Page 1 of 1

EPIA & Me (part 1) – An Adventure in Small Form Factors

Posted: Sat May 06, 2006 9:37 pm
by eitje
Crosslink - Intro, Part 2

Whenever I see reviews of the EPIA boards, I never see gaming benchmarks. Whoever it is writing the review points out how fast a file could be encrypted, how much CPU was used playing back a DVD, how long it took to copy files from one place to another. All of which are excellent measures, as long as all you plan on doing is encrypting your DVDs and transferring them places.

I want to see how these boards will perform as a standard desktop. We already know they're sufficient for those applications mentioned above, but what about games? What about raw performance?

We'll start with a look at some artificial benchmarks from SiSoft Sandra 2004 SP2b and 3DMark2001.

Image

Kicking things off, our 5 year old system is beating the snot out of the much newer EPIA boards. With a theoretical bandwidth cap of 2100 MB/s for the Duron & MS10000 systems which are using DDR266 memory, 30% ALU bandwidth is not an impressive yield for the MS board.

More importantly, the SP13000 is using DDR400 memory on a faster front-side bus, but only shows a marginal improvement on the MS10000. If we compare actual to peak theoretical, with the SP board only getting about 21% of maximum, we're actually taking a step back. That indicates that the EPIAs have a bottleneck somewhere – probably the C3/Eden CPUs.

Image

Image

The results from the arithmetic & multimedia CPU benchmarks seem to agree with that hypothesis, as the 700 Mhz Spitfire continues to whallop the competition. Although the SP does show marked improvement in the multimedia results, its C3's arithmetic FPU performance does nothing to seperate it from the Eden processor in the MS10000.

Overall, it seems like just a little more speed is necessary to push the C3 processor up to the same level as the Duron – perhaps this is something that's been resolved with the inclusion of the 1.5 Ghz C7 in the latest EPIA boards.

Posted: Sat May 06, 2006 9:39 pm
by eitje
Image

First, a note about laziness. Now, you may not have noticed, but there's a lot of information in the picture above. I certainly could have taken the time to split it all out, and make it graph real nice. I could have even worked through each standard resolution setting and recorded all of the results. But, instead, I got lazy, and I decided to just barf the values onto the screen in one big heave.

Now, for the real dig-in – maybe I'll just sort these results in order of awesomeness, and see what falls out. And maybe add some colors to speciate the video cards – just pretend the grey is chrome!

Image

Well, there's definitely a few things that quickly become apparent. First is that ATI & Nvidia are – as would be expected – in a completely different class compared to the trailing XGI & VIA offerings. Second, and more surprisingly, is that the discrete XGI card barely beats out the integrated VIA video solution.

Another point to make about the XGI offering is that, in the case of the SP13000, the system would actually reboot during testing; I think it occurred while transitioning into Nature. I tried a number of different configurations, as well as replacing the standard 60W power supply with a beefy 300W beast, but all to no avail.

Fnally, it's worth mentioning that we can see the SP13000's first swipe at the Duron system – when combined with the ATI 9250, it comes within a mere 5 3dmarks of a tie. Nvidia's performance on the EPIA platform leaves something to be desired, but this is an artifical benchmark after all; maybe they'll be able to pull a switcheroo in actual gameplay.

Image

Image

Image

Which brings us to the sad part of this story. While bechmarking, I was also testing out a variety of 40mm fans for the SP13000. :$ I know, I know, one thing at a time. Bad tester! Bad! Anyway, the moral of the story is, don't leave a benchmark looping while you go to dinner if you... forget to put a fan back on the system. So, we don't have gaming benchmarks for the SP13000, which almost made me call the whole thing off, but I decided to go ahead with what I have.

Also of note is the lack of a measure for the Unichrome Pro. Apparently, Doom 3 thinks...

Doom 3 wrote:
ERROR: The current video card / driver combination does not support the necessary features.

Posted: Sat May 06, 2006 9:39 pm
by eitje
With that in mind, let's look at these in order of average framerates, and see if we can extrapolate anything from the numbers. Consider this more a view of the graphics cards compared to one another.

Image

First, Doom 3. Looking over past reviews on Tech-Report that used Doom 3, i see verbs like “romps”, “owns”, and even just “beats” in use - none of which really apply to this result set. Let me tell you, playing Doom 3, even at 640x480, was a pain. For your consideration:

“Oh my god, a zom... bie! Quick! Shoot..... it!” *pull trigger* .... ..... ...bang!

Needless to say, I wasn't sad that the integrated video wouldn't run. And the only reason XGI was playable was because i spent most of my time running around staring at walls and the floor. The good news is, the MS10000 pulls in front of the Duron system; I can only imagine how the SP13000 would have performed.

The MS10000 had some interesting problems with both the Nvidia and ATI cards. Using Nvidia, there were problems with static and popping in the audio. Whatever was causing that may well have also given the MS+NV combo its edge over the competition. Also, MS+ATI had some serious artifacting going on along the walls and floors (brightly colored circuit-board patterns). Improperly rendered polygons also might have given the MS10000 an edge over the Duron.

Image

Next, Quake 3. It's a pretty close race in this one. Honestly, I believe that the variance in the average framerate in the Quake 3 tests is close enough to tolerance to say each combination offers “pretty close to equal” performance. The minimum framerates are nearly equal, and max framerates would stick at 92 FPS on all platforms, so I don't really think the motherboards, processors, or cards were really a limiting factor. I WAS using the demo version, so perhaps it's frame-capped.

Nonetheless, again, ATI and Nvidia stand ahead of XGI & VIA, even in a game that's so old the source code has been released.

