Discussion in 'General Hardware' started by lemonadesoda, Mar 31, 2009.
chuck216| Athlon 64 X2 5600+ Brisbane| 2.9 GHz | 6.49 |2.2379| Stock speeds
Here's the e1200@ 2.66Ghz (C1E/EIST enabled)
JrRacinFan | e1200 | 2.66 Ghz | 7.12 | 2.67 | Dual Channel DDR2 @ 800mhz CL5
I get raped by an i7...
thebeephaha| Q6600 | 3.60 Ghz | 18.09 | 5.025 | Dual channel DDR2
HammerON | E8500 | 4.33 | 12.76 | 2.947 | Dual Channel DDR 2
Melvis | FX-57 | 2.8GHz | 3.39 | 1.210 | Eat that P4's lol
Didn't like the high core voltage on last clock so here is a run at 4.0 GHz:
HammerON | E8500 | 4.0 | 11.77 | 2.942 | Dual Channel DDR2
Not sure if I did the math right on that score/clockspeed especially considering the number of cores involved...
FordGT90Concept - BY-2005 | Core i7 920 | 2.776 GHz (2.666 GHz w/ Turbo ON) | 19.21 | 6.92 | 4 physical, 4 virtual cores (HT ON) - Windows XP Professional x64 Edition - 3 x 2 GiB @ DDR3-1066 spec (tri-channel)
FordGT90Concept - SERVER | 2 x Xeon 5310 | 1.6 GHz | 16.54 | 10.3375 | 8 physical cores - Windows Server Standard x64 Edition - 8 x 1 GiB @ DDR2-667 spec (quad-channel)
hoss331 | Q9650 | 4.34 Ghz | 25.51 | 5.880 | DDR2
This is my 24/7 setup but ill do a ~4.8 run tomorrow.
Might want to put that chart in order. Just seems like random numbers, I assume pre Ghz Calc would be showing how much your getting per Ghz, don't know if you would want to order it like that, or by the score. Will be more understandable then.
btarunr | X4 9750 | 2.40 GHz | 11.64 | 4.85
Phenominal performance on those Xeons
The crunch per Ghz is very high. Could you say a little more about this system? Are you sure it is a 1.60Ghz, or is that the speedstep speed on idle? How many FBDIMM slots have you filled, and is the RAM x4, x8, single or dual rank? Thanks.
Could you kindly Geekbench so I can look at the components of your Dual Xeon performance more closely. THANKS.
I would love to sort it on score or score per Ghz, but the sort algorithm is buggy. It sort alphanumerically, but doesnt sort numbers very well. Try clicking the column headings to sort... and see what a good (or bad) job it does.
I wont sort manually. It takes too long, in fact, needs to be done for every new post. That really doubles up the work of keeping the table up to date. Just use the sort feature and "fix" it in your own minds eye.
User | CPU | Clockspeed | Score (Relative) | Score per clock Ghz | Any Comment
CyberDruid|920|1729.5 Corespeed X13 multi|20.94|7.8426966292134831460674157303371|Assumed 2.67 (stock) CPU speed. New to i7 lol.
Not sure if I did that score per ghz correctly or not.
EDIT Ah crap apparently the 920 is running at 1.7ghz now I gotta figure that out...
Here's the firstname.lastname@example.orgGhz
JrRacinFan | E5200 | 3.33 Ghz | 9.53 | 2.86 | Dual Channel DDR2 @ 800mhz CL4
Newtekie1 | Athlon X2 4400+ | 2.8GHz | 6.46 | 2.30 | Vista x64 Socket 939 Dual-Channel DDR1
1.6 GHz is stock on Xeon 5310.
It has 8 FB-DIMM slots, all of them (8 sticks total) have 1 GiB Kingston DDR-667 sticks in them. There is an OCZ memory cooler sitting on top of the memory providing direct airflow across the modules. It has two Delta 60x60x20 fans in it (the original fans died about a month or two ago, followed by the power supply ). They should be running in a pair of quad-channel configurations.
Heh, everything multithreaded is through the roof. Exactly what it was built to do.
