Phenom 1090T slows down my HD6870

[Cristian]

Hey!

I recently upgrade (¿?) my Phenom II X2 550 BE to a Phenom II X6 1090T BE.
I also have a Gigabyte GA-MA785GM-US2H, 6GB DDR2, and the HD 6870.

The thing is that in 3DMark 11 the graphic test goes wors in the 1090T than in the 550:thumbsdown:. Even when the 1090T has higher freq., more cores and the Turbo Core enabled :\

Look:
http://3dmark.com/compare/3dm11/192978/3dm11/196508

It is the exaclty the same system except for the processors.

Any ideas?
Maybe a botleneck in the memory?

Thanks!

Pd: The temps are fine.

 

[Smoove910]

Maybe an indication the test is not CPU specific.

 

[Cristian]

Maybe an indication the test is not CPU specific.

I just have tested the GTA IV and goes wors with the Phenom II X6

 

[eternalone]

Might have gotten an evil cpu I will gladly take it off your hands and punish the heathen.

 

[Lorne]

I see one problem and a possible another that has affected me the same way.
First I have found there is not enough bandwidth from DDR2 even at 1066 for 4 cores let alone 6, YOu can see this in a WinRAR benchmarking, I went from a Phen X2 550 OC to 3.4 and would get 100% both cores as long as it was running, Upgraded it to a Athlon 640 X4 and can only get 70-80% on all core usage.
You might be able to get a little more performance if you can push the NB to memory up if you are not already past 2-2.6Ghz that most DDR2 MB's already are.
I have a POS Dual P4 at work that totally chokes from this and its DDR1800 FB memory with HT turned on.

Second, I have found that Win7 64 has developed a latency after changing CPU's only (I have a thred elsewhere on this same subject) on 2 AMD machines that I use.
GUI and graphics responds has become horrible and Ive already replaced drivers for chipset, gfx and even removed the CPU and had it reinstall in the device manager and no fix.

 

[Kenmitch]

I had something similar happen to me when I switched a cpu with a Win7 install. But it was a no hyper threaded chip to a hyper threaded chip. Windows didn't see the hype

Maybe try the show hidden devices trick and look under the cpu and delete the ghosts of cpu past and see what happens....Worked for me

 

[Rifter]

I had something similar happen to me when I switched a cpu with a Win7 install. But it was a no hyper threaded chip to a hyper threaded chip. Windows didn't see the hype

Maybe try the show hidden devices trick and look under the cpu and delete the ghosts of cpu past and see what happens....Worked for me

I have had this happen as well. I would re-install windows just to rule out it being a software issue.

 

[BLaber]

I just have tested the GTA IV and goes wors with the Phenom II X6

Try disabling cool & quiet & CE1 state in bios and check.

Also increase NB speed up to at least 2.6Ghz and raise NB voltage accordingly to make sure system is stable at increased NB speeds.

Try one step at a time and check.

 

[Cristian]

Try disabling cool & quiet & CE1 state in bios and check.

Also increase NB speed up to at least 2.6Ghz and raise NB voltage accordingly to make sure system is stable at increased NB speeds.

Try one step at a time and check.

I tried and doesn't work:\

Under GTA IV, the GPU load is on 40-45% and the CPU load is on the same level (I have a BIG bottleneck somewere...). I also tried to disable four cores to see what happens and the 550 runs better.
The only thing that improves the performance (100 points in 3dmark 11) was to set the ECC Firmware in NORMAL.

Under GTA IV, the GPU load is on 40-45% and the CPU load is on the same level (I have a BIG bottleneck somewere...)

I will try a clean install of Win 7, but if it doesn't work I don't have many more options...

regards,

cristian

 

[Martimus]

I seem to remember the same thing happening when I updated my processor from a 3500+ to a X2 4200+, and it was because of the OS. A fresh format and install of XP corrected the problem.

 

[SirGCal]

Just thinking out loud;

With three sticks of memory, you can't run dual-channel effectively. That could be part of your bottleneck. DDR2 is slow enough as it is... Again, just thinking out loud, but try removing one stick to get it into dual-channel mode... (Remove the proper stick for one stick in each channel)

 

[cusideabelincoln]

You should try reinstalling just the video card drivers before doing a clean reinstall. And then update to or reinstall the latest motherboard chipset drivers.

