trouble.

 

[videogames101]

When comparing across machines, you need to measure execution time for the specific task because everything else is moot. If you can get the software and examples of each machine it should be easy to decide. If you can't, you basically you need to profile the application in question. We could guess all day at which might be faster but without some kind of data on the application it won't be helpful.

As a complete guess, single thread performance will probably be more important, so I'd go with the Pentium chip.

 

[wsaenotsock]

..

 

[Phynaz]

Seriously, your colleague is looking to cheap out on a DICOM system?

Just get the i7 as recommended by the software provider. Care to provide the location of this dental surgeon? I want to stay far away from a guy that's so willing to invest so little in his practice.

 

[serpretetsky]

" Developers want some assurance that their product will perform to client expectations, so it's reasonable to assume that some extra padding goes into any list of system requirements."

You are correct, and it's usually a good idea to follow that "extra padding" for professional applications.

 

[piasabird]

Scanner does all the work.