Intel Skulltrail Benchmark and Analysis

Tom's Hardware has a detailed benchmark and analysis of Intel's new Skulltrail offering, taking a look at 8 vs 4 cores. The comparison uses games, A/V applications, office applications, and 3D rendering tools to help demonstrate benchmarks. "We were disappointed by the Skulltrail platform. Although we have tested and reviewed numerous Intel products, we have never had such a half-baked system such as this in our labs. If this sounds harsh, bear in mind that all we have to base this conclusion on is the Skulltrail system itself in its current state, which Intel provided as an official review platform. We do not know whether Intel plans to revise and improve the platform before the final versions ship to retail."

A question

[warrior_s]

that is very important

Are these games and benchmarks actually making.. you know.. use of all the 8 cores? i.e. were they modified so that they can make use of multicores efficiently.

Multicore machines are useful when either you run multiple applications or if you want to run single app and make use of the cores, then the apps have to be updated so that they can make use of these multiple cores.

Parallel programming now!

[TheSync]

Guess what guys? We've run out of GHz (mainly a power/heat problem). Start writing parallel programs.

Here is what the article says:

To be fair, though, it is not Intel's hardware that is at fault here, but today's software. If a program only uses four of the eight processor cores, then the Skulltrail system is noticeably slower than a single-socket quad-core computer. Since there are practically no current games or desktop applications around that can utilize more than four cores (if that many), the Skulltrail system does not offer any benefit here.

Read The Landscape of Parallel Computing Research: A View From Berkeley [berkeley.edu] which has the description of why, this time, there is no getting around parallel programming.

Also examine NVIDIA's CUDA [nvidia.com] platform, which scales from a handful of processors on your PC's NVIDIA chip to the 128 processor NVIDIA Tesla [nvidia.com] card. Scalable parallel processing is the future.

Yes and no

[wandazulu]

I'm not familiar with writing games on a multi-core game system a la the PS3, but I have written multi-threaded apps in Windows and I can tell you that the answer is:

Maybe.

The problem is that your app might be multi-threaded up the wazoo, but you're at the mercy of the OS (Windows here) to actually put the threads on separate processors/cores. You can *request* a thread on a separate processor (SetProcessorAffinity(), if I recall..it's been awhile), but the docs state that this is merely a request, and the operating system is free to ignore it if it thinks it can do better. A lot of time I observed that Windows doled out threads to other processors very grudgingly, and I was told that it's because to Windows, the overhead of keeping track of what thread is on what processor was, under a lot of circumstances, more expensive (read: slower) than if it just kept them all on processor 0 and just context-switched (which it was going to be doing anyway)

Most games have been, as I've seen, multi-threaded for awhile now; the complexity of these games means they'd have an event loop that's a million lines long if they didn't (and probably do anyway), but your performance is always going to be only as good as the hardware, and the operating system, let you.

3D Rendering...

[podperson]

The only real test to show the benefit of Skulltrail was the 3D rendering section where the Skulltrail machines really did post decent results. Even for video encoding you reach a point where the problem becomes IO-bound (and you can't compress video frame n independently of video frame n+1 because of interframe compression). Of course, the next question is whether a Skulltrail machine is cost effective against slightly cheaper machines used in parallel for 3D rendering.

No, they aren't

[Sycraft-fu]

At this point 2 cores is about all you'll really useful in a gaming rig. A lot of games are still single thread, especially old ones. However there are a good number of games out there that can make efficient use of 2 cores. Past that, it gets questionable. There are some games that claim quad core support, but in general it seems they don't make efficient use of it yet. Thus far, I've never seen any game that claims 8 core support, much less any benchmarks to back it up.

I think this is mostly targeted at the "My ePenis is bigger than yours," crowd. There are a non-trivial number of people out there who are willing to just drop obscene amounts of money on gaming rigs, and Intel wants to suck every dollar they can out of their pockets.

