This thread is Spectacular:up: and full of Gigagurgle Goodness:shocked:
Printable View
This thread is Spectacular:up: and full of Gigagurgle Goodness:shocked:
Yes, I've been playing some games on it. Take note that Arma II does run with just one CPU installed with this amount of RAM, so I don't think that's the culprit.Quote:
Originally Posted by Chumbucket843
That sounds like a BI-studios game, allright:rolleyes: They make spectacular games and engines that scale amazingly with hardware, but there are all sorts of crazy bugs. They get most of them fixed in a few patches, though.Quote:
Originally Posted by Particle
That's another reason I'm a little weary of CPUs that don't have cores by the power of two, quite a few programs do not seem to utilize the cores appropriately when there aren't 2/4/8/16 etc cores.
Arma II uses incredibly small amounts of ram, at least the demo version. I've never seen it use more than 1GB on a 4 GB 64-bit Win 7 system, actually I can't remember seeing it utilize much more than 800 MB. With only 2 GB he'll probably experience a lot of page file usage in other games, though. He did say he'd get the ram addressed (:ROTF:) in a few days.Quote:
Originally Posted by Chumbucket843
I've got 4GB for now, Frodin. One 2GB dimm (per CPU). The extra RAM will provide some extra capacity but also allow me to go dual channel.
Ah, sorry, must have overlooked that. Still, it'll be interesting to see whether dual channel makes much of a difference, especially with those hexa-cores.
have u seen this particle?Quote:
Originally Posted by MechromancerPARTICLE! I have news from the PhenomMsrTweaker thread!
Quote:
Originally Posted by kink
Give that a try and let us know if the service allows you to change core VID :D. Also, post in that thread so kink knows just who the heck you are. I seriously don't know how he missed your Istanbul thread.:rofl:
Yes, it's definitely a lead. I'm not sure if it'll be able to exceed MaxVID (I don't know where the input rejection comes in--OS layer, power management layer, or lower [maybe even physical?]). I also still have to sort the problem of the program outright crashing on startup.
It seems you need to send kink a PM about his program there. I'm sure he would appreciate your input as you're our native Istanbul owner. I think you can get some productive work done with him:yepp:.Quote:
Originally Posted by Particle
I'm giddy just thinking about what you'll be able to accomplish when you get that program up and running for you. Also, Kink suggests making the modification directly in the registry. You should see PhenomMsrTweaker as a service when you type "services.msc" in the run menu. Disregard the GUI and head straight for the registry like a pro :D.
stuff still crashing on startup? man what's with that? chipset drivers? chipset is cool? idk why stuff crashes for u so much.
have you thought about quad socket for a new mobo? they support a whopping 128 gigs of memory
128 GB of fast ECC/REG memory will cost more than the mobo.
Also the cpu's will easily fetch about 2000 each.
http://www.newegg.com/Product/Produc...82E16819105255
http://www.newegg.com/Product/Produc...e=&srchInDesc=
im not saying buy 128 gigs but it would be totally hardcore if he bought $2k worth of ram! non techies would think its the fastest computer ever
That cpu is a 2p, not 4/8p
The 4/8p ones are usually the 8xxx series ones. They're easily over 1000 each.
Opteron 2435 + Particles HTT overclock and you'd have 3 Gig Istanbul ;) that would be sweet!
Thanks to justapost's help, my situation has become clear:
MaxVID is controlled by a read-only register: MSRC001_0071
The processor itself references any change to the current VID register against that one and the following takes place before signaling a change to the voltage controller:
If MaxVID < NewVID Then CurrentVID = MaxVID
ElseIf MinVID > NewVID Then CurrentVID = MinVID
Else CurrentVID = NewVID
Only processors with a MaxVID/MinVID of 00h are allowed to set to anything [except for 0V].
---
Anyone want to help me figure out a hardware mod? Perhaps it could be as simple as a pencil trick?
i think it will be possible :up:Quote:
Originally Posted by Nedjo
if it had that unlocking multipliers, it would be easy.. I could bet on 3,5ghz :D
but otherwise 3ghz sounds realistic
Dang that sucks on the voltage being a processor hardware limitation. Hopefully someone on here will know of a mod you can do to the CPU or board.
That is absolutely horrible! I suppose PhenomMsrTweaker WON'T be the solution to your problem anytime soon. Goodluck with the volt modding.Quote:
Originally Posted by Particle
Guess it makes sense that the voltage would be locked on server chips.
