Dual-Core or Faster Hyper-threaded (What does CAOS Think?)

[R. Andrew Lamonica]

The CS Dept. is about to buy some new computers for EB2029 (and other places) and I was interested in CAOS’s opinions and experience with Dual-Core Technology.

With the specifications what they are, we can buy either Computers with:

2.8Ghz Intel Pentium D Processor 820 (Dual-Core) with 2x1MB cache

Or (the slightly cheaper)

3.2Ghz Pentium 4 Processor 640 (Hyper-threaded) with 2MB of common cache

As you can see, the Dual-core technology is more expensive and splits the cache between the â€Ã...“coresâ€Ã,. Potentially, making single-threaded programs (like many of the ones CS students write) run slower.  However, this same technology is supposed to make everyday use of the computers and multi-tasking run much faster.  Additionally, Intel is moving away from the older Pentium 4 Processors so it might be best to invest in the newer stuff.

I have not personally had any experience with Dual-Core computers and I like my old HT computer so I figured I would ask CAOS what they have seen.  I am particularly interested in hearing from anyone who has used both DC and HT computers and has noticed a difference.  I am NOT interested in starting a debate about Intel vs. AMD or Dell vs. Compaq or anything like that.

[Geoff Schreiber]

First, I think the 640 processor is about $15 dollars higher than the 820 on average...

I think they should hold out for the 840 :) Solves all problems.  Dual-core with Hyper-threading!

I've been happy with HT, but also, have not worked on a dual-core machine yet for any extensive time period.  If execution speed of the CS programs is the only problem, I don't see why it would matter - most of the programs developed are such small snippets of cpu time anyways, it usually wouldn't make a difference.  Where it would show would be upper level OS/CG classes most likely...

[Justin Camerer]

I had relevent posts on this topic... whered they go?

[R. Andrew Lamonica]

Opps.  I deleted them by mistake.  My bad.

[R. Andrew Lamonica]

The prices and processor information I have is from two quotes from Dell.  They are not going to change these prices even if I present them with an argument that the 640 should be more expensive.  I suspect that their prices are based on a combination of what they think people will pay and what they have agreed to in our contract.  We don’t pay the same as regular customers because we do a lot of business with Dell and we have different warranty requirements (5 years on workstations with self-maintainer status).

For the curious, my quotes have the 3.2GHz HT computers costing $948 less per 10 stations than the 2.8GHz DC computers.  

The problem with holding out for the 840 is that it still splits the cache between the two cores.  In an AI program or other processor intensive application the amount of cache available to the application can make a big difference.  You could, of course, solve this problem by writing your program to split its work between two processes, but must CS students do not take the time to do this.

[R. Andrew Lamonica]

Yikes.  

I just ran a quote for the 840 processor model.  It would add about $270 per/computer or $2700 to the total bill.  It think we will stick with the two options posted above.

[Peter Motyka]

Wouldn't the usefulness of a dual core processor be contingent on the running of Symmetrical Multi-Processing (SMP) aware applications?  I realize that the Windows NT kernel is SMP aware and distributes load across CPUs, however, I'm not certain an app like Visual Studio does the same.

A quick Google search suggests that special application frameworks are required to truly harness the SMP nature of HT/dual-core systems...

http://msdn.microsoft.com/vstudio/tour/vs2005_guided_tour/VS2005pro/Framework/CPlusOpenMP.htm

So, are these chips marketing hype, or are they really *faster*?  I'd love to hear what Dr. Stephen would have to say about this :)

P.S. Andrew, you never said anything about starting a debate about the perceived performance aspects of HT/Dual-core ;)

[R. Andrew Lamonica]

A debate about the perceived performance aspects of HT vs. Dual-core is exactly what I want.  From a purely technical standpoint I can see the difference between the two and I think that for typical CS projects that require High performance the HT processors are probably better (especially for the money we save).  However, the actual benefit of HT processors some-of-the-time could be rendered meaningless if DC processors have a perceived benefit all-of-the-time.  

I would much rather work on a computer that seems faster 99% percent of the time and is defiantly slower 1% of the time than one that is defiantly faster 1% of the time and seems slower 99% of the time. ;-)

What I don’t know is if the DC processors seem any faster than the HT ones during normal use.  I do know that the Pentium D processors have other new technology that aids security (execute-disable-bit) and power consumption, but those are a side issue.

[R. Andrew Lamonica]

Another thing to consider is that if very few students have access to Dual-Core computers at home (or work), then that is a vote for getting some in the lab.  This way we can provide something that is not otherwise available and thus make the labs more valuable.

[Peter Motyka]

Perception is a big issue here.  If I were to lay out the coin for a DC processor, you bet I’d be desperate to find some performance benefit as to not feel swindled into buying something simply due to marketing hype.  I genuinely believe that those that jump on bleeding edge technologies are convinced they see a performance boost because they’d have to submit to buyer’s remorse otherwise.

In my experience, the true performance of a workstation is usually constrained by the I/O subsystem.  I’d rather fork over extra bucks for a 10,000RPM hard drive than a faster/more advanced CPU.

[William Grim]

lamonica wrote:

As you can see, the Dual-core technology is more expensive and splits the cache between the â€Ã...“coresâ€Ã,. Potentially, making single-threaded programs (like many of the ones CS students write) run slower.

Playing devil's advocate, I'd say that this won't make single-threaded apps any slower.  1 MB of cache sounds pretty large, and I'm sure Intel tested to ensure that for the expected use, the cache will give us a high hit percentage.

So, let's think of it like this: the HT processor can be assumed to use both areas of the HT processor all the time.  If it takes one thread 1 MB of cache, on average, to achieve a high hit ratio in the cache per thread, then 2 MB of cache makes sense for a single core HT chip.  On the other hand, a dual core, single-thread CPU would need 1 MB of cache per core to achieve the same thing.

Now, I don't have the average CPIs for the dual core or HT handy, but the dual core is probably a simpler core than the HT core.  Therefore, the improved CPI may offset the slower clock rate enough to give the dual core the same or more execution speed than the single core HT CPU.

Again, I'd have to see real numbers and the cache hit ratio to determine if there is any real benefit, but I'm leaning towards yes.  Though, the difference in price may not make the performance/cost ratio high enough to warrant the purchase of the dual core CPU.

[alep85]

All I'm gonna say is I love my Core Duo laptop, and it blazes my friend's P4 HT Dell when running Windows.  HT is slower than dual-core, to me.  Buy based on what the programs you are going to use support, and how much that extra speed matters to you and your wallet :)

[John]

I am in the market for a new computer as well and have REALLY wanted an AMD processor. (I'm not tring to start a debate just explaining myself.) However after a brief sampling of what the Core Duo brought to the table, I was so impressed that I have decided to put off my AMD dream and purchase a Core Duo laptop. The sampling of the Core Duo that I had wasn't very extensive at all and completely got me hooked. Even though I haven't used a HT computer, I would still recommend the DC.

[R. Andrew Lamonica]

This is kind of an old, we have actually already purchased the new computers.  We did go with the DC ones over the HT ones.  There was a bit of an ordering mix-up so they probably will not be installed for a few more weeks.  But anyone with access to the special projects lab (EB2029) will be able to see them then.  

[R. Andrew Lamonica]

As if by magic, as soon as I announce my prediction to when the computers will arrive, I am proven wrong.  The new Dual Core computers have arrived.  I hope to have five of them in EB2029, by the end of the week.