1 |
Ok, I made some progress on this: |
2 |
|
3 |
I was able to get the same performance of X on the 945G with 1.x intel |
4 |
driver (620GX) and the 965Q with 2.0 intel driver (745) with the |
5 |
following setup of the MTRR registers: |
6 |
|
7 |
echo "base=0x0 size=0x80000000 type=write-back" >| /proc/mtrr |
8 |
echo "base=0x80000000 size=0x10000000 type=write-combining" >| /proc/ |
9 |
mtrr |
10 |
|
11 |
(Originally I didn't have the second line) |
12 |
|
13 |
These post were very helpful (don't know why I didn't find them |
14 |
earlier): |
15 |
|
16 |
http://forums.gentoo.org/viewtopic-t-563647-highlight-.html |
17 |
http://forums.gentoo.org/viewtopic-t-501708-highlight-optiplex.html |
18 |
|
19 |
Now the only thing that remains (I guess) is to find a proper setup of |
20 |
the MTRR registers on the 755. While playing with that I was a bit |
21 |
unluck... I was only able to make the box freeze so far. |
22 |
|
23 |
|
24 |
On 14.01.2008, at 10:01, Dirk Gassen wrote: |
25 |
|
26 |
> Thanks for your quick reply. Unfortunately, the email |
27 |
> got stuck in a spam filter that's why I reply so late. |
28 |
> |
29 |
> --- Donnie Berkholz <dberkholz@g.o> wrote: |
30 |
> |
31 |
>> On 18:45 Fri 11 Jan , Dirk Gassen wrote: |
32 |
>>> I currently have a performance problem on a |
33 |
>>> Optiplex 745/755 and the |
34 |
>>> onboard graphics card. We would like to run tvtime |
35 |
>>> as a TV application. |
36 |
>>> Unfortunately, when tvtime starts X consumes >30% |
37 |
>>> CPU, which results in a |
38 |
>>> somewhat sluggish system (responses to user |
39 |
>>> actions are slow, closed |
40 |
>>> captioning looses characters, etc.). |
41 |
>>> |
42 |
>>> This system did already work well on an Optiplex |
43 |
>>> 620GX (X CPU usage <15%). |
44 |
>>> |
45 |
>>> Here's some more info about the systems: |
46 |
>> |
47 |
>> To simplify things, I'm just going to point out what |
48 |
>> I think are the |
49 |
>> important facts. |
50 |
>> |
51 |
>>> Optiplex 620GX (works): |
52 |
> [SNIP] |
53 |
>> |
54 |
>> 945G with the 1.x intel driver |
55 |
>> |
56 |
>>> Optiplex 745 (performance issues): |
57 |
> [SNIP] |
58 |
>> |
59 |
>> 965Q with 2.0 intel driver |
60 |
>> |
61 |
>>> Optiplex 755 (also performance issues): |
62 |
> [SNIP] |
63 |
>> |
64 |
>> Q35 with 2.0 intel driver |
65 |
>> |
66 |
>> You've tested the new hardware with the old driver, |
67 |
>> and it didn't work |
68 |
>> for unrelated reasons |
69 |
> |
70 |
> I'm not sure if the reasons were unrelated but with |
71 |
> the old driver I get two bars with noise, one at the |
72 |
> right side of the screen and one at the bottom (looked |
73 |
> somewhat like uninitialized memory?). I wasn't able to |
74 |
> get rid of by changing tvtime configuration options, |
75 |
> so I thought the best way would be to upgrade the |
76 |
> driver to 2.0, which fixed *that* problem. |
77 |
> |
78 |
>> Could you test the old |
79 |
>> hardware with the new |
80 |
>> driver, to test whether it's the driver version |
81 |
>> change creating a |
82 |
>> slowdown? |
83 |
> |
84 |
> Nope, doesn't seem to be related to the driver change. |
85 |
>> 30% X CPU usage with 1.7.4. |
86 |
> |
87 |
>> Also, could you try 2.2 instead of 2.0? If |
88 |
>> that doesn't work, |
89 |
>> try 2.1. |
90 |
> |
91 |
> Same thing (I was already looking pretty often if |
92 |
> there's a new driver version and jumped on each new |
93 |
> version) |
94 |
> |
95 |
>> It's also possible that the driver's just not as |
96 |
>> good on 965 and Q35 as |
97 |
>> on 945. |
98 |
> |
99 |
> Would that be a hardware restriction or would that be |
100 |
> fixable in the driver (and what would I need to |
101 |
> do/contribute to fix it?) |
102 |
> |
103 |
>>> I suspect there's something wrong with the MTRR |
104 |
>>> setup on the 745/755 (wild |
105 |
>>> guess): |
106 |
>>> localhost ~ # cat /proc/mtrr |
107 |
>>> reg00: base=0x00000000 ( 0MB), size=65536MB: |
108 |
>>> write-back, count=1 |
109 |
>>> reg01: base=0x3d600000 ( 982MB), size= 2MB: |
110 |
>>> uncachable, count=1 |
111 |
>>> reg02: base=0x3d800000 ( 984MB), size= 8MB: |
112 |
>>> uncachable, count=1 |
113 |
>>> reg03: base=0x3e000000 ( 992MB), size= 32MB: |
114 |
>>> uncachable, count=1 |
115 |
>>> reg04: base=0x3d500000 ( 981MB), size= 1MB: |
116 |
>>> uncachable, count=1 |
117 |
>>> reg05: base=0x40000000 (1024MB), size=1024MB: |
118 |
>>> uncachable, count=1 |
119 |
>>> reg06: base=0x80000000 (2048MB), size=2048MB: |
120 |
>>> uncachable, count=1 |
121 |
>>> |
122 |
>>> ... and X reports in its log: |
123 |
>>> (WW) intel(0): Failed to set up write-combining |
124 |
>>> range |
125 |
>>> (0xd0000000,0x10000000) |
126 |
>> |
127 |
>> Yeah, this could cause problems with speed. If it |
128 |
>> worked in the old |
129 |
>> versions with otherwise identical software |
130 |
>> configurations (e.g. kernel |
131 |
>> features), this could be a driver bug. |
132 |
> |
133 |
> How would I go about setting it up manually? Or does |
134 |
> that need to be fixed in the driver? |
135 |
> |
136 |
> Do you think I should contact the driver developers or |
137 |
> file a bug report for the driver? |
138 |
> |
139 |
> BTW, to rule out any other interference from other |
140 |
> pieces of software (e.g., OpenBox) I did start only X |
141 |
> and tvtime, still the same problem. |
142 |
> |
143 |
>> Thanks, |
144 |
>> Donnie |
145 |
>> -- |
146 |
>> gentoo-desktop@l.g.o mailing list |
147 |
>> |
148 |
>> |
149 |
> |
150 |
> |
151 |
> |
152 |
> |
153 |
> ____________________________________________________________________________________ |
154 |
> Looking for last minute shopping deals? |
155 |
> Find them fast with Yahoo! Search. http://tools.search.yahoo.com/newsearch/category.php?category=shopping |
156 |
> -- |
157 |
> gentoo-desktop@l.g.o mailing list |
158 |
> |
159 |
> |
160 |
> |
161 |
|
162 |
-- |
163 |
gentoo-desktop@l.g.o mailing list |