I'm sure they don't want some shady companies overclocking and selling rack systems at a higher price that may be prone to failure....
Good luck with the hardware mod, looks like the only option for now. :(
Just curious, did you ask justapost about the addy of the NBvid?
It's probably locked too, but it wouldn't hurt to try... ;)
He pointed me to the right documentation, but I looked it up myself. As such, I'll see what I can find since I know where to look now. :D
It appears as if the NB is subject to the same MaxVID and MinVID as the processor itself. At least this means I should be able to jack the NB to 1.225V if nothing else. It defaults to 1.1x. Who knows if that's what was actually becoming unstable instead of my cores. They clock synchronously on my system. CPU Clock = NB Clock
There are a lot of Phenoms that dont like NB higher than 2.4-2.6 range, so it easily could be the problem on these 6 core chips.
Also you can use a lower NB multi, check the BKDG there are pci registers for NB and HT speed whom you can change with bar edit. Dave did that 07 on an dual barcelona board with an broadcom chipset.Quote:
Originally Posted by Particle
This :rolleyes:Quote:
Originally Posted by EniGmA1987
My one can't do more than 2398MHz on default 1.175V
CPU is fine at 3GHz and 1.15V :)
If you can lower NB multi that will definitely gain you some MHz :up:
I know that attempts were done in ASUS L1N board thread ...
Hope it helps :)
As I recall, changing the NB divisor didn't actually change me, but that was really early in this project. I'll try again when I get home and see what happens.
That's already promising though for the quality of this silicon if my NB works fine at default VID (1.175V) at over 2.5GHz.
Wow, there's some really neat information in here. Good bye speculations of yesteryear--here are the required frequency relationships between NB and CPU:
- CPU can not run below 400 MHz
- NB frequency must obey: >= half of CPU Frequency AND <= 32 * CPU Frequency
So with a CPU frequency of 4000 MHz, the NB theoretically must be between 2000 MHz and 128,000 MHz. (Not that the silicon itself would do that upper limit, but it's required for the logic. This max is effectively a non-issue.)
You can change the NB multi in the pci register F3xD4 bits 4:0 (see pg 240 BKDG) after you changed those bits with baredit you must reboot before the value applies.Quote:
Originally Posted by Particle
http://www.abload.de/img/beditycqm.jpg
http://www.tweakers.fr/baredit.html
Can you be a little more specific on how to tell where to edit?
Exactly the last five bits as shown in above screenshot. I tested that yesterday with an Athlon XII here changed the last three bits from 110 to 100 rebooted and had an NB speed of 1600MHz instead of 2000MHz.Quote:
Originally Posted by Particle
With the new baredit version you must select "Configuration Space" in the first drop down, "Bus 0, Dev 24, Fct 3" in the second one and 32bit in the third one.
Then select row "D0" column "04" and last five bits should be 00111 change em to 00110 for 2GHz or 00101 for 1.8GHz or 00100 for 1.6GHz. Click write and reboot.
Excellent! Just regular big endian then. :up: I'll try this when I get home. Sorry for dragging this out, but as I've not been into the raw side of my computer before I didn't want to risk changing the wrong bits. My guess is that I'd just crash, but who knows? Maybe I'd flip the bool bit from 0 to 1 for the self destruct device? :D That would be just my luck.Quote:
Originally Posted by justapost
NP, I think you'll have to change it for both chips. The other one should be accessible via B0 D25 F3.Quote:
Originally Posted by Particle
NB FID seems to take effect without a reboot. (I rebooted anyway but couldn't tell a difference in benchmarks between 1600 and 2200 MHz.) No appreciable difference in overclocking. Thanks for the help though, justapost. Looks like I'm in need of some vcore.
Particle: "I have a fever and the only prescription is more vcore"
Get the circuit diagram for your mobo and get to volt modding! I think we have an entire forum for that here at XS :D.
More RAM is on the way! Single rank stuff this time...costs more, but it's all about getting the best (or: Newegg quit selling the cheaper stuff so this was my only choice. You decide. ;) )
Gom, sorry it took so long to do your request:
wPrime 1024M
http://www.pcrpg.org/pics/computer/o...wprime1024.png
why so i see 1.504v, did you get to overvolt it?
I think CpuZ is not showing the vcore correctly.
It doesn't. We're still at default vCore (1.225V).
any news?
Well if you are all up and running now, could you run some benches on this program and post your scores?
http://www.xtremesystems.org/forums/...d.php?t=221773
See how you compare to those high speed (some dual socket setups) Intel CPUs...
