I don't know much first-hand about running these dual CPU systems, but the gut feeling is that games didn't make use of multiple threads at that time or for some time after.

Even if you take the dual CPU factor out, its still rather odd that the 1Ghz Coppermine is nearly keeping uk with a Tualatin 1.4S , which is what is so odd about it

That margin could be within chipset vs. chipset comparisons and motherboard BIOS optimizations...

Well, the 3DMark results are similar to RTCW results - 20% more. While the theoretical difference would be 40%, but I guess these benchmarks depend on other factors, like memory bandwidth which is equal for 815 and 133A.

GL1zdA wrote:

Well, the 3DMark results are similar to RTCW results - 20% more. While the theoretical difference would be 40%, but I guess these benchmarks depend on other factors, like memory bandwidth which is equal for 815 and 133A.

I could I suppose try a more powerful GPU , like a Geforce3, but its still very very odd, if it is the chipset, then thats a pretty poor showing from VIA.

Darkman wrote:

Even if you take the dual CPU factor out, its still rather odd that the 1Ghz Coppermine is nearly keeping uk with a Tualatin 1.4S , which is what is so odd about it

The issue with these dual-socket motherboards is that if you try and run a single-threaded task on them (Some benchmarks like the early 3dmark ones, and some games that don't use more than 1 core/cpu), their single-threaded performance is slower than one of their cpu's in a stand-alone single socket motherboard.

That is how they're designed, essentially these older ones split the performance of both chips in half via the chipset, as most of the early dual-cpu systems didn't have the ability to run dedicated chipsets for each cpu and mesh them together like our modern ones do.

So if you're running a single-threaded task/benchmark/game on your dual-S370 system, expect the performance to be about half for one of the cpu's.

I looked for GA-6VTXD real fast in google and yes, that's a single-chipset system so yes each cpu is split in half.

This is why (At the time) there were these exotic server systems from hp/compaq that cost some $10,000 new because they employed multiple chipsets, dedicating one to each cpu socket and basically yours is the cut-down version sold to consumers for cheaper.

At that resolution, you shouldn't see much of a difference.

Try your benchmarks again at 512 x 384.

On the Dimension 4100 board, you will need to cross-flash the last intel bios for the motherboard as the Dell A11 bios does not support the Tualatin. Check out my still in progress thread here for more theory and details. Dimension 4100 Intel BIOS

If one CPU isn't accessing memory then it shouldn't tie up the chipset in any significant way. However, there is the issue of cache coherency which can cause some minor penalty if the process keeps bouncing around.
For benchmarks which are single threaded, you could try using the task manager to manually assign the benchmark process to a specific CPU. Supposedly the NT kernel should already do a pretty good job of managing this by itself though. It keeps track of "CPU affinity" for this reason.

Have you checked what version of the 815 chipset you have on that Dimension board? I don't remember the list but some of the S-spec codes include Tualatin support. I know on the Dell GX150 boards, some of them support Tualatins and some don't, and I'm not sure if Dell ever advertised it.

kithylin wrote:

The issue with these dual-socket motherboards is that if you try and run a single-threaded task on them (Some benchmarks like th […]
Show full quote

The issue with these dual-socket motherboards is that if you try and run a single-threaded task on them (Some benchmarks like the early 3dmark ones, and some games that don't use more than 1 core/cpu), their single-threaded performance is slower than one of their cpu's in a stand-alone single socket motherboard.

That is how they're designed, essentially these older ones split the performance of both chips in half via the chipset, as most of the early dual-cpu systems didn't have the ability to run dedicated chipsets for each cpu and mesh them together like our modern ones do.

So if you're running a single-threaded task/benchmark/game on your dual-S370 system, expect the performance to be about half for one of the cpu's.

I looked for GA-6VTXD real fast in google and yes, that's a single-chipset system so yes each cpu is split in half.

This is why (At the time) there were these exotic server systems from hp/compaq that cost some $10,000 new because they employed multiple chipsets, dedicating one to each cpu socket and basically yours is the cut-down version sold to consumers for cheaper.

I would not say it's fair to say it's "splitting each CPU in half" - but where you can run into bottlenecks is because the CPUs are sharing the FSB <-> each other and the Northbridge, which is still presenting a single memory controller and memory bank. If both CPUs were accessing heavily, it can become a significant bottleneck. shamino's point on caching is also a consideration.

Overall SMP machines don't do anything special for older games - the games just don't utilize multiple cores/threads, and while Windows should load balance, it's unlikely you're running enough background tasks (especially on a gaming-only retro machine) for that to be a significant factor. The biggest thing you may notice, subjectively, is less stutter. It isn't surprising at all that the two machines here are performing similarly enough either - the SMP machine is probably worse-off "per clock" than the Dell. If you can run the dual 370 board with a single CPU mounted, throw the 1GHz CuMine in there and see what it does - if it performs slower than the Dell you'll have your validation, if it performs faster (or identically) then there may be some other problem/factor at play. I agree with Phil on trying at a lower resolution, or trying a more powerful graphics card, to ensure you aren't GPU bound on the 5500.

