For a new Intel i7-980X six-core system with Windows 7 64-bit, what is the optimum RAM?
I will not be using RAM for TBs, but rather an SSD drive.
One of our experts, Kullberg, has recommended 12 gig. of RAM.
In order to arrive at a good answer to the main question, there are several subsidiary questions that come first.
1) How much RAM can R4 use on a 64-bit O/S? I have often seen the answer "only 3 gig. of RAM", perhaps even from Vas. On the other hand, some experts have answered 1) with "As much RAM as you can afford."
2) Large Pages seem to have been one of the success stories of R4. I have inferred from postings that Large Pages can profitabely use a great deal of system RAM in addition to the RAM allocated to the R4 processes. Is that true, and if so, would that not conflict with "only 3 gig. of RAM for R4"?
3) What is optimum hash size for R4? If R4 overall can use only 3 gig. of RAM, then then answer to 3) would seem to be 2048 MB.
Thanks in advance to all the experts.
Here's a screenshot of Rybka with 8 GB hash and 117 GB of tablebases. No other programs are running.
98gb ram woohoo
2 x xenon x5680
btw especially purchased for rybka
Note that 96 GB works out to 103,...,...,... bytes.
It also works out to 98,... MB (assuming that MB is 1024*1024).
Possibly M$ doesn't care about those things and uses MB and MiB randomly (so the displayed Ram is not in MB, but in MiB, even if MB is displayed), but since 2007 you usually don't use those wrong units anymore: http://en.wikipedia.org/wiki/Binary_prefix
Do something to use more
>1) How much RAM can R4 use on a 64-bit O/S? I have often seen the answer "only 3 gig. of RAM", perhaps even from Vas. On the other hand, some experts have answered 1) with "As much RAM as you can afford."
The correct answer is 2 to the power of 64, which gives an answer of 18,446,744,073,709,551,616 bytes or 17,592,186,044,416 GB. This is the maximum amount of RAM which a 64-bit OS can have. This is why many expert will say "as much RAM as you can afford".
As for Rybka RAM usage, it depends on the GUI you are using. Fritz GUI is a 32 bit GUI, therefore, it will only allow 3 GB of RAM maximum. If you can find a Fritz 64-bit GUI, then you will be able to use more than the maximum 3 GB...
I am using Aquarium GUI.
If I understand Lukas' screenshot, he has 24 GB of system RAM and was using about 12 GB of RAM in that screen shot.
So my conclusion is that I should get either 12 GB or 16 GB of RAM.
What say you, Lukas?
That is good enough for me. 16 gig of RAM is official now!
The memory you need also depends on the number of cores and your operating system.
For an i7 980X you could go for 12 GB (6x2 GB) or 16 GB (4x4GB) or 24 GB (6x4 GB).
The problem is these 4 GB DDR3 modules are quite expensive. So probably 6x2 GB is a good choice.
Having determined the RAM issue (6x 2 GB of DDR3 RAM) to go with my i7 980X, my next step before going into the shop to have the system built was to have Phil Harris, who is on this board, recommend a detailed list of parts for the system.
Please feel free to add your 2 cents, if you think that for the money I could build a better R4 system with another approach than my starting point ( i7 980X).
I saw little old ladies at a computer shop who were going to put a system together themselves...ok that was uncalled for, but true.
Personally I would wait if I was going to go with Intel until the non-extreme versions of the 6-core processors came out and save some money overclocking that. If I were buying today, I would go with a duel Magny-Cours Opteron system. http://secure.newegg.com/WishList/PublicWishDetail.aspx?WishListNumber=9591569 And if I was going to wait 9 months to a year and a half, Bulldozer might be interesting. Most the fear mongers are worried about the floating point but chess programs don’t use much if any floating point. Sandy Bridge might be interesting too but we won’t really know until either are here. There is always something on the horizon.
I like the Magny-Cours chip because it uses a new platform and will likely be in use for years to come and it has an advantage: the sockets are big! That means AMD can cram more CPU on the chip even if its lithographic process is larger. That means more bang for the buck for a few years, unless Intel comes up with something vastly superior. Magny-Cours has 12 cores already, full cores not that hyperthreading nonsense. And AMD is not really loosing much with bigger chips because there are several new silicon foundries starting up because of increasing demand for solar panels, which brings down the price of silicon for both industries.
