I'd be interested in a complete reverse set of tests, where the same memory module is used but at different timings. I haven't seen a memory benchmark based on timings in years, from any source.
Announcement
Collapse
No announcement yet.
DDR3-800MHz To DDR3-2133MHz Memory Testing With AMD's Kaveri
Collapse
X
-
Originally posted by Rexilion View PostI want my new setup to be futureproof
Comment
-
Originally posted by jakubo View Postwhat about timings? dont they have influence on the performance?
seeing the results i really must wonder why AMD did choose not to go all the way and go for GDDR5 and put some pressure on Intel. Or at least triple-channel...
Basically blame Intel, if Intel had backed GDDR5m you'd see it everywhere already, but since it was only AMD pushing for it it never made it out of R&D.
Comment
-
Originally posted by Rexilion View PostI want my new setup to be futureproof, and here are the downside's of the AMD FX 8350:
- No PCIe 3.0 support
- Bad single core performance
2) Single core performance is fine in everything today -- if you want 'futureproof,' well, all future games are threaded for 8+ cores. See how the PS4 and Xbox One both feature 8 core AMD processors.
Comment
-
Originally posted by mmstick View PostThe timings don't have much affect since the bottleneck is primarily a bandwidth one rather than a latency one. I remember that it was originally planned to have GDDR5 embedded on the chip but that was canned for the first gen Kaveri -- perhaps due to budget constraints. Instead of triple channel, I don't see why they couldn't just go straight for quad channel.
AMD even warns about this since many people new to overclocking do it very poorly.
Comment
-
-
Memory timing
I'd relly like to hear more on how the memory timings were set. Latency and bandwidth both can make large differences on different workloads. Many of the things already said in this thread hint at it, but I'd like to say it once and for all and in a clear way. There is one chart in the test that makes me think that the latench cycles aren't being held steady as the highest clock speed is slower than the one just one step slower.
Latency is best measured in absolute time. You can easily calculate that from the # of clocks and the duration of one clock. The first is normally specified in the spec for the memory (and can be read from the SPD on the memory) and can be set in the BIOS. The latter is just the reciprical of the clock speed 10^9/clock speed MHz. Multiply that by the # of clock cycles and you have the absolute latency in nanoseconds.
Testing really needs to do both variables independently. If you hold the # of clocks stead and change the clock speed, you alter both the absolute latency and the bandwidth of the memory and that confuses the results. It'll take longer--especially since messing with settings in the BIOS are never fun--but the results will be more meaningful.
Comment
-
Originally posted by Kivada View PostThe timings are the latency, you have to calculate the timings against the Mhz you are running the ram at to see if you are actually gaining anything. Push the timings too high for a higher Mhz rating and you end up being slower then if you backed off on the MHZ and tightened up the timings.
AMD even warns about this since many people new to overclocking do it very poorly.
Comment
Comment