MSP7 and Hyperthreading Benchmarks

[Hulk]

Brian,

Hello! Over the last few years that I've been testing MS Pro I've never noticed a performance difference between operating systems or amount of memory, provided that MSP isn't memory starved. Everyone who runs the test always has 512MB or more. I will add this information to the test page.

Take care,

Mark

[Brian Ellis]

Mark

In general, I would agree with you, but the fact that Intel says that one cannot get full HT performance except with XP does make me suspicious. Also, I've a niggling suspicion that, with HT, there may be a higher memory loading, so that those with, say, 1 Gb may see an improvement over 512 Mb with HT on. I cannot substantiate this notion, at the moment, but...

[Dr Mordrid]

Videomaker magazine had a review of a Dell Precision 650 video workstation loaded with dual Xeon 3.06 HT's in this months issue. Their analysis after extensive testing in numerous configurations was that HT was of little advantage in rendering video;

We are not confident enough in our margin of error to conclude that HT hurts rendering performance, but we are prepared to declare that it doesn't help.

Still, the dual-virtual-quad CPU machine was eminently usable even while we were wailing away on the CPUs, RAM and hard disk. At no point did the system crash or grind to a crawl and our DVDs came out just fine. This performance is analogous to having a car with a 60MPH maximum engine. If we add another 60MPH engine (or three more), the car will still have a maximum speed of 60MPH, but it will still have a lot more power than the single engine car.

Which indicates to me that most of the advantages are in multitasking.

Dr. Mordrid

[Hulk]

Doc,

I read that review, carefully. They didn't even get a significant boost from dual processing, much less HT, as can been seen by the slight drop in performance when one cpu was disabled. I don't think their test stresses much beyond MPEG encoding, i.e. rendering overlays, titles, color correction, and other effects. We have seen larger dually gains in Premiere with YOUR test.

I don't think Premiere is optimized as well as MSP7 for dual processors (for MPEG), or HT at all. We do have solid evidence of how well MSP7 does with dualies (Kirill) and HT (Sciascia and Brian). P4 duallies provide a huge boost, nearing 100% with HT, and HT alone looks to provide 12 - 25% depending on the motherboard. More data will clear this up.

Brian,

The only issue with Win2000 and HT is that 2000 "sees" HT as another physical processor as far as the number allowed for the particular version of Win2000 is concerned.

Win2000 Pro - Will allow 2 non-HT cpus or 1 HT cpu
Win2000 Server - Will allow 4 nonHT or 2 HT
WinXP Home - Will allow 1 cpu with or without HT
WinXP Pro - Will allow 2 cpus with or without HT
WinXP Server- Will allow 4 cpus with or without HT

See what I mean? It's only a licensing issue. The coding of the OS to take advantage of multithreading is the same for both. I gleamed this information from the Anandtech P4 3.06 review if you wish further information.

As far as the memory issue. I have yet to see a difference in scores due to memory from 256MB to 1GB, or whatever. This assumes no other apps are running and no previously closed apps have hijacked memory. If you run the test from a clean boot state scores will not differ due to memory. Yes, in the real world more memory will keep things moving along faster, but it won't affect what we are measuring if the test is run as indicated above. HT is new to the test, and if I receive solid contrary evidence, I will change my view and start reporting memory as well!

I will keep your thoughts in mind.

- Mark

[Dr Mordrid]

Premiere 6.5 isn't optimized to even use SSE, much less SSE2 or HT, for anything except for when it uses DirectShow filters. Got that from Adobe during the 6.5 beta.

Dr. Mordrid

[wkulecz]

Bang/buck?

This would be my most important figure of merit.

The dual Xeon 2.8GHz seems the clear performance winner despite being "only" 533 fsb. Dell Precision 450 with dual 2.8 Xenon is ~$2800.

the second best is about 3/4 the performance, how much under 0.75 X 2800 = ~$2100 is it?

--wally.

[wkulecz]

A quick trip to pricewatch has:

2.8GHz Xeon/533 @ $468
Tyan 2668AN dual MB @ $266
2GB DDR-266 as 4X 512 @ $200

So for ~1500 I could upgrade recycling everything else from my current system
or about $1200 dropping to 2.66GHz Xeon/533 processors.


3GHz P4/800 @ $407
about $200 for 875 MB
2GB PC3500 DDR400 as 4X 512 $264

About $875 for the P4 update.

To answer my own question, about 2X the cost to gain 1/3rd more bang with the Xeon update.

--wally.

[Hulk]

Wally,

You can pick up an Abit IS7 motherboard, which has PAT activated for $105 shipped (retail) from Newegg or some other similar on-line retailer.

It's a shame that Intel puts such a price penality on dual processing by not allowing P4 to do dually.

- Mark

[wkulecz]

Mark,

Abit is really pretty low on my list of MB makers, I know overclockers love 'em but the ones I've experience lacked stability. Also, Abit got hit really bad by the "leaky" capacitor problems a while back.

I'd rather pay more to get something a bit better, YMMV.

AMD is no better wrt dualies, priced Athelon MPs vs same speed Athelon?

How much performance would I give up dropping from 2.8 to 2.66GHz Xeons?

I don't really have time to tear things apart to upgrade right now anyways, but I periodically check if a really compelling bang/buck upgrade would force the issue.

--wally.

[Hulk]

Wally,

I know a lot of people had bad experiences a few years back with the good old BH6. I had a BH6/Cel 300A>450 o/c combination myself. I guess I was lucky, I used that board for 3 years without incident.

Anandtech.com just published the first of a series of motherboard comparisions today and rated the IS7 number 1 for speed and stability. If I remember correctly, the Asus 865 board was ranked number 2 and just as fast. Both of these boards have "workarounds" that enable PAT. I know many people who swear by Asus so you can also pick up this board for little over $100 and still get 875 performance (actually a bit better) at 865 price.

Here are some MSP7 performance extrapolations based on the performance of Kirill's dual Xeon 2.8 and Sciascia's 2.6C systems.

DV Test:
Dual 3.06 Xeon - 5:10
Dual 2.8 Xeon - 5:40 (actual test data)
Dual 2.66 Xeon - 5:56
Dual 2.4 Xeon - 6:36
Dual 2.2 Xeon - 7:12
Dual 2.0 Xeon - 7:55

Single 3.0C - 7:36
Single 2.8C - 8:09
Single 2.6C - 8:46 (actual test data)
Single 2.4C - 9:30

Assuming Kirill's and Sciascia's scores are repeatable, and we have no reason to assume they are not, even a dual 2.4 Xeon would require a single 3450MHz P4 to match it. A dual Xeon 2.66 system would require a single 3850MHz P4 for equal performance. Just for kicks, it would take a 4.4 P4 to equal the performance of a dual 3.06 Xeon system. It would take a dual Xeon 6.0 system to perform this test in real time!

Although pricey, those dual scores are really quite impressive. Especially when you consider the fact that the performance delta is even GREATER for the MPEG test. That is, the dual Xeon systems even to better in the MPEG test.

If I were to buy and off the shelf Dell system I would opt for an 875/HT enabled P4 system. With coupons and rebates, you can work a deal on one of these systems for $600 or so. If I were to build, I go dual Xeon since it's hard to find anyone on-line that discounts these types of workstations.

- Mark

[wkulecz]

Since MPEG encoding is my main bottleneck, I'll be giving the dual Xeons a very hard look when its time to update!

Thanks for the estimates.

--wally.