You imply that I'm wasting money with getting the i7 980X (not overclockable?), so why not a different i7 chip, since many posts on this board say that the i7 is better for R4 than any AMD chip?
The processor in your recommended system was the AMD Opteron 6128 Magny-Cours 2.0GHz 8 x 512KB L2 Cache. It seems to be 8-core but only 2.0 GHz, so can it compete with a much faster i7 chip?
Also, I think water cooling is tricky to install. I have it in my quad core.
My tournament starts in 9/10 so I don't have too much time to get a system.
http://www.cpubenchmark.net/high_end_cpus.html compares CPUs. On the chart the i7 X980 gets a score of 10,191. The Opteron 6128 gets 5,143. With an average overclock of 4302mhz the i7 X980 gets a score of 13,340. There is a multiple CPU chart http://www.cpubenchmark.net/multi_cpu.html but that and the chart we were just looking at will greatly underestimate the advantage of more cores because most apps make poor use of additional cores. Chess is an exception. Somewhere on this site Vas says what the gain due to more cores is. I can't remember what that is I am guessing it is something like 80%. I would bet that the 5,143 is not accurate when it comes to chess. The jump to 8 cores from 6 will likely be larger for chess. I suspect a 7000 or 8000 for chess but this is pure speculation on my part. Then add 80% of the first to the second and we get anywhere from 12,600 to 14,400. But it is hard to say; if anyone has access to such a machine please run a test! It could be higher. Some of the benchmarks incorporated were likely aided by hyperthreading which does not help Rybka. Just by itself the 8-thread could be up to 9,000 equivalent. If that were true, it in relative performance at chess might be as high as 16,200. But this is pure speculation until we actually get some numbers. I think if you went with the 12-core it would be virtually certain but I think there is a limit for the Fritz interface of 16 if I am not mistaken. I don't think Aquarium has any limit we are likely to reach though. There are faster versions of the 8-core but I doubt the speed-up is going to be worth the extra money.
But in a year when 16-cores or 24-cores come out you can probably get 12s or 16s for another $600 or $1000. That i7 X980...it will probably be left in the dust. But that is just a guess.
I don't know how useful this is but it is as closest to a head to head comparison as I can find: http://www.anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon/1
There is nothing growing in distilled water. The colored coolants they sell for $9 and more are what cause bacteria growth. They use propylene glycol which is just food for the little buggers.
"Besides cooling system breakdown, biological fouling also occurs. Once bacterial slime starts, the corrosion rate of the system increases. In systems where a glycol solution is maintained on a continuous basis, regular monitoring of freeze protection, pH, specific gravity, inhibitor level, color and biological contamination should be checked routinely. Propylene glycol should be replaced when it turns reddish in color." from Wikipedia http://en.wikipedia.org/wiki/Antifreeze
If you want color, add some food coloring ;)
The times I have opened it there was no odor or anything suspicious. There is no reason to be afraid of the light. I always use distilled water so nothing alive enters the system or very small amounts. Distilled can pull some metallic ions out of the copper but as long as it is a tight system it grabs what it can and then an equilibrium is reached fairly quickly it won't continue to erode the metal. If you are nervous you can have a piece of copper in your distilled water jug for a few weeks prior to using it. But be sure it is actually copper and not just appearing to be copper like some coins. If you are in a rush just use some sand paper or a file and put fine particles of copper in there an hour before and filter it with a piece of clean cloth before you pour it in.
You can get corrosion other ways...by having more than one kind of metal where you are in essence setting up a sort of battery. Most modern components are either plastic or they have coatings to avoid messing with copper.
I run three closed systems that have reservoirs, and algae formation can be a problem.
I suggest using this http://www.petrastechshop.com/pepcobi1.html
That has worked well in my systems.
If clean distilled water is used in a closed cooling system, it is not merely that there are no living things in it, it is that the components are not there from which to build cells. If you use the store-bought coolants they provide what is needed.
In my book I would not call those organisms that do grow in systems algae, as bacteria fits better, but call it what you want.
It is like you people are trying to convince a person walking up to you that they are crippled. There is nothing in my system but water! No algae! And I live in sunny San Diego...it gets all kinds of light.