Just looking at your graphics scores shows the 1090T slightly slower, but if you look at the Physics scores your 1090T is clearly working faster there. I doubt it's working at its full potential with just DDR2 RAM, as already mentioned.

 

[richierich1212]

Well your 1090T is clocked nearly the same as your 550 was. Put good use to that unlocked multiplier and raise it up over 4GHz. The 550 + 1090T is based off of the same platform except for two more cores. Since they both have the same amount of L3 cache, the 550 has 3MB allocated to each core while the 1090T has 1MB. Plus the Thubans have a really strong IMC, so try pushing for 2.8GHz NB if you can run the processor @ 4GHz.

 

[Cristian]

Just thinking out loud;

With three sticks of memory, you can't run dual-channel effectively. That could be part of your bottleneck. DDR2 is slow enough as it is... Again, just thinking out loud, but try removing one stick to get it into dual-channel mode... (Remove the proper stick for one stick in each channel)

Yes, I also tried that and it doesn't work either.

You should try reinstalling just the video card drivers before doing a clean reinstall. And then update to or reinstall the latest motherboard chipset drivers.

Just looking at your graphics scores shows the 1090T slightly slower, but if you look at the Physics scores your 1090T is clearly working faster there. I doubt it's working at its full potential with just DDR2 RAM, as already mentioned.

Yes indeed! the Physics goes faster there, but the main question is why the fps in the graphics test in slower in the 1090T than in the 550. Moreover, in GTA IV there is no additional gain in perfomance with the extra cores of the 1090T, even when this game is very CPU limited:\.

I will try to reintall the drivers, but I'm not quite sure if any will change.

Well your 1090T is clocked nearly the same as your 550 was. Put good use to that unlocked multiplier and raise it up over 4GHz. The 550 + 1090T is based off of the same platform except for two more cores. Since they both have the same amount of L3 cache, the 550 has 3MB allocated to each core while the 1090T has 1MB. Plus the Thubans have a really strong IMC, so try pushing for 2.8GHz NB if you can run the processor @ 4GHz

The NB is already in 2.6GHz with out a problem, but there is no gain in performance. I will not overclock the Turban with the stock cooler. Regarding the second concern, I believe you are right. The L3 cache may be the main cause in the drops of the fps in the 1090T.

Nevertheless, after trying in reinstalling the drivers, I will make a clean win instalation.

regards,

cristian.


Pd: Just one curious thing that I was able to observed. If I use another sofware to count the fps of the 3Dmark 11 bench (IZ3D in mono mode), the fps of the two test that assess the cpu (Physics and Combined) differ in both programs. In other words, the fps count of the IZ3D is higher (4-7 fps) than the one that the 3Dmark 11 tests shows. This only occurs in the test that involve the CPU. In the Graphic test 1, 2, 3 and 4 the fps count of both (3Dmark and IZ3D) are the same. Moreover, looking at how the test is performed in my PC, I believe that the fps correspond to the ones showed by the IZ3D drivers, since the scene seem to be more smooth than the fps showed by the futuremark software. This behavior makes one to believed that the CPU tests in the 3Dmark are wrong. Could any try the same on an Intel processor?:twisted:

 

[soccerballtux]

The NB is already in 2.6GHz with out a problem, but there is no gain in performance. I will not overclock the Turban with the stock cooler. Regarding the second concern, I believe you are right. The L3 cache may be the main cause in the drops of the fps in the 1090T.

You know, I noticed a huge performance increase in changing the BUS speed.
This impacts the CPU-NB (L3 cache) and CPU frequency.

at 2.1ghz and overclocked via the bus, my CPU feels as fast as it did when I left it sitting at 3.5ghz. Firefox is much smoother. This likely means at 3.5ghz, it's MUCH faster.

You can overclock this and lower your multiplier and maintain the same thermals. If you so desire. I would recommend it.

 

[richierich1212]

The stock cooler is decent, I'm sure you can hit 3.5GHz easily with stock voltage.