I'd really love to see that bench run too!Quote:
Originally Posted by EniGmA1987
+1 :up:Quote:
Originally Posted by Mechromancer
PPD for Folding would be nice as well. Specifically under Linux because the client is running the new A2 core which can handle more threads than Window's A1 based client.
After last night, I'm pretty well convinced that what is holding me back is the motherboard itself. Reminds me of the old days when overclocking HTT was tough because the boards just couldn't go that far.
HTT multi down from 5x to 4x.
CPU multi down from 11x to 9x.
NB multi down from 11x to 8x.
NB voltage up from 1.175 to 1.225V.
CPU voltage stock at 1.225V.
RAM timings at maximum board supports, DDR2-400 (physical memory is DDR2-800 spec).
PowerNow off.
I crash randomly when HTT is in the 230 range even when the above settings are active. Sample speeds versus stock at crash:
HT Ref: 235 [200]
HT Link: 940 [1000]
CPU: 2115 [2200]
NB: 1880 [2200]
RAM: DDR2-470 @ timings more relaxed than spec for DDR2-800.
The only thing overclocked at this point is the HT reference clock itself, and it's still crashing. As such, I don't think vcore is going to save me. I've got a fever, and the only cure is not just more vcore but a new RS800 motherboard this fall. It's too bad I can't just blow and trace bridges these days on the CPUs to get the multiplier I'm after. I wonder if it's possible to do a pin mod.
Of course its the Motherboard:) After all its a server motherboard and the Bios are locked for overclocking features and its not supposed to be overclocked at all:p:
Talk to Movieman he can help you overclock that thing I think he uses Systool that was developed by Wizard.
Also go to 2CPU Forum you can find out much about overclocking server boards on that forum.
Oh, I think I wasn't clear. I mean I think the board is physically incapable of generating an HT reference clock exceeding about 230 MHz with stability. It won't matter what software I use to attempt it. As for 2CPU, I don't think they're interested. My "planning a dual istanbul" thread over there went over like a lead balloon. No posts after two weeks.
You've got me wondering... I actually considered bringing out the multimeter, before I realised I wasn't really up to testing 1456849 combinations on at least 2 processors :rolleyes:Quote:
Originally Posted by Particle
Bet there's some help to narrow it down in the tech docs from AMD though..
if he tells amd he plans to have the fastest 2P system in the world, they might just send him unlocked CPUs
Hehe--good idea.
http://support.amd.com/us/Processor_TechDocs/31412.pdf
Look at page 22, karbon!
Is it possible to increase the chipset voltage? Maybe some water cooling and more voltage will do the trick. Chances are you've already gone down this obvious path so this post is probably in vain.
PS: Please give y-cruncher a try and let us know your scores. All us geeks want to know.
Y-Cruncher x64 SSE3:
12 x 2508 MHz
1M: 1.061s
2M: 1.761s
4M: 2.981s
8M: 5.184s
16M: 9.278s
32M: 16.710s
64M: 33.620s
My times hardly look competitive with anything.
Edit: Looks like y-cruncher doesn't hit all cores very well. It's often only using 40-60% CPU time.
hmm, disappointing :(
Wonder what is wrong with it that the program isn't hitting all 12 cores
The program author did a nice job explaining in his thread. Basically, he says I'd have to do a run in the billions of digits to see solid CPU usage.
I thought Y-Cruncher was hitting 100% on 16-thread Xeon systems. If they're pulling times like they are without 100% utilization, that says something.
EDIT: Particle, you need to horde ram like a dragon hording gold. You got a fever, and the only prescription is MORE RAM!
I'd be inclined to think real core count would matter more in terms of how badly the problem surfaces. For the HT people, they'll each have two threads running per physical core, and that's going to offer a much better chance of keeping them all loaded up.
might be hitting Hyper Transport bottleneck too.Quote:
Originally Posted by Particle
is at 1000 mhz?
1.14GHz HT between CPU and NB, but that isn't terribly important. The CPUs negotiate their own HT3 link offering 4.8GT/s.
Nvidia's replacement for Ntune has everything but vcore control. I, like you have hit an oc wall on my motherboard (Tyan S2927-E) but mine is needing some additional vcore to bump my 2389s past 3.45ghz.
