GA-970A-DS3 slow memory benchmarks

[drbob]

New system with GA-970A-DS3 motherboard, FX-8350 processor, XFX Radeon 7850 2gb graphics, two 4gb Patriot Viper PC3-15000 memory modules rated at 9-10-9-27. MB has latest bios. Autodetected memory as only 1600mhz, manually set to 1833mhz and correct timing settings, but memory still benchmarks (Passmark at only 1473, close to half what I would have expected. According to CPU-Z, memory timing is correct and running in dual channel mode.
Individual Passmark memory results are all in line with overall benchmark, except that latency is about double what I think is should be (55.1ns) and threaded throughput looks very good at 23084 MBs. Any ideas??? CPU benchmark is right on the money at 9502. Graphics is a bit lower than I would expect, but I will address that later.

[David (PassMark)]

I didn't look up the specs of the RAM, but you might get better latency in the benchmark by lowering the clock speed and also lowering the timings. (assuming you can get tighter timings by lowering the clock speed on these modules).

[drbob]

Nope. Running memory at default detection (1600mhz) gives identical overall memory Passmark rating, latency actually goes up slightly. Then tried just fine tuning timing specs before changing clock speed and got no significant change. FYI, if I manually adjust clock speed up to 1866, but allow timing to remain set to auto, MB reports a boot up error. If I then manually change timing to listed SPD specs, machine runs just fine. Part number for memory is PV38G186C9KRD. Tried moving memory to other pair of slots, no difference. MB also has a nifty feature called EOCP (dram Enhanced OverClocking Profile). I tried using it and it works just fine to automatically set dram to 1866mhz etc., but came up with same test results. I've got a query in to Gigabyte, but no answer as yet. Memory is fairly new part number, and I found no reviews to speak of yet. Looking deep into mfg specs for similar memory it appears that it is not uncommon to label memory as 1866mhz while SPD is actually set to 1600. Motherboard model has gone thru several revisions, mine is version 1.1. There are tons of posts on Overclock.net for various issues, haven't seen anything regarding slow memory yet, but LOTS to slog thru.

[hawks]

AMD memory is slow no matter what. The highest score that i see on passmark is # 5312 and they overclocked the crap out of it. This is a big reason intel is so much faster I bet if you put the same memory in a intel the scores would shyrocket. But who cares about the passmark score i bet your pc is still fast. Anyone with more info please feel free to help this guy out

Memory Mark
#6991 - 16GB G Skill Intl PC3-12800 1714

Memory - Database Operations
#6991 - 16GB G Skill Intl PC3-12800 59.2

Memory - Read Cached
#6991 - 16GB G Skill Intl PC3-12800 21376

Memory - Read Uncached
#6991 - 16GB G Skill Intl PC3-12800 8218

Memory - Write
#6991 - 16GB G Skill Intl PC3-12800 6579


Memory - Available RAM
#6991 - 16GB G Skill Intl PC3-12800 14717

Memory - Latency
#6991 - 16GB G Skill Intl PC3-12800 50.6

Memory - Threaded
#6991 - 16GB G Skill Intl PC3-12800 20749

[David (PassMark)]

Unfortunately we don't have any results at all for the PV38G186C9KRD module. So it is hard to do any good analysis.

But Hawks is correct in saying that the current generation of Intel CPUs do significantly better that the current AMD CPUs in the memory test. However the slightly older Intel's (like the Core2 6600) are about on par with the current AMDs.

We'll look at adding some additional charts, or splitting up the existing charts by Intel and AMD, in order to make breakdown more obvious.

[hawks]

Yea you look up a ram benchmark it looks good and think great i will buy that, then go home and benchmark it and find out that great score was on a intel not a amd.

[David (PassMark)]

Partially true. Actually the score is an average and AMD and Intel (at the moment).
The RAM still has the potential to run as fast, or faster, than the numbers we are listing however.

Would be and identical problem if you stick you brand new high end RAM in a slightly older Intel system. Bottleneck will be the CPU and not the RAM.

We are going to do a separate AMD chart in any case. Should be up on the site next week.

[David (PassMark)]

We have done an initial split of the memory benchmark charts today.

Intel latency results are here,
http://www.memorybenchmark.net/latency_ddr3_intel.html

AMD latency results are here,
http://www.memorybenchmark.net/latency_ddr3_amd.html

These are a sub-set of all the results, taken from systems with newer (fast) CPUs. Systems wtih slow CPUs have been excluded.

The AMD DDR3 results are really not greatly better than DDR2. Meaning that,
A) The current AMD CPUs aren't really pushing the limits of current DDR3 RAM.
B) You can use pretty much any DDR3 RAM in an AMD system and not limits the systems CPU performance significantly. With the possible except of massively overclocked systems.
C) The AMD chart (above) isn't really all that interesting, as RAM performance doesn't matter much.
D) I am guessing a reasonable portion of the differences in the AMD charts are still related to configuration / BIOS settings and CPU, and not the performance of the RAM itself. This will probably become more evident over time as we get more results in.

Still the charts are useful from the point of view of knowing what to expect (and what not to expect) from an AMD system.

I should also mention there is a secondary reason to go for better RAM in an AMD system, even if it doesn't improve the CPUs performance much. On AMD chips that have an integrated GPU, the GPU also shares the bus and the RAM when doing 3D. The GPU can actually push the RAM harder than the CPU can in some cases.

So if you are planning on using AMD integrated graphics and doing gaming, picking good RAM is still important. But you might see the results appear as better 3D frame rates rather than better CPU throughput.