Same sort of deal with nVidia's new triple SLI boards. At this point even 2 card SLI isn't a great idea because it costs so much (literally twice what a single card does) and the benefits aren't that great. There isn't a lot of need for 3 card SLI. However, people will spend the money, so nVidia will happily make a product to take it from them.

Why it works for video compression

[MosesJones]

Most video compression approaches use Keyframes which are uncompressed (across frames) in order to make sure the compression doesn't drift to far from the actual content. So doing Multi-core is actually pretty easy on video as you just dedicate a core as working on one key-frame to key-frame section. Given that key frames often occur as much as once a second (or on decent connections once every three seconds or so) then there is a huge amount of work that could be done in parallel and its not very difficult to make the encoders work in that way.

So video compression isn't one of the areas where it isn't an advantage to have multi-cores.

Re:Parallel programming now!

[h4rm0ny]

It's going to depend on whether those ten applications are actually making ongoing use of your processor. Encoding a movie whilst listening to music and editing photos - yes, proper use of multiple cores will see big benefits. But if you're talking about some spreadsheets, word documents, browser and an email client, then less so because no matter how quickly you think you're switching between these applications, it's going to look like slow motion to a CPU swapping processes. With this sort of usage, a CPU is actually sitting idle a lot of the time waiting for the next eternity between keystrokes to end. I'm not saying you wont see a benefit, but the benefit really kicks in when you've got multiple applications that are really doing something. A lot of applications (and probably the ten you have open at work) simply don't fall into that category.

Re:Yes and no

[afidel]

Windows will happily keep everything on processor 0 until such time as a scheduling threshold is reached on processor 0 at which time it will move the thread to another processor if available. It will continue to use the same process until all of the processors have a full load. I imagine in the average case of a desktop system this probably IS the most efficient algorithm, but if you have lots of short lived high resource consuming threads it's probably not due to all the state copying going on. Also in Windows 2003 the kernel is aware of memory locality and so will try to keep processes on the processor closest to their largest pool of memory in a NUMA system. Also the reason that affinity requests aren't hard is that otherwise it would have to throw an error if that processor wasn't available either due to hardware issues or due to the process attributes being set so that it can't see that processor.

Re:A question

[milsoRgen]

Are these games and benchmarks actually making.. you know.. use of all the 8 cores?

No they are not, the article goes on to say the 2nd processor is basically left unused and even current quad core designs are out performing skull trail.

The problem lies in the fact Intel released this platform as a gaming platform. However they reached into their workstation kit to pull out this hardware. Dual processors are a nice bragging right for enthusiasts, but only if the performance is in the very top tier with software actually in use. And using fully buffered memory, is simply a big no-no when it comes gaming rigs. On top of that, there are no BIOS options for overclocking, something any respectable board designed for even modest gaming comes with these days.

The article goes on to state that this is supposedly a review grade board, but has some serious stability issues and consumes far more electricity than the performance justifies.

AMD has faced a similar problem with its Quad FX line (aka 4x4). It's thermals were also way above the norm with dubious performance benefits as well. However at the time of it's release it was holding it's own (as best AMD can mangage these days) against the current Intel top offering. But just one look at the thermal images of a system built around AMD quad leaves you wondering if it's going to spontaneously burst into flames. However at least AMD had the sense to not implement FB-DIMM's in a gaming platform.

In short, it looks like a really half baked package for gaming enthusiasts, which might of worked if it had overclocking support, DDR3, didn't crash, and could outperform AMD and Intel's top quad core parts. It just leaves one wondering why? Intel is the current performance leader it makes no sense to release something like this to the reviewers.

Re:Page 1 of 25

[Mad Merlin]

At the bottom of the linked page I saw "Page 1 of 25" and I gave up. Bad submitter! Bad! Bad!

Tip: add print.html to the end of any THG URL, and you can read the entire thing on one page. THG would be completely and utterly useless otherwise...