Ran AIDA64 benchmark on a new build with a 5775c and ddr3-2400 C10 memory. The read/write/copy throughput are nearly identical for L4 and main memory. Was expecting an order of magnitude difference. Does this basically mean the 128mb edram was just a gimmick?
Broadwell L4/edram only as fast as main memory?
- Thread starter rimshaker
- Start date
Solution
you can overcome that write/copy deficiency of L4 by overclocking Uncore ratio. It affect directly write and copy rating of L4, but not affect read (which is already way higher than DDR3 2400Mhz) that much. and, even if you downclock your Uncore to 3.2, Read will still be way higher than DDR3 2400.
funny, 100Mhz of Uncore give exactly around 1000Mb/s on Write of L4 (on my pc)
. try it in Aida64 and see it yourself.
I also tested 5775C with DDR3 2400Mhz C10, with Cinebench C15. CPU was overclocked at 4.4, and Uncore was 4.0
Edram disabled, with XPM disabled, and Ram speed downclocked to 1066Mhz, it get average result of 890. so, it confirm that this test also depends on Ram.
Edram disabled, with XPM enabled, so Ram speed is...
funny, 100Mhz of Uncore give exactly around 1000Mb/s on Write of L4 (on my pc)
. try it in Aida64 and see it yourself.I also tested 5775C with DDR3 2400Mhz C10, with Cinebench C15. CPU was overclocked at 4.4, and Uncore was 4.0
Edram disabled, with XPM disabled, and Ram speed downclocked to 1066Mhz, it get average result of 890. so, it confirm that this test also depends on Ram.
Edram disabled, with XPM enabled, so Ram speed is...
Ecky
Illustrious
- Feb 3, 2013
- 7,592
- 2
- 38,960
The difference is that latency is far lower on Broadwell's L4 cache, which can improve performance dramatically. Bandwidth isn't the whole story.
Broadwell's L4 cache is only an eviction cache, which (mostly) means that anything that won't fit in L3 and would have ended up in system memory instead, hits L4 first, but Skylake's eDRAM is going to be different.
Broadwell's L4 cache is only an eviction cache, which (mostly) means that anything that won't fit in L3 and would have ended up in system memory instead, hits L4 first, but Skylake's eDRAM is going to be different.
L4 latency was 41.7ns, and memory was 53.4ns. Read throughput was 35% higher for L4, but actually slower on Write/Copy. Still doesn't make the L4 look all that impressive when coupled with fast system memory. Guess I should've went Skylake for the ddr4 headroom instead.
honey2015
Reputable
- Oct 8, 2015
- 87
- 0
- 4,640
you can overcome that write/copy deficiency of L4 by overclocking Uncore ratio. It affect directly write and copy rating of L4, but not affect read (which is already way higher than DDR3 2400Mhz) that much. and, even if you downclock your Uncore to 3.2, Read will still be way higher than DDR3 2400.
funny, 100Mhz of Uncore give exactly around 1000Mb/s on Write of L4 (on my pc) . try it in Aida64 and see it yourself.
I also tested 5775C with DDR3 2400Mhz C10, with Cinebench C15. CPU was overclocked at 4.4, and Uncore was 4.0
Edram disabled, with XPM disabled, and Ram speed downclocked to 1066Mhz, it get average result of 890. so, it confirm that this test also depends on Ram.
Edram disabled, with XPM enabled, so Ram speed is 2400Mhz, with timings 10-12-12-30-1T, it get average result 919.
Edram enabled, with XPM enabled, it get 936-938 score.
so, L4 matters and it is still worth it. but, take into account that Uncore was overclocked at 4.0 so that improved much L4 write and copy. similar testings need to be done with stock (3.3) Uncore speed, to see effect of L4.
funny, 100Mhz of Uncore give exactly around 1000Mb/s on Write of L4 (on my pc)
. try it in Aida64 and see it yourself.I also tested 5775C with DDR3 2400Mhz C10, with Cinebench C15. CPU was overclocked at 4.4, and Uncore was 4.0
Edram disabled, with XPM disabled, and Ram speed downclocked to 1066Mhz, it get average result of 890. so, it confirm that this test also depends on Ram.
Edram disabled, with XPM enabled, so Ram speed is 2400Mhz, with timings 10-12-12-30-1T, it get average result 919.
Edram enabled, with XPM enabled, it get 936-938 score.
so, L4 matters and it is still worth it. but, take into account that Uncore was overclocked at 4.0 so that improved much L4 write and copy. similar testings need to be done with stock (3.3) Uncore speed, to see effect of L4.
TRENDING THREADS
-
-
News Windows 11 will reportedly display a watermark if your PC does not support AI requirements- Started by Admin
- Replies: 18
-
Question Can I plug a docking station to a USB-C 2.0 port to have video output?- Started by jameli
- Replies: 2
-
Discussion What's your favourite video game you've been playing?- Started by amdfangirl
- Replies: 3K
Facebook
Twitter
Instagram