Do we loose Windows optimization with mobiles in desktops?

[The Coolest]

It really depends on the BIOS, OS and how the specific program itself is written. Windows just uses whatever BIOS has to say about the naming of the processor, the reason for that is simple: Say tomorrow we see some new processor that windows doesn't "know" so you'll get unrecognized CPU in windows, but if it pulls the name from the BIOS (well actually its pulled out from the CPU itself, but the BIOS is what puts it there.) then the CPU will still be recognized as the BIOS is supposed to recognize the chip.
What I personally think, from reading the K8 BIOS developer's guide, is that the BIOS reads the CPUID of a chip and the capabilities it has, and most likely turns them on. the only reason that the chip isn't recognized is because its a mobile chip, which have a single bit set to '1' instead of '0' and when the BIOS tries to compare it to the CPU names it has pre-programmed, it can't find any, and displays the CPU as "Unrecognized". I had the same thing happen to me when I had my Tbred in my A7V133 board. At first it would read the chip as a Duron, because the bios couldn't recognize it. after I updated the BIOS the chip was recognized correctly, but I didn't see any performance boost, or anything like that.

The bottom line of what I am trying to say here is that I don't think we're losing any performance or feature because the CPU is displayed as unrecognized.
most programs will not look at the CPU name either, but the instruction sets that the chip supports. so the CPU name, is again, irrelevant.

 

[Gnufsh]

Does someone with a mobile that displays as unknown want to display their feature flags (I know WCPUID used to tell you which ones were enabled)? That way you can check to see if your BIOS is enabling SSE support.

tdbone1, both those support articles state that th driver is available from windowsupdate. Go to windowsupdate.microsoft.com scan for updates, and make sure it's selected (if it isn't already installed). Then, download and install.

 

[dands42]

OK, My cpu says Unknown in both the bios and Windows. If I remember correctly (I'm at work now) all but 1 flag was enabled in the CPUID program. It was an SSE flag. There were two I believe and one wasn't selected. I guess that would be SSE2 from what I read here.

 

[hitechjb1]

This is from a mobile barton 2400+.

WCPUID, CPU-Z, GCPUID, Windows XP System Information all detect it as
- model 10 (which stands for barton)
- L1 data cache size 64 KB
- L1 instruction cache size 64 KB
- L2 cache size 512 KB
- the cache size, cahce line size, associativity are the same as desktop barton
...
- supports for SSE, MMX, MMX+, 3DNow!, 3DNow!+, as in desktop barton
- some utilities (WCPUID, sisoft) explicitly said it is a "mobile" xp

The only unknown is the CPU name string

 

[hitechjb1]

Further, Sandra cache and memory benchmark was used to compare the mobile barton set to the reference CPU (available in Sandra)
- Tested mobile 2400+: 200 x 11 (which is same frequency as 3200+ CPU)
- Reference1: 3200+ barton (512KB L2), running at PC3200 (200 FSB), CAS 2.5
- Reference2: 2800+ barton (512KB L2), running at PC2700, CAS 2.5
- Control CPU: 2200+ tbred (256KB L2), running at PC2100, CAS 2.5

So the mobile barton set to a PR rating of 3200 measured
- very close to the 3200+ cache-memory performance
- better than a 2800+ cache-memory peroformance
- way above a 2200+ CPU with only 256KB L2 (for control test)

 

[hitechjb1]

When the mobile barton is set to 200 x 12 = 2400 MHz, it indeed delivers better cache-memory performance than a 3200+ (512KB L2) at rated 200 x 11 = 2200 MHz.

 

[hitechjb1]

Managed to run it at high voltage of 2.22 V to get to 2.79 GHz.
SLK-947U and 80mm Tornado (with fan speed control)
MB is NF7-S rev 2.0.
512 MB x 2 PC 3500.


At such speed, from the Sisoft Dhyrstone Integer and Whetstone Floating Point benchmark,
it scores 7%+ higher than a P4C 3.2 GHz in both integer and floating point (w/o SSE2).

For pics,
http://www.ocforums.com/vb/vb/showthread.php?s=&postid=2578290#post2578290

 

[mongoled]

Thanxs for the info

 

[Gnufsh]

Well, it looks like the only problem is with the name string (supplied by the BIOS), which it wouldn't make sense to base optimizations on. Far better to check the flags, especially in the case of things like SSE on Athlon Xps, which can be disabled. I doubt there is any performance hit/feature loss with this. If there is, however, can you cut the correct L5 and have it e recognised as a regular Barton?

 

[hitechjb1]

If looking at the post on cache-memory benchmarking I posted few post back carefully, the mobile barton set to run at 3200+ rating, its cache-memory performance is a little bit below that of a 3200+, under same reference condition of 200 x 11 CPU setting and cas 2.5. This may be attributed to

- the reference curve of 3200+ in sisoft is off a bit
- the test run I did at 3200+ setting may not be exactly the same as the reference setup

Note: The cache-memory benchmarking includes SSE feature of a CPU.

I will be doing another run (later today) of the cache-memory run again to cross check the result posted earlier, attempting to explain the small discrepancy of cache-memory of the 3200+ run between the sisoft reference and an actual mobile barton. Will include some other cross checking using CPU benchmarking, ... to calibrate potential setup discrepancies, ....