That's socket 940 :(. I find references to the lga1207 funtional datasheet everywhere (#31118 and/or #41696), but not the papers themselves...Quote:
Originally Posted by Particle
Surprising, considering Socket F was designed to be used with FPGAs aswell as x86 CPUs...
Looks like the Istanbul processors use Fr6 which is document #45603.
Hawk, what HT ref do you get up to before crashing?
Quote:
Originally Posted by Particle
Well I just tested it and dropping the Multi down gave me no additional overclocking, so looks like anything over 234-235htt it's not stable.
I also noticed with clockgen that this motherboard (chipset likely) has no PCI lock.
That could be an issue if anything is on the PCI bus. All I have on mine is a XFI Audio Card, but I am not getting any wierd looping sound crashes that are typical with audio card problems.
I will say that my Asus KFN32-D SLI will do more than 240htt... but no shanghai support there.
Looks like we're mostly stuck in the same boat them. I don't think vCore is going to help us. Have you tried overvolting your chipset yet? I think that may be out only bet until new boards come out this fall.
I see you're using the Kingston ECC/Reg DDR2-800 like myself. Were you the chap buying out all of Newegg's stock a while back? I had to wait weeks to get a full compliment. heh
No chipset voltage control. No voltage control of any kind.Quote:
Originally Posted by Particle
No they were out of stock.. i bought directly from Kingston. 4 sticks atm, will buy 4 more next month.
I think the Newegg supply issue was due to Kingston changing over from dual to single rank.
I have the dual rank stuff:Quote:
Originally Posted by Particle
KVR800D2D8P6K2/4G
4GB 800MHz DDR2 ECC Reg with Parity CL6 DIMM (Kit of 2) Dual Rank, x8
The dual rank stuff went OOS for a week or so, then it deactivated entirely. A couple days later Newegg listed the modules again under a new listing and it was the single rank stuff (for $5 more, naturally). I've got a mix.
isn't the HT 3.0 for core to core even then it would start at 2400 stock at 235 you'd but to 2800. HT to Nb is also chip to chip isn't it?Quote:
Originally Posted by Particle
I don't know for sure if the CPU to CPU HT link is based off of the same reference clock or an internally generated one. HT links are always point to point, so in the case you asked, from CPU to NB.
Hows the system coming along? Had a chance to populate those memory slots and run a slew of benchies yet?
Anyone have any experience with ASA Computers? They're a Supermicro partner and all, but I'm curious as to why they're the first and only people to have the new H8DAi+ motherboards so far. I emailed them and they assure me the boards are in stock instead of a preorder. The price is fair.
http://www.asaservers.com/product.asp?pf_id=H8DAi%2B
What do you guys thing? Yay or nay on this vendor? I'm anxious to get started overclocking, using crossfire, and my IOMMU project, but I don't want to get burned for $400.
I thought I read you got the new boards with new chipset? Come on lets see some 3.2ghz :)
I've got the new board. I'm still working with the creator of SetFSB to see if he can add support for my clock generator.
Use the donate button -might speed up the work flow,lookingQuote:
Originally Posted by Particle
forward to the results/pics/etc from the new board :up:
Everest has some numbers with the 12c Istanbul's
already:
CPU CPU Clock Motherboard Chipset Memory CL-RCD-RP-RAS Score
12x Opteron 2431 2400 MHz Supermicro H8DI3+-F SR5690 Unganged Dual DDR2-800R 6-6-6-18 CR1 49041
8x Xeon X5550 HT 2666 MHz Supermicro X8DTN+ i5520 Triple DDR3-1333 9-9-9-24 CR1 42606
Nano L2200 1600 MHz VIA VB8001 CN896 Int. DDR2-667 SDRAM 5-5-5-12 CR2 41788
8x Xeon E5462 2800 MHz Intel S5400SF i5400 Quad DDR2-640FB 5-5-5-15 41561
http://www.xtremesystems.org/forums/...&postcount=110
That's curious. When I try to run Everest, it just BSODs my system (as it also did for my last motherboard) when it tries to measure CPU frequency. I put 3250 yen in his coffer. Maybe that'll help if he notices it was my email address. :) (That's about $35)
says on http://en.wikipedia.org/wiki/AMD_700_chipset_series
that Sp5100 is Sb700.
I don't know if that helps. :shrug: but it's worth a try.
CPU-Z lists my SP5100 as an SB750, actually. It's kind of funny. :) Unfortunately no, though, I need support for my specific clock generator. It's a different one than those used on virtually all 700-series boards.