For locking an application to a single processor, you'll have to use the affinity settings in Task Manager, as Windows 2000 does not support the /affinity hook thru the start command. This can be bothersome/problematic for games that do not run (or run properly) on SMP machines. Windows XP (and higher) will support the /affinity hook.

^^ that and the Via33a is not that optimized to run Dual Tualatin. It will work, but an Appollo 266 or ServerWorks ServerSet III HE-SL Chipset preform much better. The Via133a is first designed for a Single CPU.

http://www.os2museum.com/wp/the-tualatin-story/

Do what phil says: try a lower resolution. You can max out a voodoo5 with a 1ghz cpu no problem as long as you set the deatails of the game high enough.

Those games are pretty demanding for a voodoo5 and it could very well be a gpu limited situation. And the via is about 5% slower when compared to the intel 815 (at least in my tests, both are asus boards).

Or try both systems with a card like a geforce4 ti.

Also make sure the board can cache the 1,5gb ram.

Thats actually quite interesting in regards to the dual CPU bottleneck , it does make a bit of sense. I actually did try a lower resolution and the results were not that much better (the Tualatin was faster , but still relatively slower "per clock")

if I am going to go with a single CPU I may as well go with an Intel Chipset (like the i815E or such) since I generally trust them more for stability.

the optimal comparison: test the dual board with the coppermine 1ghz and see how it goes... preferably both cpu's in single setup (or both dual, but not single vs dual).

Since the tualatin is held back a little by the 133mhz fsb it does not scale perfectly, and on a dual board some things are shared, degrading performance a bit further.

Darkman wrote:

Thats actually quite interesting in regards to the dual CPU bottleneck , it does make a bit of sense. I actually did try a lower resolution and the results were not that much better (the Tualatin was faster , but still relatively slower "per clock")

if I am going to go with a single CPU I may as well go with an Intel Chipset (like the i815E or such) since I generally trust them more for stability.

Generally I like Intel chipsets (I actually can't think of one I dislike) - they did make SMP chipsets for P3, like the 840, which may offer better performance. I've also heard of DDR-compatible boards from QDI and perhaps others, but never seen one myself. 😊

Roman78 wrote:

^^ that and the Via33a is not that optimized to run Dual Tualatin. It will work, but an Appollo 266 or ServerWorks ServerSet III HE-SL Chipset preform much better. The Via133a is first designed for a Single CPU.

http://www.os2museum.com/wp/the-tualatin-story/

Blast! Beaten! I was going to mention this.

In This Thread: Stuffing a Server-Class CPU into a Desktop-Class motherboard and wondering why the system runs at Desktop-Class speeds.

Honestly, OP get one of the above chipsets so you get dual-channel memory. Even if you need to run Registered ECC, it should outperform your i815, significantly.

Here's a good article to help explain what's going on, here: http://www.anandtech.com/show/715/4

Logistics wrote:

Blast! Beaten! I was going to mention this. […]
Show full quote

Roman78 wrote:

^^ that and the Via33a is not that optimized to run Dual Tualatin. It will work, but an Appollo 266 or ServerWorks ServerSet III HE-SL Chipset preform much better. The Via133a is first designed for a Single CPU.

http://www.os2museum.com/wp/the-tualatin-story/

Blast! Beaten! I was going to mention this.

In This Thread: Stuffing a Server-Class CPU into a Desktop-Class motherboard and wondering why the system runs at Desktop-Class speeds.

Honestly, OP get one of the above chipsets so you get dual-channel memory. Even if you need to run Registered ECC, it should outperform your i815, significantly.

Here's a good article to help explain what's going on, here: http://www.anandtech.com/show/715/4

never said I expected server class speeds, obviously a server chipset would perform better, I was just rather perplexed how a Tualatin can nearly get beaten a Coppermine thats 2/3 the speed with half the cache.

couldnt have imagined the VIA would have crippled it that much.

Tualatin isn't really a server CPU - it is just a PIII with double L2 cache size. 😀
Also I don't think desktop board is holding the performace back. Check this - PIII-S 1400 + Asus TUSL2 = 53.3 fps ; PIII 1000EB + Asus TUSL2 = 40.9 fps ; 2x PIII Xeon 1GHz (cascades-256 core) + i840 board + dual-channel RDRAM = 40.7 fps.

http://hw-museum.cz/pom/ut_6_2015.png

Darkman wrote:

I suppose try a more powerful GPU , like a Geforce3, but its still very very odd, if it is the chipset, then thats a pretty poor showing from VIA.

Try lower resolution, - this will reduce factor of video card. And software mode.