 |
 |
 |
|
 |
You are not logged in.
It may be true that all of these benchmarking tools are "misreading" the CPU frequency, but the question I have is why the EEE scores almost double in 3Dmark2001SE with the only bios revision that makes manually setting the frontside bus's speed to 100mhz possible. If what Noeee says is true, then the only thing changing the FSB's speed would do would make benchmarking programs correctly report the CPU's clockspeed. If we're already receiving the extent of the processing power in 0204 and 0401, then where did the 90% performance jump come from in 8804 when benchmarking via 3Dmark2001?
Or am I missing something entirely...
Offline
I think there's more to the 8804 BIOS than just the FSB. Even at the same speeds, there's a 500-point jump going from 0401 to 8804.
0401, 630MHz = ~1500
8804, 630MHz = ~2000
8804, 900MHz = ~3000
Windows XP also launches a lot faster under 8804, even though the raw SSD benchmarks are nearly identical.
Offline
Trousers wrote:
It may be true that all of these benchmarking tools are "misreading" the CPU frequency, but the question I have is why the EEE scores almost double in 3Dmark2001SE with the only bios revision that makes manually setting the frontside bus's speed to 100mhz possible. If what Noeee says is true, then the only thing changing the FSB's speed would do would make benchmarking programs correctly report the CPU's clockspeed. If we're already receiving the extent of the processing power in 0204 and 0401, then where did the 90% performance jump come from in 8804 when benchmarking via 3Dmark2001?
Or am I missing something entirely...
Exactly...very well said. Clearly the FSB is at 70MHz. I would say this is the case in both Xandros and WinXp.
Offline
TwisteR wrote:
Trousers wrote:
It may be true that all of these benchmarking tools are "misreading" the CPU frequency, but the question I have is why the EEE scores almost double in 3Dmark2001SE with the only bios revision that makes manually setting the frontside bus's speed to 100mhz possible. If what Noeee says is true, then the only thing changing the FSB's speed would do would make benchmarking programs correctly report the CPU's clockspeed. If we're already receiving the extent of the processing power in 0204 and 0401, then where did the 90% performance jump come from in 8804 when benchmarking via 3Dmark2001?
Or am I missing something entirely...Exactly...very well said. Clearly the FSB is at 70MHz. I would say this is the case in both Xandros and WinXp.
Isn't 3DMark designed to primarily measure 3D performance? So maybe the BIOS is tweaking the graphics component?
Offline
rs-px wrote:
Isn't 3DMark designed to primarily measure 3D performance? So maybe the BIOS is tweaking the graphics component?
3D performance depends on more than just the GPU - CPU, RAM, FSB, and almost every other part of the system has an impact on the score.
The GeekBench2 score increased as well, though not nearly as much:
http://browse.geekbench.ca/geekbench2/similar/23621/6
My score is the 689, vs the ~560 results that seem to be coming from "stock" clocked Eees - a 23% increase.
Offline
Chris wrote:
rs-px wrote:
Isn't 3DMark designed to primarily measure 3D performance? So maybe the BIOS is tweaking the graphics component?
3D performance depends on more than just the GPU - CPU, RAM, FSB, and almost every other part of the system has an impact on the score.
The GeekBench2 score increased as well, though not nearly as much:
http://browse.geekbench.ca/geekbench2/similar/23621/6
My score is the 689, vs the ~560 results that seem to be coming from "stock" clocked Eees - a 23% increase.
Yes, I know that 3DMark measures more than 3D performance 
But the evidence you're presenting here (quote: "The GeekBench2 score increased as well, though not nearly as much") really do indicate that the BIOS update tweaked the graphics somehow.
I'm no expert on this but I do know that the Linux kernel is very robust, and very rarely incorrect. I also know that Windows is flaky and using apps like CpuZ is pushing Windows in a non-standard way.
I'm inclined to believe that the hardware is running at 900MHz.
Offline
rs-px wrote:
Yes, I know that 3DMark measures more than 3D performance
But the evidence you're presenting here (quote: "The GeekBench2 score increased as well, though not nearly as much") really do indicate that the BIOS update tweaked the graphics somehow.
I'm betting the GMA900 got its core clock raised - that would explain the discrepancy in the 0401-to-8804 results at 70MHz. I'd have to compare some numbers in detail to find out. Shame I don't have two Eees to do side-by-side comparos ... 
I'll post a thread later in the XP forum asking for some help.
I'm no expert on this but I do know that the Linux kernel is very robust, and very rarely incorrect. I also know that Windows is flaky and using apps like CpuZ is pushing Windows in a non-standard way.
I'm inclined to believe that the hardware is running at 900MHz.
Run GeekBench at 900MHz and post your result - if they're truly running at the same speed, the scores should be within a reasonable margin of error.
Offline
Chris wrote:
Run GeekBench at 900MHz and post your result - if they're truly running at the same speed, the scores should be within a reasonable margin of error.
I get 555 running GeekBench both with the p4-clockmod module in the kernel and without (actually I get 556 without it in there). Note the caveats at the bottom of this post.
With the p4-clockmod kernel inserted, the speed is reported in /proc/cpuinfo as 900MHz. Without it, the speed is reported at 630MHz.
So I've got to conclude that the p4-clockmod is causing the kernel to incorrectly report the CPU speed. I read elsewhere that the primary purpose of p4-clockmod is to "insert 'waits' between clock cycles, so workloads take longer to complete, but generate less heat whilst doing so" (see here).