 

[oldswab]

Pd: Just one curious thing that I was able to observed. If I use another sofware to count the fps of the 3Dmark 11 bench (IZ3D in mono mode), the fps of the two test that assess the cpu (Physics and Combined) differ in both programs. In other words, the fps count of the IZ3D is higher (4-7 fps) than the one that the 3Dmark 11 tests shows. This only occurs in the test that involve the CPU. In the Graphic test 1, 2, 3 and 4 the fps count of both (3Dmark and IZ3D) are the same. Moreover, looking at how the test is performed in my PC, I believe that the fps correspond to the ones showed by the IZ3D drivers, since the scene seem to be more smooth than the fps showed by the futuremark software. This behavior makes one to believed that the CPU tests in the 3Dmark are wrong. Could any try the same on an Intel processor?:twisted:

subscribed. i have noticed a tremendous boost going from an e6600/ddr2/x38 system to a 1090t/ddr3/890fx system, but i have nothing else to compare it to. (sorry this wasn't helpful, but i'd like to find out what you find out!)

 

[nenforcer]

I believe this has something to do with going from 2 X 512Kb L2 and 6MB L3 spread over 2 processors = 1 X 512Kb L2 Cache and 3MB L3 cache per core.

The Phenom 2 X6 has the same amount of L2 cache per core at 512Kb but now distributes the same 6MB L3 cache over 6 cores so you only have 1MB L3 cache per core.

This is exactly why I don't like what AMD has done with there 6 core processors, since for any software which is not highly multithreaded i.e. 3DMark 2011 etc. you will actually see worse performance.

They really need to increase the amount of L3 cache but this was probably not possible with 45nm and will have to wait until 32nm Bulldozer to see better performance. I assume you ran the Phenom ][ X2 950 @ what 3.1GHz? and the Phenom ][ X6 1090T at 3.2GHz is nearly identical in clock speed.

Either that or wait for software to finally catch up to the hardware!

I'm still running a dual core and don't see any reason to upgrade.

 

[Arkaign]

I believe this has something to do with going from 2 X 512Kb L2 and 6MB L3 spread over 2 processors = 1 X 512Kb L2 Cache and 3MB L3 cache per core.

The Phenom 2 X6 has the same amount of L2 cache per core at 512Kb but now distributes the same 6MB L3 cache over 6 cores so you only have 1MB L3 cache per core.

This is exactly why I don't like what AMD has done with there 6 core processors, since for any software which is not highly multithreaded i.e. 3DMark 2011 etc. you will actually see worse performance.

They really need to increase the amount of L3 cache but this was probably not possible with 45nm and will have to wait until 32nm Bulldozer to see better performance. I assume you ran the Phenom ][ X2 950 @ what 3.1GHz? and the Phenom ][ X6 1090T at 3.2GHz is nearly identical in clock speed.

Either that or wait for software to finally catch up to the hardware!

I'm still running a dual core and don't see any reason to upgrade.

Eh, it doesn't work this way as far as L3 cache access.

PhII L3 cache works in 'exclusive' design, using 48-way set associativity. Any of the cores can access the contents dynamically, and no cache data is duplicated whatsoever.

Performance will not go down at all with more cores as long as they have an equal amount of L2 cache (dedicated per core). The shared L3 is dynamic, so on an equal load/app, you have no difference in capability, just more cores to do processing duties.

If a PhII X6 with the same L2 per core and same L3 cache size is performing worse at equal clock speed to an X2/X3/X4, then something is very wrong. In the worst single-threaded apps, you can expect identical performance. In practice, there is usually a boost anyway due to the OS process scheduler handing out background processing to lesser-used cores (background windows stuff, AV, etc).

OP's OS/Driver config is broken most likely, unless it's a bios issue (update to latest and do a full reset to check), or broken hardware.

 

[Lorne]

I suggest reinstall Win7 again and hope it fixes the problem, Im hoping it does because this would prove the point that changing CPU in Win7 causes a fluke problem just like the Win7 2D taskmanager fluke.

 

[Cristian]

Thanks for all the suggestions!!!

I just reinstalled Win 7, turned off the C1E state, Virtualization and Turbo Core (this really affect in a negative way the 3dmark 11 score :S), and set the vcore manually at 1.28 (in order to reduce the temps), and the things improves almost 300 points in 3Dmark 11.

new windows installation:
http://3dmark.com/3dm11/258227;jses...m11/258227?key=qy6PXBu6p7ULejcRurUJ3JTE7wtJty

However, only two of the previous tweaks REALLY improves the performance.

