So is a 6m50s TPF (PPD=972k) on 8102 any good? :)

How is it going quickz, are you doing any progress? I have been thinking about why my rig is so much slower then yours, and i am wondering if it could be the memory. I use only 8 sticks, so the memory works just as 2-channel. Could this be why the machine folds slower then it should?
 
The number of memory channels plays an important role. If you can increase these to triple or quad, you will see a nice bump in performance.
 
As an eBay Associate, HardForum may earn from qualifying purchases.
You are using only 8 sticks of memory? If it was achieved with 8 memory sticks, your current score is not bad at all. On the contrary, I think it's really good.
I'm eager to see your score with 16 memory sticks.

How is it going quickz, are you doing any progress? I have been thinking about why my rig is so much slower then yours, and i am wondering if it could be the memory. I use only 8 sticks, so the memory works just as 2-channel. Could this be why the machine folds slower then it should?
 
Me too, I guess the new memory should arrive late next week, and I will be back with an update.

You are using only 8 sticks of memory? If it was achieved with 8 memory sticks, your current score is not bad at all. On the contrary, I think it's really good.
I'm eager to see your score with 16 memory sticks.
 
Upgrade from 8 to 16 sticks 2GB memory, from 2 channel to four channel. The power consumption rise from 705 to 725Watt, and the TPF went down from 7:25 to 6:52 for P8102. The amount of RAM is the same as before, at a total of 32GB. I am quite pleased, but hope that the TPF will be better after some time. If the rig can make 6:42 then I will pass one million points PPD. The PPD is for now 965K for P8102.
 
Good PPD increasement. Congrats!
But I think 1M PPD is very difficult for the current 4P E5-4650. I'll wait for the forthcoming 4P Opteron 6300s (ES). This one would probably beat 4P E5-4650 and pass the 1M PPD line (provided that we could OC it to a satisfied frequency).

Upgrade from 8 to 16 sticks 2GB memory, from 2 channel to four channel. The power consumption rise from 705 to 725Watt, and the TPF went down from 7:25 to 6:52 for P8102. The amount of RAM is the same as before, at a total of 32GB. I am quite pleased, but hope that the TPF will be better after some time. If the rig can make 6:42 then I will pass one million points PPD. The PPD is for now 965K for P8102.
 
Thanks!

Does the amount of memory have any influence on the performance?
I was under the impression that it generally does not have that big of an impact as long as you have "enough." memory bandwith, otoh, is a much bigger factor
 
Thanks!

Does the amount of memory have any influence on the performance?

As long as you have enough memory for the WU, OS, and Client, and have the appropriate number of sticks to meet the maximum multichannel memory requirements of the CPU in question, you will be fine.

With 48 cores on a GUI Linux install running a -bigadv work unit, I am using 3.4GB of ram.

More threads will take more memory, but not significantly more.
 
Eight GB for each CPU should be plenty then, regardless the number of cores / treads & OS!
 
Eight GB for each CPU should be plenty then, regardless the number of cores / treads & OS!

Oh yeah! That will be fine. You should not have any problems with memory if 4 slots per CPU are occupied.
 
Out of curiosity, I will attempt to populate all the 32 memory slots on the mobo to see if this has any effect on folding performance. This is a 4P G34 6176 rig, not on 4650 rig. If this makes the performance better I will try it also on the 4650 rigg. I will be back with an update later to day, or to morrow.

Oh yeah! That will be fine. You should not have any problems with memory if 4 slots per CPU are occupied.
 
In theory it would be no change for reg ram and hurt performance for non-reg ram. But I'm still curious about your real-world comparison result.


Out of curiosity, I will attempt to populate all the 32 memory slots on the mobo to see if this has any effect on folding performance. This is a 4P G34 6176 rig, not on 4650 rig. If this makes the performance better I will try it also on the 4650 rigg. I will be back with an update later to day, or to morrow.
 
It could possibly lower performance. 4 sticks is enough to max the memory controller. Going to 8 will force the memory controller to switch between dimms if the data is stored across more than 4 dimms per controller.

This is theory though. If there is a reduction in performance it would most likely be negligible. Might not even be detectable beyond the margin of error for our testing ability.

edit: Magny-cours (IL/PD) uses two dual channel controllers to handle 4 channels.
 
14% progress after I upgrade the memory for 16 to 32 sticks the TPF has gone from 14min to 13:55. The best I have seen on this rig with 16 sticks is 13:54, so the performance is about the same, no change I will say. Maybe this will change to something better when a brand new WU comes along, I hope, but time will show! The memory is Kingston Value none EEC 1GB RAM sticks, http://www.valueram.com/datasheets/KVR1333D3N9_1G.pdf

The power consumption did rose with 18Watt or so, as I asume comes from powering the 16 more sticks, from the specs for the spesific memory I can see that one stick use 1.14W, so that should be about right.