If I limit Windows to 4 CPUs, AOD will load. I just don't have control over HT ref. Of course, it would be of limited usefulness to only be able to run with 4 CPUs, but it would have been a start. Apparently AOD crashes if it doesn't have enough display spaces for the number of CPU cores or GPU cores in a system.
:rofl::rofl::rofl: :ROTF::ROTF::ROTF::ROTF:Quote:
Originally Posted by Particle
Sorry, that's just so ridiculous it almost hurts. :D
Software Devs and their short-sightedness ("Who cares if it works, we can always patch it later, right? Let's go home now") :rolleyes:
Fire up your ICQ some time, by the way. ;)
Look at the RAID support. SB600/700/710 don't have RAID5, SB750 does.
SB700S -> SP5100 by the looks of it. Way before SB750 was a twinkle in our eye that Wikipedia article also mentioned a SB710S in the works with something like ACC.
For now, I'm sticking to it being SB750 due to the following:
http://www.pcrpg.org/pics/computer/s...5690_cpuz2.png
Attaching some results with "12x Opteron 2431 (2400 MHz)
Supermicro H8DI3+-F ".
-Since everest wont run for you (yet).
Have you got these thing overclocked yet ?!?
Can I make a request for this thread and others to follow-If a Thread this Awesome gets to this length-can we get it converted to an audio file so if I decide to reread it again in the future i can listen to it instead.Thank you:cool:
Particle did you ever get your Istanbul CPU's over 2.5GHz
Quote:
Originally Posted by Hawkeye4077
Hawkeye, so using ntune you were able to get a pair of shanghai's over 3GHz :clap:
Might Ntune work with a pair of Istanbul CPU's on an nforce chipset, or would I run into the same problem that particle is having without being able to increase the vcore.
My problem was never vcore. I couldn't get HTT up high enough for that to matter. Those ancient nVidia chipset boards (MCP55/nForce 3600 Pro) have unstable clock generators just like all the other AMD boards designed around that time. Remember those initial Socket 939 days where people still had to worry about how high they could crank HT ref? Same deal. Boards seem to get to ~220-230 MHz.
I've bought a new board based on AMD's SR5690 chipset, but nothing seems to exist that will allow me to overclock with it. AMD Overdrive crashes when there are > 4 CPUs in a system. I've been talking to the creator of SetFSB and even paid him a good chunk, but for whatever reason he hasn't ever added support for this clock generator. Now his page says something along the lines of he's not going to work on it anymore. I think I'm just boned. All I need is something that can change my HT reference clock for goodness sake.
Anyone know of anything other than SetFSB? I've got money. I can pay someone to develop this thing if that's what it'll take.
I think a lot of us here would like that. I know I am waiting for the new server socket to be launched to buya board based on this chipset, and it would be nice to finally be able to overclock again since I moved to dual socket computers.
Wow, that's morally bankrupt. He should give you the money back at the very least. It's a man's word worth anything these days?
David
Quote:
Originally Posted by Particle
It isn't quite like that. The financial relationship is an asynchronous one. He's got a donate button on his page that you can use. While I figure it can be used to get his attention, it doesn't make any promises for supporting whatever clock generators you'd like to see supported. He seemed disinterested, as every time I emailed him it was like starting all over again. He didn't appear to remember that I'd already contacted him and that he was going to work on it.
Thanks to a member of this forum, however, there is some hope. I've been in contact with some BIOS engineers who have been kind enough to take an active interest in my pursuit. They're going to try to get something going with any luck.
Has anybody succefully moded the bios on this board?
Hi particle,
I received one of those S8212 dual socket boards yesterday and I'm abit lost atm gettinge this kvm stuff working. I assume your supermicro board also has the aspeed 2050 chip with bmc and kvm capabilities. Do you have any monitoring or management software? tyan only shiped the board with an basic driver cd.
I was equally perplexed, unfortunately. I don't have any idea how that stuff is supposed to work. I looked around for a software front-end that could actually do something with that capability but didn't see anything from Supermicro. I'm not sure how they expect people to use it.
Well I asked tyan support and it's so easy. :)Quote:
Originally Posted by Particle
Quote:
The IPMI / KVM function does not require any tools. When the board POSTs, you can read the IP address on the POST screen. Also you can set the IP address in the Bios of your mainboard.
Then go to another PC and insert this IP in the address bar of your browser. You can then login with user: root and password: superuser.
In the webinterface you have the option to read the sensors and also to start the remote console which is the KVM function.