There's only two ways to fix it:
1) Zero-pad the values (e.g. values 10000 and 900 would require 900 be changed to 00900 because 0 comes before 1).
2) Explicitly declare the column as an integer or floating point value which sorts by value instead of by alpha/numeric order. I doubt that feature is supported though because it complicates things as much as it simplifies them. XD
Newtekie1 | Celeron 325 | 2.53GHz | 1.47 | 0.58 | Server 2003
Fact is, you are getting better performance per Ghz than my E5420's with a bigger cache. So, there must be something in your FBDIMM setup that is giving you milage. Could you kindly give more information. Are they x4 or x8. Are they single or dual RANK?
I found no such option (unfortunately) in the table add-in for vB. If you know of a way to do this; great! Tell.
Ah sorry about that, I didn't realize that the chart had the ability to auto sort, very nice!
I have no idea.
It has four of these kits in this motherboard.
You got your processors overclocked 500 MHz. Work/clock is a measure of efficiency. The farther they are overclocked, the less efficient they become. If you return the clocks to stock, it may match or exceed mine. The harder the processor is worked, the more often it has to clear the pipes and try again (lost efficiency).
you missed mine
It's a conspiracy chuck!
Your StreamScore is very high. (Geekbench). Much higher than mine even though memory is at the same speed and my CPUs are nearly twice the speed. It is therefore a memory bottlenecked bechmark.
The only logical reason for your superior performance is the RANKING of your FBDIMM. Your sticks are dual rank, which are known to be faster than single rank (mine). So with 8 sticks you are running 16 ranks vs. my 4 ranks. That means you can handle more simultaneous reads and writes.
We all know that single access read or write on FBDIMM is slow. But it is interesting to see that for multi-threaded independent read and simultaneous write there IS some scaling with the FBDIMM memory architecture. But to get the most out of it you need more RANK.
So, if I want an extra bit of performance, I need to swap my single rank FBDIMMs for dual rank.
What? Just kidding, got you for the most part, but I think I'm going to write this down and say it to random people and see what kind of judgment they pass on me
Yeah, LOL. FBDIMM is a PITA to understand and manage. This whole channel and rank business is a real pest. ESPECIALLY since many people use passive RAM heatsinks that mean you can only use alternate RAM slots. If you dont use big passive sinks you HAVE to use a fan cooler on the memory.
With FBDIMM there are the following issues to "manage"
1./ Number of channels (you can do single, dual and quad channel not just dual channel like regular DDR). On some pimp server boards I believe octo channel was also possible.
2./ Number of ranks. You can have one, two, or four ranks per channel, however, many board limit you to a maximum of 16 ranks in total, so you cant over rank you memory channels. (Still with me? LOL)
3./ There are x4 and x8 arrangement of the memory. The x4 is a higher density RAM, has the benefit of a newer feature set, BUT, usually has lower rank, ie. single rank. So while it is newer and better, you need more sticks to get the best performance.
4./ 1.5v and 1.8v. FBDIMM gets horribly horribly hot (talking 80°C easy unless there is active cooling) and draws a hell of a lot of power. This is due to the on-stick memory controller. Everyone wants the 1.5v sticks fpr obvious reasons but they are not compatible with older chipsets that require 1.8v sticks. But the 1.5v are rare and expensive and in fact have stopped being manufactured for supply/demand/bankruptcy reasons.
5./ FBDIMM reads and writes at the same time using different "pins" in the socket. That is GREAT for simultaneous reads and writes, but BAD for "only" reading or "writing" due to lower bandwidth on just read or just write. In practice, FBDIMM has a higher bandwidth on "copy" and simultaneous read and write than regular DDR2 at the same clock. ie ideal for servers. BUT, FBDIMM doesnt have the ultra-high single function bandwidth of overclocked DDR2 or DDR3, and is therefore terrible at many single thread memory benchmarks.
>> The technology of FBDIMM is very similar to PCIe. ie it is serial data across multiple lines, rather than, parallel data on one bus. (are you still with me?)
OK - so now you've got an idea of the headache of FBDIMM. So knowing how best to use the memory architecture can be real hit or miss.
And now you know why people using FBDIMM are always asking DIMM questions. LOL
Separate names with a comma.