8-core SB-E ES Testing & Overclocking

pjkenned said:
Yea but you know those 8-core 3.0GHz Xeons in the wild will probably run north of $1k so interesting proposition if you were so inclined.
Click to expand...

maybe on fleabay but through retail your probably looking at north of 2k.
 
I can confirm that the Xeon 3.0 behaves the same way, anything over 107 and it starts to affect the pcie bus, and the vid card driver corrupts and gives a error 43. I do not believe it is because the chip cannot handle more, but because the x79 chipset does not work the same way with the Xeon chips. You have to also take into account that these chips have not been released officially, so the boards as well as OS have a hard time recognizing them, and there may be some core differences, pun intended, between the two chips.

In any event, I have been able to get the chip to oc to 3.5 with a 106.3 BLCK, but even then the chip throttles back to 3.1 fully loaded, and only bursts the bubble when a lesser load, i.e. graphics, is applied. Still the 3.0 delivers performance on par with heavily overclocked 3960x and 3930k, so it is not that they do not perform well.

Glad I caught this thread, was about to purchase another board thinking that my ASRock was not working properly. It may be the case that when the retail versions hit the market, they will have ability to oc a bit more, and mayhap, th
 
Archangel7 said:
I can confirm that the Xeon 3.0 behaves the same way, anything over 107 and it starts to affect the pcie bus, and the vid card driver corrupts and gives a error 43. I do not believe it is because the chip cannot handle more, but because the x79 chipset does not work the same way with the Xeon chips. You have to also take into account that these chips have not been released officially, so the boards as well as OS have a hard time recognizing them, and there may be some core differences, pun intended, between the two chips.

In any event, I have been able to get the chip to oc to 3.5 with a 106.3 BLCK, but even then the chip throttles back to 3.1 fully loaded, and only bursts the bubble when a lesser load, i.e. graphics, is applied. Still the 3.0 delivers performance on par with heavily overclocked 3960x and 3930k, so it is not that they do not perform well.

Glad I caught this thread, was about to purchase another board thinking that my ASRock was not working properly. It may be the case that when the retail versions hit the market, they will have ability to oc a bit more, and mayhap, th
Click to expand...

I will have a piece on this soon. The 3.0GHz is like the high end E5-2600 series... expect 107 to be around the max and not being able to work with straps.
 
Well it's possible that the Core i7 3820 was designed to work with the 125MHz strap setting rather than allowing us to overclock it by multiplier. Maybe they are throwing certain niche enthusiasts who want the LGA2011 platform's features without the extra cores a bone.
 
fastgeek said:
So are these processors different than the SB EP/EX series? Been abusing two 8/16 processors in an unreleased Dell R720xd for a couple weeks now and they seem to be doing just fine. Since it's a Dell there's no OC'ing though.
Click to expand...

EP = E5 which is what these are. I think -EX is the E7 series now. Pretty big difference.
 
As of Sandy Bridge, the "EP" series now supports 4 sockets per system.
SB EX is for high-reliability systems/giant 256-way machines and will still live on Socket 1567, or so the rumors say.
 
Dan_D said:
Well it's possible that the Core i7 3820 was designed to work with the 125MHz strap setting rather than allowing us to overclock it by multiplier. Maybe they are throwing certain niche enthusiasts who want the LGA2011 platform's features without the extra cores a bone.
Click to expand...

Is this something [H] would be interested in doing an article on? It would be very interesting to see if straps work with this chip in boards where it otherwise didn't. If it doesn't work, we may be onto a BIOS implementation issue that spans several manufacturers...
 
R-Type said:
Is this something [H] would be interested in doing an article on? It would be very interesting to see if straps work with this chip in boards where it otherwise didn't. If it doesn't work, we may be onto a BIOS implementation issue that spans several manufacturers...
Click to expand...

Well until we try with a Core i7 3820, we won't know for sure. I wouldn't think very many would be interested in this CPU. If you are going LGA2011, I'd think you'd step up to the 3930K at least. Otherwise there isn't much point to it over LGA1155 setups.
 