Caveats: I ran GeekBench from the command prompt in advanced mode of the default Xandros OS on the Eee. On the webpage for GeekBench it says it's designed for Fedora Core but it seemed to work OK, although it's primarily computational in nature and didn't do any 2D/3D tests. I also ran it with Firefox open in the background and on a system that's been running for some time, so the results are rough at best.
Offline
Thanks for making an appearance here, Chris. I guess I forgot to mention that all of my information comes from your posts.
It's possible that 8804 tweaked the integrated graphics, although I doubt a 3dmark performance gain of so much is attributable to only that (especially since using the latest intel GMA900 drivers only boosted my performance by about 5-10% over the old asus-provided display drivers).
If 8804 truly increases the frontside bus's speed from 70mhz to 100mhz, it wouldn't just mean that the CPU's core clock would scale with it, but that the integrated graphics would be able to communicate with the CPU that much faster. Unfortunately for me, I'm not able to get 8804 stable in order to actually test it for myself.
Release more updates, Asus! 
Offline
Trousers wrote:
Thanks for making an appearance here, Chris. I guess I forgot to mention that all of my information comes from your posts.
Not a problem. I've got some more information here too:
3DMark01 | 630-0401 | 630-8804 | 900-8804 -------- | -------- | -------- | -------- | score | 1589 | 2071 | 3019 | (overall score) -------- | -------- | -------- | -------- | game1 | 15.1 | 24.7 | 37.0 | (game 1) -------- | -------- | -------- | -------- | game1h | 7.9 | 9.0 | 14.2 | (game 1, high quality) -------- | -------- | -------- | -------- | game2 | 23.0 | 26.0 | 37.3 | (game 2) -------- | -------- | -------- | -------- | game2h | 9.5 | 13.6 | 18.4 | (game 2, high quality) -------- | -------- | -------- | -------- | game3 | 27.9 | 42.6 | 61.2 | (game 3) -------- | -------- | -------- | -------- | game3h | 12.3 | 16.4 | 24.7 | (game 3, high quality) -------- | -------- | -------- | -------- | game4 | 16.7 | 18.0 | 26.0 | (game 4) -------- | -------- | -------- | -------- | fr-st | 404.9 | 404.3 | 571.7 | (fillrate - single-textured) -------- | -------- | -------- | -------- | fr-mt | 430.1 | 430.1 | 607.6 | (fillrate - multi-textured) -------- | -------- | -------- | -------- | hpc-1 | 2.3 | 2.2 | 3.4 | (high polygon count - 1 light) -------- | -------- | -------- | -------- | hpc-8 | 1.2 | 1.2 | 1.8 | (high polygon count - 8 lights) -------- | -------- | -------- | -------- | ebm | 57.5 | 58.2 | 83.3 | (environment bump mapping) -------- | -------- | -------- | -------- | dot3 | 55.9 | 55.4 | 76.8 | (dot3 bump mapping) -------- | -------- | -------- | -------- | vs | 19.1 | 19.2 | 27.6 | (vertex shader) -------- | -------- | -------- | -------- | ps | 57.7 | 57.6 | 80.7 | (pixel shader) -------- | -------- | -------- | -------- | aps | 17.9 | 20 | 28.8 | (advanced pixel shader) -------- | -------- | -------- | -------- | pspr | 2.4 | 2.6 | 3.8 | (point sprites) -------- | -------- | -------- | -------- |
Clock-for-clock, the 8804 BIOS is faster in the game tests. Theoretical test performance is almost identical, but the game performance gets a good bump, especially in the low-detail tests.
Cranking the Eee up to 100FSB (900MHz) is a whole different ballgame. Since the GMA900 offloads processing onto the CPU and shares system RAM, giving it a 40% boost in FSB speed brings in massive gains from the bottleneck vanishing.
Now I'm off to run Doom 3.
Offline
Trousers, I think parts of my earlier post were a bit unclear, or am I just misinterpreting other's interpretation of my post?
My points regarding ODCM are unrelated to the FSB's impact on the actual operating frequency. When I stated that "the processor frequency never changes" it was in regards to ODCM throttle states not being capable of altering the CPU frequency. I was not stating that the processor runs at 900 MHz irregardless of FSB frequency (that statement would not be in line with the 'published' CPU theorem that I worked so hard on 
)
Back to ODCM:
When somewhat idle various Linux utilities on the eeePC would report that the processor is running at 112.50 MHz (12.5% throttle), such a reported frequency is incorrect because of the following reasons:
1- The utility incorrectly assumes the FSB is 100 MHz. Since the FSB is running at 70 MHz the CPU is actually running at 630 MHz therefore a 12.5% ODCM throttle should be reported as 78.75 MHz CPU frequency.
BUT more importantly the 78.75 MHz CPU frequency is still incorrect because...
2- ODCM throttling does not alter the CPU's operating frequency.
All in all, due to p4_clockmod and cpufreq modules the CPU frequency reported in linux/Xandros is always FUBAR.
Okay...I'm tired of theorem, where are the damn black eeePCs 
Offline
NoeeePC4MeWellMaybe... wrote:
Okay...I'm tired of theorem, where are the damn black eeePCs
Fo' realz. 
Offline
Trousers wrote:
NoeeePC4MeWellMaybe... wrote:
Okay...I'm tired of theorem, where are the damn black eeePCs
Fo' realz.
Going to be sitting in my lap in a few hours.
Offline
So, is there any software out there we can run which will let us adjust the FSB while the system is running (linux or windows) so that we can up the FSB without changing to 8804 bios?
Offline
This seems like a obvious BIOS issue. It's not taking advantage over the whole 900 MHz. I'm guessing a BIOS update could fix this
Offline