 

[The Coolest]

Yes, if you want to turn it back into XP you will have to break the 3rd L5, this will make it an MP tho. now if you brake both 3rd and 4th L5s it will become XP

*EDIT:* Sorry had a typo, typed L3 instead of L5 in the first mention of the bridges.

 

[dands42]

Yes, if you want to turn it back into XP you will have to break the 3rd L3, this will make it an MP tho. now if you brake both 3rd and 4th L5s it will become XP

So you break L3 #3, L5 #3 and L5 #4. All of these or just L5's?

This will still keep it as a Barton XP Right?

You are also saying that the Bios will read it as an XP if I read this correctly.

 

[The Coolest]

breaking L3 #3 will achieve nothing, if you mean L3 #5 then this will give you a default multiplier of x14.
If you have the DFI Infinity/Lanparty or ABIT NF7/AN7 or a Soltek nF2 boards you have nothing to worry about as you'll get access to all multipliers, if you have the ASUS or something else, and will want higher multiplier (x13 and higher) you will have to either do a wire trick or just break the 5th L3.
But when u break L5 #3 and #4 you get an unlocked desktop barton.

*EDIT:* Sorry had a type at the first mention of the bridges in the previous post, I mean the 3rd L5, 3rd L5 is the mobility bridge, and the 4th L5 is the MP bridge, break them both and get a "desktop" barton.

 

[hitechjb1]

Redid the the cache-memory benchmark of a mobile Barton set to the similar setup as a desktop 3200+ Barton
(200 x 11 = 2200 MHz, PC3200 CL2.5) in Sandra.

- The results look the same, being 1% lower than the reference 3200+.
- Also did a CPU benchmark for comparison w/ a 3200+, CPU benmark also 1% lower.

The 1% lower may be due to measurement discrepancy, some background task in the computer, ....

So, from the cache-memory benchmark and CPU arithmetic benchmark of a mobile Barton,
and the various CPU detection utilities such as XP System Information, WCPUID, GCPUID, CPU-Z,
I do not see any significiant deviation in the mobile Barton (set to 3200+ setting) from a traditional 3200+ Barton.



Reference: few posts back in this thread
The Sandra cache-memory benchmark uses SSE.

 

[c627627]

Thank you again for a thorough battery of tests.

Would you say that CPUs do not have to be detected correctly in BIOS to take advantage of what's described with "AMD and Microsoft worked together to optimize applications..."

The applications themselves then do not rely on BIOS for that and automatically take advantage of the system CPU's capabilities...

 

[hitechjb1]

Thank you again for a thorough battery of tests.

Would you say that CPUs do not have to be detected correctly in BIOS to take advantage of what's described with "AMD and Microsoft worked together to optimize applications..."

The applications themselves then do not rely on BIOS for that and automatically take advantage of the system CPU's capabilities...

I cannot answer that specific question, about AMD and Microsoft, ....

What I have shown is that:
(1) from cache-memory benchmark, CPU arithmetic benchmark,
I do not see any difference greater than 1%.
(2) from a number of CPU detection utilities - XP System Information, WCPUID, GCPUID, CPU-Z, Sandra, ...,
the key features of a Barton are shown to be in the mobile Barton by those utilities.

Originally posted by hitechjb1
WCPUID, CPU-Z, GCPUID, Windows XP System Information all detect it as
- model 10 (which stands for barton)
- L1 data cache size 64 KB
- L1 instruction cache size 64 KB
- L2 cache size 512 KB
- the cache size, cahce line size, associativity are the same as desktop barton
...
- supports for SSE, MMX, MMX+, 3DNow!, 3DNow!+, as in desktop barton
- some utilities (WCPUID, sisoft) explicitly said it is a "mobile" xp

The only unknown is the CPU name string

 

[tb2496]

But when u break L5 #3 and #4 you get an unlocked desktop barton.

I did that yesterday, only broke L5 #3 and it shows in BIOS and WinXP as Athlon XP now.

 

[repilce]

Managed to run it at high voltage of 2.22 V to get to 2.79 GHz.
SLK-947U and 80mm Tornado (with fan speed control)
MB is NF7-S rev 2.0.
512 MB x 2 PC 3500.

At such speed, from the sisoft Dhyrstone Integer and Whetstone Floating Point benchmark, it scores 7%+ higher than a P4C 3.2 GHz in both integer and floating point (w/o SSE2).

For pics,
http://www.ocforums.com/vb/vb/showthread.php?s=&postid=2578290#post2578290

heh.. ummm ... that's 7% faster than... TWO 3.2C's SMT'n... (multi procs).. if you notice in sandra there's also an option for the 3.2C (same thing) without the [2 SMT]..

 

[Fr3ak]

According to WinXP Pro, my oc'ed 1700+ also is a unknown CPU type...
But there aint a differance in Booting, etc when its detected as 3200+

 

[hitechjb1]

heh.. ummm ... that's 7% faster than... TWO 3.2C's SMT'n... (multi procs).. if you notice in sandra there's also an option for the 3.2C (same thing) without the [2 SMT]..

Thanks for pointing that out.

Actually, it is still one P4C but with symmetric multi-threading turned on.

The instruction per cycle (IPC) ratio for integer arithmetic for XP to P4C (w/ 2 SMT) is about 1.22, meaning 1 XP cycle is about 1.22 P4C cycles.


What is IPC and how to compare cycle or Hz for different CPU architectures (page 19)