cnzdrn - Sempron 150 @ 6133MHz - 3373 points Geekbench3 - Multi Core

[I.nfraR.ed]

5 hours ago, ObscureParadox said:

You can still get around 3500 with an L3014 if you maximise everything. This for sure isn't game over for the Xeon It's also almost impossible to get much faster than this with a Sempron on DDR3 because of the IMC limitations. All the same I'm really glad someone managed to get this fast with a Sempron, it for sure makes the game really intersting. I'm looking forward to the battle

450 HTT is 900 mem and 6.5GHz with Sempron 150 (I have a promising one), so there's still a headroom for memory frequency.

Based on my testing on air, if it scales ok, then ~170cb might be possible. Don't know about Geekbench, but mem should definitely help. Not sure if the IMC would be a bottleneck, though. Have to test. Looking forward to the new scores as well, I can only watch at the moment. Don't forget Sabertooth is one of the worst boards for AM3 CPUs, at least for Pi, Geekbench and CB might be different.

Some examples (not perfect, but you get the point). Both are excellent benchers:

DDR2 + 4969MHz https://hwbot.org/submission/3252830_noxinite_cinebench___r15_sempron_145_119_cb

DDR3 + 4662MHz https://hwbot.org/submission/2518880_strunkenbold_cinebench___r15_sempron_145_120_cb

Edited November 13, 2018 by I.nfraR.ed

[cnzdrn]

Sabertooth is slow in geekbench. Compared to gigabyte 890fx based. 

[yosarianilives]

I would think anything memory related would be heavily board dependant. There are a number of memory timings that are set by bios that you don't have access to set so certain boards will just perform better on platforms like AM3 where relatively few timings are exposed to the user. That's why for example the Crosshare IV Extreme has been shown to have better "IPC" than the Crosshair IV Formula in 32m and probably anything else that memory heavily impacts.

[cnzdrn]

15 minutes ago, yosarianilives said:

I would think anything memory related would be heavily board dependant. There are a number of memory timings that are set by bios that you don't have access to set so certain boards will just perform better on platforms like AM3 where relatively few timings are exposed to the user. That's why for example the Crosshare IV Extreme has been shown to have better "IPC" than the Crosshair IV Formula in 32m and probably anything else that memory heavily impacts.

And why you need to bin ram with 1:2 divider 

[ObscureParadox]

5 hours ago, I.nfraR.ed said:

450 HTT is 900 mem and 6.5GHz with Sempron 150 (I have a promising one), so there's still a headroom for memory frequency.

Based on my testing on air, if it scales ok, then ~170cb might be possible. Don't know about Geekbench, but mem should definitely help. Not sure if the IMC would be a bottleneck, though. Have to test. Looking forward to the new scores as well, I can only watch at the moment. Don't forget Sabertooth is one of the worst boards for AM3 CPUs, at least for Pi, Geekbench and CB might be different.

Some examples (not perfect, but you get the point). Both are excellent benchers:

DDR2 + 4969MHz https://hwbot.org/submission/3252830_noxinite_cinebench___r15_sempron_145_119_cb

DDR3 + 4662MHz https://hwbot.org/submission/2518880_strunkenbold_cinebench___r15_sempron_145_120_cb

No it's a very fair point. When I made my initial statement I forgot that this wasn't run at max multiplier, so there is a lot more wiggle room. Maybe I should get out my AM3 boards and get some testing done too. But for sure first things first is to maximize what I already have

[unityofsaints]

12 hours ago, yosarianilives said:

r15 definitely gains from mem on these chips, just a small gain. Got to remember that at 6.7 on ddr2 it's just a few points behind l3014 on ddr3, and at 6.1 on ddr3 the 150 is about 10 points away from l3014. Definitely worth binning 890fx boards to find one that can support 6.5+ on a sempron.

Even taking DDR3 into account Xeon will win. There's also some fuel left in the tank in terms of BCLK on the blue side

[ObscureParadox]

There's not that much left though Don't forget too that it's almost impossible to run good memory speeds with an L3014, the boards just don't like it, hence Luumi was only running single channel memory for his runs.

[bolc]

too bad I don t live around, i d do it for free, well after borrowing your cpu