The log below shows the progress after more memory was installed:
Code:
[11:40:17] Project: 8101 (Run 2, Clone 8, Gen 40)
[11:40:17] 
[11:40:17] Assembly optimizations on if available.
[11:40:17] Entering M.D.
[11:40:23] Using Gromacs checkpoints
[11:40:28] Mapping NT from 48 to 48 
[11:41:06] Resuming from checkpoint
[11:41:37] Verified work/wudata_03.log
[11:41:38] Verified work/wudata_03.trr
[11:41:38] Verified work/wudata_03.xtc
[11:41:38] Verified work/wudata_03.edr
[11:41:38] Completed 90540 out of 250000 steps  (36%)
[11:52:36] Completed 92500 out of 250000 steps  (37%)
[12:06:36] Completed 95000 out of 250000 steps  (38%)
[12:20:37] Completed 97500 out of 250000 steps  (39%)
[12:34:34] Completed 100000 out of 250000 steps  (40%)
[12:48:34] Completed 102500 out of 250000 steps  (41%)
[13:02:33] Completed 105000 out of 250000 steps  (42%)
[13:16:30] Completed 107500 out of 250000 steps  (43%)
[13:30:28] Completed 110000 out of 250000 steps  (44%)
[13:44:24] Completed 112500 out of 250000 steps  (45%)
[13:58:21] Completed 115000 out of 250000 steps  (46%)
[14:12:22] Completed 117500 out of 250000 steps  (47%)
[14:26:21] Completed 120000 out of 250000 steps  (48%)
[14:40:15] Completed 122500 out of 250000 steps  (49%)
[14:54:09] Completed 125000 out of 250000 steps  (50%)
 
Last edited:
Hi, tear. Would you like to provide a link for this special version of TPC? Such tools are very needed for the intel platform. Thanks in advance!

I've got a TPC version that can do -dram on SB-E (has issues with node-socket mapping
but otherwise works). Flashing needs to be done in another board as we don't have I2C
support for these chips yet.
 
After 5 days with all memory slots populated I can see a better stability that shows in the way that the rig is now close to its best TPF more often then before the change. The bad side is that the power consumption rose with 18 - 25W, but it is hard to measure accurately since the power consumption goes up and down all the time. The RAM is non-reg.

In theory it would be no change for reg ram and hurt performance for non-reg ram. But I'm still curious about your real-world comparison result.
 
Last edited:
Good trying, so the previous theory of performance decreasing might be not very reliable.
However, you didn't benefit much from the extra 16 sticks either. Why not try low-latency DIMMs if you want to improve the memory performance? Replacing the memory with 16 sticks of CL6 or CL7 DRR3-1600 DIMMs you would probably get your rig significantly faster.
 
Replacing the memory with 16 sticks of CL6 or CL7 DRR3-1600 DIMMs ...

My wallet just yelped reading that. What kind of performance diff would be expected between CL7 and CL8 DDR3-1600? An extra $500 for CL7 might be hard to justify at least for myself.
 
I don't think all of CL6/CL7 memory would be so expensive. Just several days ago, I ordered some DDR3-1600 CL6 memory from my local supplier (memory model: G.Skill/F3-12800CL6D-4GBXH), the price is about $24 per stick.
However, as I pointed in another post, it's fairly hard to find low latency (and inexpensive) DDR3-1600 DIMMs nowadays. So, perhaps waiting for tear's OCNG4 BIOS is also a good choice, which would support DIMMs with non standard MTB values:
http://hardforum.com/showpost.php?p=1039269489&postcount=670

My wallet just yelped reading that. What kind of performance diff would be expected between CL7 and CL8 DDR3-1600? An extra $500 for CL7 might be hard to justify at least for myself.
 
No decreasing have taken place as far as I can se, but it has become more stable to stay close to the best TPF if you understand what I mean. That is positive! But the conclusion have to be that it is not a very good idea nevertheless, to populate all of the memory slots since the final bonus is negative, and in the form of more power use.

Good trying, so the previous theory of performance decreasing might be not very reliable.
However, you didn't benefit much from the extra 16 sticks either. Why not try low-latency DIMMs if you want to improve the memory performance? Replacing the memory with 16 sticks of CL6 or CL7 DRR3-1600 DIMMs you would probably get your rig significantly faster.
 
One question, will this be a good CPU trying to pull 1mill PPD: AMD Opteron ZS232545TGG45 ES 16Core 2.3GHz 16M/L3 Matches Opteron 6276 Tested http://www.ebay.com/itm/281014486937?ssPageName=STRK:MEBOFFX:IT&_trksid=p3984.m1427.l2649

Do you have anymore information about this version of TPC yet?

Hi, tear. Would you like to provide a link for this special version of TPC? Such tools are very needed for the intel platform. Thanks in advance!

Code:
Quote:
Originally Posted by tear [URL="http://hardforum.com/showthread.php?p=1039145713#post1039145713"][IMG]http://hardforum.com/images/buttons/viewpost.gif[/IMG][/URL]                 
[I]I've got a TPC version that can do -dram on SB-E (has issues with node-socket mapping
but otherwise works). Flashing needs to be done in another board as we don't have I2C
support for these chips yet.[/I]
 
Last edited:
As an eBay Associate, HardForum may earn from qualifying purchases.
I can't ensure that you could get 1M PPD from it, since that would need extremely high frequencies (in my estimation at least 3.6-3.7GHz, with 8102). This is possible but not too easy. However, I think it would be quite easy if you are only planning to get 800 - 900K PPD from it (with 8102).
I didn't have any new information about that special version of TPC. Perhaps tear is not ready to release it yet.