The reinstallation of win 7 doesn't affect the scores:

Old windows installation:
http://3dmark.com/3dm11/258405;jses...m11/258405?key=aQuHxejzY2UDsacvUAC3ZJbEexp977

From what I saw, the C1E and Turbo Core features were the things that slow down the performace. The results just increased 200 points turning off the Turbo Core and 100 points turning off the C1E state. Nevertheless, the HD 6870 still goes faster with the phenom X2 550 (see the first post). :thumbsdown:

However, in overall performance, I'm just 30 points bellow the lowest similar system (1090T + HD 6870) in 3Dmark 11. This difference may be related to the fact those systems have DDR3 whereas I have DDR2 (800Mhz), but I don't know the state of the C1E and Turbo Core in those systems... So, none conclussion can be stated from that....

regards,

cristian

 

[BLaber]

Thanks for all the suggestions!!!

I just reinstalled Win 7, turned off the C1E state, Virtualization and Turbo Core (this really affect in a negative way the 3dmark 11 score :S), and set the vcore manually at 1.28 (in order to reduce the temps), and the things improves almost 300 points in 3Dmark 11.

new windows installation:
http://3dmark.com/3dm11/258227;jses...m11/258227?key=qy6PXBu6p7ULejcRurUJ3JTE7wtJty

However, only two of the previous tweaks REALLY improves the performance.

The reinstallation of win 7 doesn't affect the scores:

Old windows installation:
http://3dmark.com/3dm11/258405;jses...m11/258405?key=aQuHxejzY2UDsacvUAC3ZJbEexp977

From what I saw, the C1E and Turbo Core features were the things that slow down the performace. The results just increased 200 points turning off the Turbo Core and 100 points turning off the C1E state. Nevertheless, the HD 6870 still goes faster with the phenom X2 550 (see the first post). :thumbsdown:

However, in overall performance, I'm just 30 points bellow the lowest similar system (1090T + HD 6870) in 3Dmark 11. This difference may be related to the fact those systems have DDR3 whereas I have DDR2 (800Mhz), but I don't know the state of the C1E and Turbo Core in those systems... So, none conclussion can be stated from that....

regards,

cristian

Also disable cool and quiet in bios.

 

[TekDemon]

I was also considering the cache as a possibility. Even though in theory the extra cores shouldn't be eating up the L3 there might be background tasks that kinda cause them to eat up cache. And of course the rather ancient bus honestly seems to struggle to deal with the traffic from 4 cores, going to 6 might cause a bit of a traffic jam.
Still, I'm kinda surprised to see this since generally I haven't seen benchmarks showing a drop going to Thuban.
Is the RAM running correctly in dual channel mode? The OP said he has 6GB of RAM, how many sticks is that?! If he upgraded from 2 to 3 sticks it might have screwed up his memory bandwidth.

 

[Martimus]

I was also considering the cache as a possibility. Even though in theory the extra cores shouldn't be eating up the L3 there might be background tasks that kinda cause them to eat up cache. And of course the rather ancient bus honestly seems to struggle to deal with the traffic from 4 cores, going to 6 might cause a bit of a traffic jam.
Still, I'm kinda surprised to see this since generally I haven't seen benchmarks showing a drop going to Thuban.
Is the RAM running correctly in dual channel mode? The OP said he has 6GB of RAM, how many sticks is that?! If he upgraded from 2 to 3 sticks it might have screwed up his memory bandwidth.

Very good point. Hopefully he has 2x 1GB, and 2x 2GB to allow dual channel. If not, he should remove the third stick until he can get a fourth so that he isn't completely starved for bandwidth on the new system. Especially with the extra cores.

 

[Arkaign]

Very good point. Hopefully he has 2x 1GB, and 2x 2GB to allow dual channel. If not, he should remove the third stick until he can get a fourth so that he isn't completely starved for bandwidth on the new system. Especially with the extra cores.

No review will ever show PhII X2/X3/X4/X6 as having any advantage to fewer cores at the same clock speed for any app, so long as :

Each core has the same L2 cache size
Total L3 cache is the same size
Clock/NB/etc frequencies are the same

If you ever run into the case where an X6 is performing worse than the same clock speed X4, even in a single-threaded app, then something is broken/borked, end of story.

6GB of ram is very strange, although it could be :

2x2GB + 2x1GB, which should stay in dual-channel mode. 3x2gb would tend to break it back down to single-channel mode.

You won't find a single review showing 1090T slower than PhII 955BE (both 3.2Ghz stock) in anything. Even apps that aren't multi-core optimized work a little better due to better multitasking of background stuff.