You are currently viewing our boards as a guest, which gives you limited access to view most discussions and access our other features. By joining our free community, you will have access to post topics, communicate privately with other members (PM), respond to polls, upload content, and access many other special features. Registration is fast, simple, and absolutely free, so please, <a href="/profile.php?mode=register">join our community today</a>!
I am not sure at what stepping Barcelona is now, but I just did some testing with the B2 stepping at 2.5 GHz, and it seems that contrary to what some people have been claiming, that this stepping has actually a slightly slower northbridge.
Raw Bandwidth numbers are up to 10% lower than the 2 GHz with an identical configuration (mobo, RAM). Further investigation necessary of course, but the scaling in specjbb (rather mediocre, on both 1.5.0.8 and 1.6 JVMs of sun) seems to confirm this.
Do you know if K10 still has the same issues as K8 regarding memory clock and divisors, at least in standard socket F? This might explain an anomalous result.
(Although just checking a K8 clock-memory table and I see both 2000 and 2500 will run DDR667 at full speed)
Do you know if K10 still has the same issues as K8 regarding memory clock and divisors, at least in standard socket F? This might explain an anomalous result.
(Although just checking a K8 clock-memory table and I see both 2000 and 2500 will run DDR667 at full speed)
2ghz K8 would run ram @667Mhz but 2.5Ghz would not(real ram clock would be 2500/8*2=625Mhz-so ~7% slower than DDR2-667)
Johan just said that B2 stepping produced poorer results. If AMD fixes that, then the new stepping will not be called "B2"...
It's getting confusing, what is the codename for that "dancing in the aisles" stepping? You know the one that reaches 2.9GHz and has 10% clock-to-clock advantage against current "buggy" steppings...
Johan just said that B2 stepping produced poorer results. If AMD fixes that, then the new stepping will not be called "B2"...
It's getting confusing, what is the codename for that "dancing in the aisles" stepping? You know the one that reaches 2.9GHz and has 10% clock-to-clock advantage against current "buggy" steppings...
Sorry, but something is wrong with Johan's configuration. Either the NB is clocked lower, or he doesn't have a B2. And FWIW, you have no idea how AMD names its steppings. If all the layers have been changed, it gets a new stepping number.
As you can see, 2.5 GHz is just a bad multiplier and the memory clock falls short by 6.25% of 333 MHz (double pumped to DDR2-667). 2.4 GHz is an even worse multiplier and 2.3 GHz is decent.
This might explain why the Barcelona 2.5 GHz processor only scales at 46.6% efficiency on SPECfp_rate2006 going from 2 to 2.5 GHz. For a 25% clock boost, it merely gained 12% on SPECfp_rate2006 (based on AMD's estimates used on AMD's advertising and website).
Opteron dual-core is in the 64% efficiency range scaling from 2 to 2.6 GHz. Opteron K8 dual-core had the same clock divider issue but the memory controller supplied so much extra bandwidth with respect to a dual-core processor that it didn't matter if you only got 90% of the memory clock speed. But when you take one integrated memory controller and make it perform double the work with four cores, it needs every shred of memory speed it can get and this might be the problem.
Now 2.4 works perfectly with DDR2-800 (fewer than 4 DIMMs per channel/CPU) but Sun is the only vendor that I am aware of that has DDR2-800 support. 2.3 and 2.6 GHz looks to be in decent shape with DDR2-667 but it’s still less than ideal.
Of course this is primarily a problem with bandwidth intensive applications and benchmarks and it does seem to indicate that AMD is up against some kind of wall on memory bandwidth in the next six months and they don’t have a massive cache and execution engine to overcome it. DDR2-800 might be the right savior on some systems but not everyone is going to be able to use it since their Barcelona servers don’t support DDR2-800. Even when we are using DDR2-800, there still appears to be a major issue – though better than with DDR-667 – since we’re only at 89.29% memory clock efficiency on some clock speeds.
Last edited by GeorgeOu on Sun Sep 30, 2007 12:34 pm, edited 1 time in total.
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot post attachments in this forum
