[Mikemakeet]

So I have an issue with my 4170 i can't seem to understand. I have a stable 4.9 ghz clock and it runs fairly well but i run intel burntest and the gflops are horrible. I run all my games with my 6970 on high or ultra with normally yielding around 60 fps on battlefield 3, dirt 3, and skyrim, but my gflops are between 21.7 and 22.3. since its a quad core @4.9 ghz i should be getting 4*4*4.9= 78.4 gflops. I've tried adjusting my memory frequency and timings with no luck, as well as my hypertransit frequency. Any suggestions?
And if you need any other tech info lemme know:

Asus Sabretooth 990fx mobo
Mushkin Enhanced Blackline memory-2x4gb @ 1600 and 1866 mhz (rated for up to 2000, but mobo supports up to 1866) 1.65V
AMD FX4170 @ 4.9ghz 1.4375V
Corsair H60 water cooler
Asus 6970 radeon gpu

 

[Natesters93]

Have you raised your CPU-NB? And i don't think your 4.9ghz is stable with 1.4375Volts.... Mines not even stable under 1.48Volts @ 4.7ghz, But all chips are different, so just my 2 sence. I get 20.8-21Gflops at 4.6Ghz too put my input in on that too

 

[Cannon19932006]

Quote:

Originally Posted by Mikemakeet

So I have an issue with my 4170 i can't seem to understand. I have a stable 4.9 ghz clock and it runs fairly well but i run intel burntest and the gflops are horrible. I run all my games with my 6970 on high or ultra with normally yielding around 60 fps on battlefield 3, dirt 3, and skyrim, but my gflops are between 21.7 and 22.3. since its a quad core @4.9 ghz i should be getting 4*4*4.9= 78.4 gflops. I've tried adjusting my memory frequency and timings with no luck, as well as my hypertransit frequency. Any suggestions?
And if you need any other tech info lemme know:
Asus Sabretooth 990fx mobo
Mushkin Enhanced Blackline memory-2x4gb @ 1600 and 1866 mhz (rated for up to 2000, but mobo supports up to 1866) 1.65V
AMD FX4170 @ 4.9ghz 1.4375V
Corsair H60 water cooler
Asus 6970 radeon gpu

Fx performs terribly in ibt because it only has 1 fpu per module, therefore in your 4170 you have 2, the same amount as a dual core processor.

 

[trumpet-205]

Quote:

Originally Posted by Cannon19932006

Fx performs terribly in ibt because it only has 1 fpu per module, therefore in your 4170 you have 2, the same amount as a dual core processor.

That (FX is bad at floating point operation) and don't worry yourselves over GFlops.

GFlops is a really useless measurement of CPU performance. GFlops varies depending on different workloads too.

 

[damric]

Those FLOPS numbers are right on target for FX-4100 at that frequency, at least according to IBT/LinX

This has to due with the way IBT and LinX calculate FLOPS. It's just not to date for the Bulldozer architecture.

To get an accurate reading, download Sandra 2012. It's coded to accurately test FLOPS.

 

[trumpet-205]

Quote:

Originally Posted by damric

Those FLOPS numbers are right on target for FX-4100 at that frequency, at least according to IBT/LinX
This has to due with the way IBT and LinX calculate FLOPS. It's just not to date for the Bulldozer architecture.
To get an accurate reading, download Sandra 2012. It's coded to accurately test FLOPS.

Not really. IBT does make use of AVX.

GFlops is a really useless measurement of CPU performance. GFlops varies depending on different workloads too.

 

[damric]

Quote:

Originally Posted by trumpet-205

Not really. IBT does make use of AVX.

Really. Go try your Bulldozer on Sandra.

 

[dhenzjhen]

Quote:

Originally Posted by trumpet-205

GFlops is a really useless measurement of CPU performance.

I'm just curious why it's useless? If it's useless then can I ask you, what's the best way to calculate Gflops in a CPU/GPU cluster?

 

[trumpet-205]

Quote:

Originally Posted by dhenzjhen

I'm just curious why it's useless? If it's useless then can I ask you, what's the best way to calculate Gflops in a cluster?

Like I said it varies depending on workload. Take GFlops on GPU for example, their measurements are done using simple calculation repeatedly, not an accurate reflection of modern workload. You will also notice that Radeon tends to have higher GFlops than GeForce, but that does not mean Radeon has better GPGPU performance. Many times you find that GeForce has better GPGPU performance because CUDA offers better flexibility in coding.

 

[Fooxz]

That is also not how to calculate an expected GFLoPs... Did no one notice that?

 

[iinversion]

Quote:

Originally Posted by Cannon19932006

Fx performs terribly in ibt because it only has 1 fpu per module, therefore in your 4170 you have 2, the same amount as a dual core processor.

This.

/endthread

 

[Blameless]

Quote:

Originally Posted by Fooxz

That is also not how to calculate an expected GFLoPs... Did no one notice that?

I noticed it, but I just got to this thread.

The OP is assuming each core can reliably retire four FP instructions per clock cycle, which is ludicrous.
Quote:

Originally Posted by Cannon19932006

Fx performs terribly in ibt because it only has 1 fpu per module, therefore in your 4170 you have 2, the same amount as a dual core processor.

Not the whole story by any means, but BD does have fewer FP resources per core than prior AMD CPUs did.

 

[Mikemakeet]

Thanks for all the replies guys i'm relieved that its a typical problem. I'm gonna try sandra 2012 and see what the results yield. Also @Fooxs thats just what i read how to calculate it how is it actually supposed to be calculated?

 

[damric]

Here:



Just over 40GFLOPS, like I said.

 

[Domino]

lol...my 555 @ 4.1 did 27ish...

 

[tinytim1971]

hi just wondering i have a fx4170 what would you recomend me doing i have mine at 4.5 so what voltage and anything else i should i set it to thanks

 

[tinytim1971]

hi everyone i have a fx4170 im sitting on 4.5ghz at the moment what other setting in the bios do i have to set things to like voltage and what other settings do i have to ajust to go with those settings