One question, will this be a good CPU trying to pull 1mill PPD: AMD Opteron ZS232545TGG45 ES 16Core 2.3GHz 16M/L3 Matches Opteron 6276 Tested http://www.ebay.com/itm/281014486937?ssPageName=STRK:MEBOFFX:IT&_trksid=p3984.m1427.l2649

Do you have anymore information about this version of TPC yet?



Code:
Quote:
Originally Posted by tear [URL="http://hardforum.com/showthread.php?p=1039145713#post1039145713"][IMG]http://hardforum.com/images/buttons/viewpost.gif[/IMG][/URL]                 
[I]I've got a TPC version that can do -dram on SB-E (has issues with node-socket mapping
but otherwise works). Flashing needs to be done in another board as we don't have I2C
support for these chips yet.[/I]
 
As an eBay Associate, HardForum may earn from qualifying purchases.
62xx is designed to work at higher frequencies than 61xx, I think it should also do better at overclocking. And in practice we know that brainbug has already managed his 2p 62xx running at 3.7GHz. I believe 4p 62xx at 3.7GHz or even higher is also possible, but the power delivery might be a big problem.

I was reading this http://hardforum.com/showpost.php?p=1039274752&postcount=12 topic and is wondering how he could OC that mobo & those chips til nearly 3.6GHz? Is it possible to do the same with a 4P SM mobo & 4 x 6276 ES chips? As you know, they are better for folding 810x then 61xx are. The whole thread is here: http://hardforum.com/showthread.php?t=1725074
 
It`s amazing, and I have the equipment to use 2 pieces og PSU on the same board. 1200W + 950W would keep the power need pretty good? But will this chips http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=281014486937&ssPageName=ADME:X:BOCOR:US:1123 with this stepping work in this H8QGi+-F mobo?

62xx is designed to work at higher frequencies than 61xx, I think it should also do better at overclocking. And in practice we know that brainbug has already managed his 2p 62xx running at 3.7GHz. I believe 4p 62xx at 3.7GHz or even higher is also possible, but the power delivery might be a big problem.
 
As an eBay Associate, HardForum may earn from qualifying purchases.
seller's feedback looks strange, many, many duplicate-appearing ratings
 
seller's feedback looks strange, many, many duplicate-appearing ratings
That's 'cuz you need more than one chip to fill a 4P! Don't worry - there are a lot of familiar names who have left that seller positive feedback.
 
One usually buy them all at the same time, so I agree with Linden, the names and feedback look kind of weird.

Update: I recognize some of the names to, and can see why they are repeated several times. This is not normal, but it must be the the system that is set up to make this layout as many number of items that were purchased.

That's 'cuz you need more than one chip to fill a 4P! Don't worry - there are a lot of familiar names who have left that seller positive feedback.
 
Last edited:
Looks to me like he listed each one separately instead of having the box to select how many you want.
Then everyone gave feedback for each piece.
I don't read anything bad into it.

I also recognized 2 of the buyers names.
 
Looks to me like he listed each one separately instead of having the box to select how many you want.
Then everyone gave feedback for each piece.
I don't read anything bad into it.

I also recognized 2 of the buyers names.

Agreed. Take a look a tthe auction numbers and you will see that they are all different, thus reinforcing what Brother john mentions above. ;)
 
Looks to me like he listed each one separately instead of having the box to select how many you want.
Then everyone gave feedback for each piece.
I don't read anything bad into it.

I also recognized 2 of the buyers names.

He had listed a ton of auction style listings. That's why he has similar ratings as a seller. I purchased from him 8 chips (the same as in that listing), but not on ebay. All are good, he ships fast with EMS and his communication is very well too.
Further, it seems like he won a ton of auctions from the same seller, and that explains the long list of positive ratings he got as a buyer.
 
The chips should be working fine on that board. The capacity of your PSU might be enough, but the board would be unlikely to accept so much power. I hear that H8QGL are doing better at power delivery.
PS, this might be a bit OT, open a new thread about 62xx extremely building would be fine. :)

It`s amazing, and I have the equipment to use 2 pieces og PSU on the same board. 1200W + 950W would keep the power need pretty good? But will this chips http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=281014486937&ssPageName=ADME:X:BOCOR:US:1123 with this stepping work in this H8QGi+-F mobo?
 
Last edited:
As an eBay Associate, HardForum may earn from qualifying purchases.
My E5-4650 has made an improvement in TPF with 3 seconds down to 06:49. I beleive I have seen 6:48 also but I forgot to make a picture of it.

What I hope is that tear has had time to work on the OC util for the Xeon E5-46xx & 26xx chips so some one could finally overcome 1 million record of PPD on a 4P machine?:D
 
Back
Top