They sell you propylene glycol to both make money selling you it and make money buying more to replace it when it goes bad and of course sell you some treatments so it won't go bad as fast. Of course I have to say that is just one person's hypotheses, for legal reasons. And you may want antifreeze to actually stop the water from freezing. I would not know about that, living in San Diego ;)
What; you think the quote from Wikipedia I gave is nonsense, that I just made it up? Read it for yourself: http://en.wikipedia.org/wiki/Antifreeze Read the last paragraph under the heading Propylene glycol. Glycol is an alcohol and like any alcohol has energy stored in bonds like food. It is made of oxygen, hydrogen, and carbon...the stuff that life requires. Distilled water has just water. Yes that is Oxygen and Hydrogen but it is at a low state there is no energy to obtain from breaking it down and of course it is missing the carbon.
If all you needed was sunlight how is it you can set out a bottle of distilled water in the sun for years and nothing will happen to it except a few dogs using it as a fire hydrant. No green no nothing...it's water! And it still has that oh so dangerous air at the top.
Since I started adding a couple of drops of copper sulphate to each one, I have not had any problems.
Oh, and the blue you see in the little water bottle; that is just a reflection from the LEDs in the power supply.
I used it in my Skulltrail and it worked very well. The only downside is the tubes turn pink after some time.
Also very thin layers do not generally help with conduction unless there would be a gape without them. That's why it is nice to have large heavy heatsinks...all that mass helps pull away the heat. Silver is the best metallic conductor but copper is not far behind. Silver is 429 and copper is 401 W/(m·K). Diamond is 900-2320. The jump from aluminum to copper is pretty big though. Aluminum is only 237 W/(m·K).
World's "blackest" material:
I think it would be cool (if somehow safe and stable) to have a jacket made out of that. It would be so dark that it would look very strange.
Pic is my lapped CPU.[img][/img]
Copper could be replaced by Silver and you get a thin silver layer (good) plus Copper chloride in the water :)
> Fritz GUI is a 32 bit GUI, therefore, it will only allow 3 GB of RAM maximum. If you can find a Fritz 64-bit GUI, then you will be able to use more than the maximum 3 GB...
I don't think this is correct: with the 32-bit Fritz GUI, running in a 64-bit Windows (XP Pro x64 for instance), you should be able to run and use more than 3 GB RAM.
See Lukas' screenshot just above: he has 24 GB, 8 GB are used for Rybka hash.
And for all I can see, he is using the 32-bit Fritz 11 GUI...
I assume that (with enough fast RAM in my W5560 machine) and with large-pages enabled, the more hash the better performance for Rybka, irrespective time-control?
I was trying to test this but did not reach conclusive result. But my logic tells me that 4GB of hash with large pages (using RTSingleServer) should be better than 1GB under any time control.
While I'm at it can I ask you another question? I have no clue how to determine optimal tablebase hash setting.
I have a very fast SSD RAID array with all 5 & 6 tbs. I thought about trying a RAM disk and put some of them there (but didn't try it yet).
At the moment I load into the engine both 5 & 6 tbs, and I do need to wait a short while for the engine to load. However when trying to measure engine performance with different tablebase hash size setting (in Fritz GUI) I could only see changes to the engine load time, but nothing much anywhere else. You probably know the absolute truth abou this topic :)
Thanks in advance.
>Lukas - do you run RTSingleServer as well as the UCI-Client on the same machine (127.0.0.1) in order to use more RAM for Rybka's hash within playchess or fritz GUI?
Yes. To use the same computer you can put in its name into the ini file instead of an ip.
>I assume that (with enough fast RAM in my W5560 machine) and with large-pages enabled, the more hash the better performance for Rybka, irrespective time-control?
There's only an advantage at longer time controls. But with large pages enabled there never is any adverse effect of using much hash.
As a rule of thumb doubling hash size adds 5 Elo points.
>While I'm at it can I ask you another question? I have no clue how to determine optimal tablebase hash setting.
Very good question. I've seen ill effects of using too big tablebases caches like sluggish unloading of Rybka. But if you can afford a reboot before the next start, that shouldn't matter.
>I thought about trying a RAM disk and put some of them there
I did that a long time ago using a 2x Opteron 275 computer. It worked very well. Way back then it was much faster than a 15k rpm SCSI drive. But nowadays there's next to no benefit if you use a fast SSD.
The task manager tells you how much performance you loose while accessing tablebases. If CPU usage is not below 95% this performance loss is negligible.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill