AMD FX-8350 "Vishera" Linux Benchmarks

[curaga]

Ivy Bridge was shown not to measurably scale with RAM speeds on Windows. So not surprising you didn't see it on Linux either.

[Banu22elos]

It's not like Intel has 8 FPUs and 4 integer cores...

[Kano]

ddr3-2133 is not really cheap, the money you spend for that you could instead invest in a nv gfx card without onboard vga faster ram over ddr3-1333/1600 is hardly noticeable. OEM boards often run with ddr3-1333 setting all the time (you can not select the ram speed with those boards). Lately i got 4gb ddr3-2133, it was 1-2 fps faster with TF2/mesa 9.1 and intel hd 4000. I would not say that when it jumps from 51 fps to 53 fps it was a needed upgrade (usually i use 8gb ddr3-1600 with that board) - well I switched back to the slower but more ram after some benchmarks.

[Matt607]

You are wrong because its not a short period of time you can force this output all the time.

[juanrga]

For APU tests Michael did test the RAM scaling.

Yes, and he wrote:

It's just not with the graphics performance though where the AMD A10-5800K APU performance really desires fast memory, but for memory-intensive applications there is also a big impact when moving to DDR3-2133MHz speeds.

I would like to know what memory timings were used in the Vishera Benchmarks for the FX, the APUs, and the Intel chips. My impression is that the Intel chips got an extra advantage from the AMD chips running with underclocked RAM.

[curaga]

For APU tests Michael did test the RAM scaling.

[juanrga]

Memory timings matters very little for most benchmarks, nothing for practical use. The reason for this is the fact that no program writes to RAM and then immediately after reads it back in the next few CPU instructions. And if so, it would be in CPU cache, so it would not matter any way.

People keep complaining these kinds of tests are run without DDR3-1866, but that does normally not matter. In general the improvements will be less than 2%, sometimes none at all. You could actually run the memory at 1333 MHz and it would still not hurt too much. The exception is the APUs, which are a bit more sensitive to memory bandwidth.

Do you actually know what XMP is? XMP would not affect performance at all, it's just stored recommended settings embedded in the flash on the memory chip. The user still has to select it in the BIOS menu, and there is nothing preventing the user from running the same settings on an AMD board. "AMD Performance kit" is just marketing bullshit, any module following specs will do. And for your information, many SB/IB boards actually defaults to running DDR3-1333 even as CPU and Memory support more, so this should be an disadvantage for Intel!

I know that memory can affect AMD FX-8150 CPU. For instance, gaming under windows you would lost up to a 8% performance if you run memory at 1333 MHz instead of at stock speeds



And the performance lost would be larger in more memory intensive tasks.

Moreover, how many users will be overclocking the FX chip @ 4.8 but underclocking the ram? Memory @ 2133 seems a more adequate timing and then differences would be a bit larger.

As you say, the APUs are much more sensitive to memory bandwidth. I read reports where the gain in performance is of up to 20% by using faster memory under windows. I know AMD chips usually run faster under linux. That is why I asked about what memory is being used in phoronix tests. This is important info (at least for me) which is lacking.

So far as I know AMD mobos support XMP via emulation. Enthusiasts users say me that the best results are obtained with AMD optimized RAM. I do not know more about this issue, and this is why asked to test some AMP profile.

[efikkan]

Even my High End Workstation P9X79 WS defaults to 1333 MHz, I had to manually adjust speed and timings. And if my computer reboots for whatever reason (e.g. power outage), it says "overclock failure" and defaults back to 1333 MHz.

[Kano]

it depends on the board, when you use oem boards as you get in retail pcs/laptops ram is most likely running @ 1333 for Intel systems. If you buy oc boards then those support of course xmp profiles or at least manual overrides for timing and speed settings and voltage control. Basically you can prove many things with benchmarks, if you use a test that is highly ram sensitive or with gpus running with shared memory then you can see a diff. I am sure many would not even correctly identify if dual or single channel ram is used with onboard gpus - that usually gives a nice boost compared to 1 piece - but you need to run benchmarks.

[efikkan]

The review only says that 8 Gib of memory were used for all the chips, but timings are not reported.

Would I assume that the review run the i7-3770k at stock speed (1600) but the A10 and FX-8350 were run with underclocked ram (1600) instead of stock speed (1866)? If this is so, then one would add some score more for the AMD chips.

Memory timings matters very little for most benchmarks, nothing for practical use. The reason for this is the fact that no program writes to RAM and then immediately after reads it back in the next few CPU instructions. And if so, it would be in CPU cache, so it would not matter any way.

People keep complaining these kinds of tests are run without DDR3-1866, but that does normally not matter. In general the improvements will be less than 2%, sometimes none at all. You could actually run the memory at 1333 MHz and it would still not hurt too much. The exception is the APUs, which are a bit more sensitive to memory bandwidth.

Also memory brand and profiles are not reported. I assume that both Intel and AMD chips used some Intel optimized memory kit (XMP enabled) but not memory kits AMP enabled. It would be interesting to see how the AMD chips perform with an AMD performance kit memory kit.

Do you actually know what XMP is? XMP would not affect performance at all, it's just stored recommended settings embedded in the flash on the memory chip. The user still has to select it in the BIOS menu, and there is nothing preventing the user from running the same settings on an AMD board. "AMD Performance kit" is just marketing bullshit, any module following specs will do. And for your information, many SB/IB boards actually defaults to running DDR3-1333 even as CPU and Memory support more, so this should be an disadvantage for Intel!