![]() |
|
|
#12 |
|
Nov 2010
Germany
3·199 Posts |
Interesting.
LaurV, for you I'll add 'RV7xx' to the list of known GPUs . This way I can also teach mfakto how to calculate the number of stream cores this card has (320 instead of the reported 640). However, you seem to have some special low-power version as HD4650 ist officially listed with two versions of 600 and 650 MHz respectively. If mfakto got the clock speed right, then yours is running at just 450MHz.Also, I suggest you use VectorSize=4. This will not make the screen lag much more, but may bring you into the 20GHz-days/day range that James lists for the card. Anyway, that GPU was not a high-end one 4 years ago, and it still isn't. Yet I assume it's faster than the HD3000. To make that one work, it appears you need to install the Intel SDK and remove all other GPUs from the system plus get the special mfakto version, see the mfakto thread. |
|
|
|
|
|
#13 | ||
|
Romulan Interpreter
"name field"
Jun 2011
Thailand
41×251 Posts |
Quote:
Quote:
Related to my HD4650, we also wondered how come that the performance of the card seems to be higher (almost double) on James' site, and, in spite of the fact that GPU-Z shows that the card is maxed (98%) with a single instance, we also tried to launch 2 instances, but it did not help at all, of course. We got few more GHz by increasing the max number of primes in the sieve to 1M, it went to 1M after few rows, and to almost 17GHz, but still far from what James site says. Now after you told us about the clock difference, THIS would explain (beside of us messing with settings of the program) the bad performance! The value is "in the range" of the fraction 450/600. Now we may know the reason too, but we can not fix it until we will be back home tonight: the card has some jumpers to select the clocks, with which, most probably, we played in the past... (the settings are not written o the PCB), so we have to find the booklet, which is another raid into the pile of trash in our wardrobe...
Last fiddled with by LaurV on 2013-02-01 at 02:00 |
||
|
|
|
|
|
#14 |
|
"Mr. Meeseeks"
Jan 2012
California, USA
1000100001112 Posts |
I assume 640 is 640 "compute cores"? The 7770 I have has 640 too!
Last fiddled with by kracker on 2013-02-01 at 02:04 |
|
|
|
|
|
#15 |
|
Romulan Interpreter
"name field"
Jun 2011
Thailand
283316 Posts |
As the man said above, because the card is too old, and unknown to the program, this number is detected wrong. New version of mfakto will be "taught" to "read" this card correctly in the future. There should be 320.
|
|
|
|
|
|
#16 | |
|
"Mr. Meeseeks"
Jan 2012
California, USA
37·59 Posts |
Quote:
![]() Sorry.
|
|
|
|
|
|
|
#17 |
|
May 2011
810 Posts |
I run a box with that exact model card. From the factory, my Sapphire (POS) HD4650 is clocked 600/700. I OC the snot out of it though (900/875) and I pull 32-36 GHz out of it, depending on the bit range of the TF I run.
So yeah, if you don't want that card, I've got a free PCIe slot. |
|
|
|
|
|
#18 | |
|
May 2011
23 Posts |
Quote:
In Windows, I enable Overdrive in CCC, and adjust the clocks in GUI. I would be surprised if you couldn't do that as well. *Linux: aticonfig --odenable aticonfig --odgc #reports clocks aticonfig --odsc=600,700 #sets 600 core, 700 mem aticonfig --pplib-cmd "set fanspeed 0 100" #set's 100% gpu fan, if you are going to OC... LOL |
|
|
|
|
|
|
#19 | |
|
Jun 2019
Boston, MA
3·13 Posts |
Quote:
|
|
|
|
|
![]() |
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| 8675309^2-1 just passed index of 3000 | pdazzl | Aliquot Sequences | 10 | 2015-02-14 17:42 |
| 3000 < k < 4000 | otutusaus | Riesel Prime Data Collecting (k*2^n-1) | 2 | 2012-05-09 19:27 |
| 3000 and beyond! | guido72 | Teams | 5 | 2009-11-30 16:02 |
| Optimising work for Intel Core 2 Duo or Quad Core | S485122 | Software | 0 | 2007-05-13 09:15 |
| how do graphics cards work so fast? | ixfd64 | Hardware | 1 | 2004-06-02 03:01 |