Intel 4770k-4970k Gflops help!

[ScribbleMike]

I've been wondering what the average amount of Gflops is on an i7 4770k, I get around 100 gflops stock and I see it bump up to 110 when I go to 4.2Ghz. I wanted to see what other people are getting

So the rules are simple, you have to have a 4770k or 4970k and use Intel Burn Test to record gigaflops. I would prefer you guys take a screenshot so we can see what it says but you don't have to.

I just wanted to know average amount because sites on the net don't report their average gflops at all.

So post away! I would love to know the average!

 

[Traciatim]

Nobody reports on it because nobody cares. Floating point math is probably best done on a videocard anyway, especially if it's highly parallel. This is why things like OpenCL and CUDA exist.

 

[mapesdhs]

It's not reported because as Traciatim suggests, it's not a meaningful benchmark anymore.
Apart from GPU acceleration being very relevant these days, the fp performance of a CPU
for any real-world task depends on a whole host of factors, including data set size, cache
access issues, whether lots of array lookups are needed (needs strong int to keep up),
RAM speed/latency, etc. The c-ray test, for example, fits inside any CPU's L1 cache (tiny
data set), whereas I've seen some AE raytracing renders gobble 40GB RAM, and that's
small compared to scicomp workloads or film production environments (500GB I/O for one
frame is often typical).

Tests like Linpack are still used for rating supercomputers (top500.org), but people use IBT,
P95, etc. for stress testing, not performance testing. The only reason one wouldn't use a
GPU is if ECC is essential (in which case there's always Tesla) or if the nature of the dataset
means a GPU's main RAM is nothing like large enough (eg. ANSYS, GIS, defense imaging, etc.)

Ian.

 

[ScribbleMike]

Nobody reports on it because nobody cares. Floating point math is probably best done on a videocard anyway, especially if it's highly parallel. This is why things like OpenCL and CUDA exist.

But isn't there a problem if your cpu does not reach average gflops.
like a 4770k getting 20 gflops lower than the average for the 4770k

 

[Traciatim]

Nobody reports on it because nobody cares. Floating point math is probably best done on a videocard anyway, especially if it's highly parallel. This is why things like OpenCL and CUDA exist.

But isn't there a problem if your cpu does not reach average gflops.
like a 4770k getting 20 gflops lower than the average for the 4770k

Kinda, you wouldn't measure in GFLops in most cases, just use any benchmarking software like Passmark's CPU test, it even has built in tools to allow you to compare to similar or other CPUs so you can tell quickly if you are near the average performance.

 

[mapesdhs]

But isn't there a problem if your cpu does not reach average gflops.
like a 4770k getting 20 gflops lower than the average for the 4770k

No, because there are too many other factors affecting performance as I explained.

If you want to test performance then use tests designed to measure performance properly,
eg. render benchmarks, etc.

Ian.

 

[ScribbleMike]

But isn't there a problem if your cpu does not reach average gflops.
like a 4770k getting 20 gflops lower than the average for the 4770k

No, because there are too many other factors affecting performance as I explained.

If you want to test performance then use tests designed to measure performance properly,
eg. render benchmarks, etc.

Ian.

I thought you could use iIntel burn test to see your gflops

 

[mapesdhs]

I've already explained why this is meaningless. If you don't do the further reading
& research implied then I can't help further.

Ian.