SSE2 not enabled for P4-M?

[pumpkin0]

Hi,

I've downloaded the new 1.2.0 client and installed it on my desktop and laptop PC's. The client itself works fine and I have no crash issues or anything, but:

My desktop is a Pentium 4 2.4GHZ with hyperthreading and 800mhz FSB.
My laptop is a 2.4GHZ Pentium 4 mobile, non-hyperthreading with 400mhz FSB.

When I run the client on my laptop, I get about 235183 cEMs/sec - and on my desktop I get almost 2.5x that speed!

Is the client detecting the P4-mobile chip as SSE2 capable? I looked at the log and saw '[Sat Jan 03 11:34:01 2004] Mobile Intel(R) Pentium(R) 4 - M CPU 2.40GHz detected. Enabling cpu specific optimizations.' - but I'm not sure sure if it's actually using SSE2. Or is that the desktop really does run way faster because of the bus speed? (HT should not be an issue because I only run one client anyway.)

Regards from New Zealand,
pumpkin0

[pumpkin0]

Oh, and the processor on the laptop is not being clock-throttled or anything, I checked it out with WCPUID's realtime clock checker and it runs at the full 2.4GHZ.

[phooey]

You may want to make sure another task isn't interfering with your laptop results. Just as a reference, I have a 2.0Ghz Pentium4 - M and I'm seeing a rate of 380,300 cEMS/sec with the 1.20 client. That's up about 33% from the 1.1.1 client, so it appears 1.20 is utilizing SSE2.

On a related note, does anybody know if there are processor benchmarks for SoB anywhere, kinda like what D-Net has done with their speed pages?

[kugano]

On hyperthreading -- the client has no support for it, so it won't make a difference in the client's performance except possibly better responsiveness when other processes are sharing the CPU, but I'm not even sure about that.

On SSE2 support -- it's buggy. It's very possible the client is not properly auto-detecting CPUs on all systems and enabling/disabling features when it shouldn't be. I believe Louie plans to look at this as soon as he gets back from vacation. I think we're stuck until then, because honestly I don't myself understand some of the low-level client issues... that's historically been Louie's domain, while I've stuck mostly with server and website code. We can ask him when he gets back.

--
David Norris
danorris@seventeenorbust.com

[pumpkin0]

Originally posted by phooey
You may want to make sure another task isn't interfering with your laptop results. Just as a reference, I have a 2.0Ghz Pentium4 - M and I'm seeing a rate of 380,300 cEMS/sec with the 1.20 client. That's up about 33% from the 1.1.1 client, so it appears 1.20 is utilizing SSE2.

Hmm, that's what I thought as well, so I checked task manager to see if another process was leaching CPU time - which turned out to not be the case. SB appears to be getting full use of the CPU as far as I can tell.

Thanks for that explanation Kugano. It definitely would be cool to have SSE2 working on this laptop, because I have seen the huge difference that it makes on my desktop

[Frodo42]

I think that a lot of the speedup on your laptop is because of the difference in the FSB as P4's are pretty dependent on the speed of the FSB to my understanding anyways. On the other hand I have a P4 2.4 GHz running on a 133 FSB and get almost 400 Kcems/s ...

Another thing you might want to look into is if the laptop is somehow slowed because it is a laptop. The P4 is made to automaticly underclock itself if it gets overheated, when you run your CPU 100% I could imagine that it starts underclocking itself since it is limited how much cooling you are able to do on a laptop.
You could try putting the laptop into a freezer, but I wan't guarentee for the consequences

[pumpkin0]

Ok I think you might have a point there Frodo42 - I've just done some benchmarking with some surprising results.

One instance with 1.2.0 client:

My Pentium 4 Desktop:
2.4GHz, 800mhz FSB, hyperthreading: ~600k cems/sec

My Pentium 4 Laptop:
2.4GHz, 400mhz FSB, no hyperthreading: ~224k cems/sec

phooey's Pentium 4 Laptop:
2.0GHz, no hyperthreading: ~380k cems/sec

One instance with 1.1.1 client - no SSE2:
My Pentium 4 Laptop: ~9000 cems/sec (not k cems/sec)

So it appears the new client does use SSE2 on my laptop, but that it's significantly underperforming. phooey's 2.0GHz P4 laptop absolutely thrashes mine.

I've double checked, and the SB client is getting full use of the CPU. Nothing else is drawing CPU time.

I wonder if it's the ATi motherboard chipset that my laptop has. My desktop uses an Asus P4P800 motherboard - i865 chipset with that Asus memory acceleration thingy that basically turns it into an i875. It's also dual channel.

In contrast, the laptop has an ATi integrated video chipset, and I've heard that it's not very performance orientated.

Still kinda funny how I get less than half the performance of the desktop on the laptop. Oh well, it's still reasonably quick