Image

Finally, Unreal Tournament 2003 was “meant to be played”, but NV still loses to ATI on both platforms, but generally by a narrow margin. Poor XGI... when paired up with the Duron, it even gets beaten by the Unichrome on the MS. Again, however, the EPIA board takes care of the Duron system in real gameplay, giving only a hint of what the SP13000 could have done in its place and beating that “artificial benchmarks” horse a little more.

Conclusions

Posted: Sat May 06, 2006 9:40 pm
by eitje
At this stage, it's hard to recommend an EPIA board as a standard desktop solution. It's a bit disconcerting that (until recently) the best VIA had to offer for SFF enthusiasts was only marginally competitive with products that were launched 5 or more years ago.

Also, I'd never owned an FX-series card, so I wasn't prepared for how hot it gets. On my testbench (read - “spare bedroom's floor”), it was fine, but I don't know what might happen to the card/system in a fanless case. In Windows, at least, there doesn't seem to be a clear winner between ATI & NV, so I'd probably recommend the Radeon 9250, unless you critically need DX9 functionality.

In the end, though – with a little assistance from a discrete graphics card – the EPIA boards can become an expensive, low-power Windows box capable of a basic level of gaming. It won't be a performance-buster, and it's certainly not something that you'll see on the front page of the Tech Report any time soon (in fact, you'll have to go back 3 years to find the last time an EPIA was an article target). However, I love that I can have a computer that I can't tell is on just by walking into the room. So, I think EPIAs will continue to be a fixture in my geeky world.

Addendum:
Just for kicks, I took the 60W power board and plugged it into the Duron system to see what would happen. At first, it couldn't boot; it would POST, detect drives, but then reboot when trying to go past that. However, once I disconnected the CDROM drive, it was able to power on and boot into Windows. It even was able to do some browsing and drive access, although I didn't try to run any games on it. Given that Morex also has a 90W and 120W adapter kit, I think I could probably more than just basic functionality out of the Duron system – but that's outside of the scope of these posts. ;)

Re: Conclusions

Posted: Sat May 06, 2006 10:35 pm
by bitvector
eitje wrote:
Whenever I see reviews of the EPIA boards, I never see gaming benchmarks. ...
I want to see how these boards will perform as a standard desktop.

eitje wrote:
At this stage, it's hard to recommend an EPIA board as a standard desktop solution.

Is gaming really a "standard desktop" function? A basic desktop computer is expected to do stuff like web browsing, email, word processing and possibly some light media tasks (play music, maybe movies, store photos, etc.). I certainly don't consider gaming a "standard desktop" task.

You've created a lot of threads on this topic, but ultimately gaming is a non-factor for EPIA's targets. It's a small, low-power solution where power and size are more important than performance (like Crusoe). When performance is de-emphasized, gaming generally goes out of the picture. I have a VIA C3 machine that I built not for performance, but because I had to cram it into a non-ventilated hall closet and an underclocked Duron kept overheating.

Re: Conclusions

Posted: Sun May 07, 2006 7:35 am
by eitje
bitvector wrote:
eitje wrote:
Whenever I see reviews of the EPIA boards, I never see gaming benchmarks. ...
I want to see how these boards will perform as a standard desktop.

eitje wrote:
At this stage, it's hard to recommend an EPIA board as a standard desktop solution.

Is gaming really a "standard desktop" function? A basic desktop computer is expected to do stuff like web browsing, email, word processing and possibly some light media tasks (play music, maybe movies, store photos, etc.). I certainly don't consider gaming a "standard desktop" task.


i understand that most people don't want to run the latest and greatest, but i think most people do at least a little light gaming on their main PC. for some people, that's party poker. for some people, maybe it's just warcraft 3. the point is:

In the end, though – with a little assistance from a discrete graphics card – the EPIA boards can become an expensive, low-power Windows box capable of a basic level of gaming.


also, i think most people don't keep their desktops in a closet. ;)

i appreciate your opinion that EPIAs are not intended for gaming, but my purpose was to demonstrate that they COULD game, to some degree.

So, I think EPIAs will continue to be a fixture in my geeky world.


thanks for the feedback! :)

Re: Conclusions

Posted: Sun May 07, 2006 8:22 am
by cheesyking
eitje wrote:
i understand that most people don't want to run the latest and greatest, but i think most people do at least a little light gaming on their main PC. for some people, that's party poker. for some people, maybe it's just warcraft 3. the point is:


Actually I would say that people who use PCs for any sort of gaming (beyond non 3D stuff like party poker) are a tiny minority of all PC users. For most people the first 3D they've ever used on a PC is going to be Google Earth which even runs ok on the onboard graphics those boards have.

Can they be used for basic gaming? Of course, but then so can a Duron 700.

So why spend on an EPIA system when you can get something old bit of junk off ebay that's just as fast but cost far less? Because you need small and cool, which is the reason we knew about in the first place.

Anyhow here's another benchmark for you:

PII333 + FX5700 = (approx) 1000 pts in 3DMark2001se

I did that a couple of years ago so I can't remember the exact numbers but adds a little to your argument I suppose.

I seem to remember there being a gaming benchmark done on all processors from PII up, was that here (or, dare I say it... Toms)?

Posted: Fri May 12, 2006 5:00 am
by mrfixitx100
"So why spend on an EPIA system when you can get something old bit of junk off ebay that's just as fast but cost far less?"

Why not just go to the landfill and rescue an old P3? You want "green", that will give you "green", rather than buying more stuff that will find itself in the landfill in short order?

"Because you need small and cool"

So, underclock the "free" P3? (that will give you cool)
Buy a Palm Pilot? (that will give you small)

Here is another benchmark for you:
P2@233, FX Intel board, 128m of EDO simms, Radeon 7500 (PCI)= 527 pts in 3DMark2001se (print server)

That kind of gets a little closer to what you see with your 1000.