Dan_D said:
Well until we try with a Core i7 3820, we won't know for sure. I wouldn't think very many would be interested in this CPU. If you are going LGA2011, I'd think you'd step up to the 3930K at least. Otherwise there isn't much point to it over LGA1155 setups.
Click to expand...

I was thinking about a more broad based 2011 overclocking article, a few different chip and board combinations to explore the options for non-multi overclocking. I understand this is easier said than done, but it would certainly be interesting.
 
I hate to say this... in the Xeon world, I totally get LGA 2011... but the enthusiast 4-core makes little sense unless for some reason you really needed a ton of memory and PCIe slots, but not much processing power... and didn't need ECC memory.

Frankly... I think all LGA 2011 should just be UP Xeons at this point.
 
R-Type said:
I was thinking about a more broad based 2011 overclocking article, a few different chip and board combinations to explore the options for non-multi overclocking. I understand this is easier said than done, but it would certainly be interesting.
Click to expand...

Well from a bus clocking standpoint, 1-7MHz is pretty useless and pathetic. I don't see it as worth while. So far, the strap settings haven't gained a higher overclock the one time I got that to work.
 
The Xeon E5 series (what you have there) is not released yet, but there are a ton of ES (like that one) around.
 
Just as a FYI, I have mine at just over 3.2GHz on 8 cores. Runs fine with a H80 cooling. 6903 in Linux it appears to be doing ~36min TPF using an old SSD I had everything setup on.
 
pjkenned said:
Just as a FYI, I have mine at just over 3.2GHz on 8 cores. Runs fine with a H80 cooling. 6903 in Linux it appears to be doing ~36min TPF using an old SSD I had everything setup on.
Click to expand...

4 minutes faster than my dual x5570 machine, 8 cores 16 threads @ 3.2Ghz.
 
pjkenned said:
Just as a FYI, I have mine at just over 3.2GHz on 8 cores. Runs fine with a H80 cooling. 6903 in Linux it appears to be doing ~36min TPF using an old SSD I had everything setup on.
Click to expand...

Thats about 2 minutes faster than 24 threads at 2.4 on L5640's, but 8 minutes slower than 24threads at 3.2 (x5670's)
 
Yea assuming it is linear, that means 18min TPF on a 6903 using two $1800 CPUs. Even using ES chips it is probably slower than a similarly priced (off ebay) AMD 4P boxen.

Also as a data point, I have to keep the H80 at high or else the thing crashes even with medium fan settings. Sounds like a leaf-blower. 150w + anything OC'd is dumping a lot of heat.
5
The big question is whether you will need the Xeon E5-2600's for the new SR-3 type boards. If so, and you are limited to a 6-7% overclock (and this was at 4%) meaning bang for the buck just isn't there unless Intel actually makes the platform worthwhile.
 
Thread resurrection...

I have picked up a couple of 2.3GHz Intel Xeon 8-core QA8E ES (LGA2011) without doing enough research. My first go was a dual lga2011 motherboard from a thinkstation d30... (not a good idea).


I am now looking at either the Asus Z9PE-D8 WS or the Asus P9X79 WS. I have found instances where both of these board have worked on this early stepping. So the next question is one of BIOS.

In general will the latest bios support these or am I likely to have to search for the magic BIOS number that works? and If so is there anyone here who knows the answer :).

Thanks,
Robert
 
You are being brave, I've got a pair of B stepping chips and its not always a pleasant experience working with them.

On a Z9PE-D8 the latest BIOS is a complete no no, anything beyond 3506 will not support QAxx cpu's. Your best bet is to try 0503 or 0703 and then see if the board is stable enough for your purposes. If not you will need to flash to 3109 and then 3506. 3302 is a buggy mess and I could not get mine to boot. If you aren't running Maxwell gpu's you should be OK on either 0503 or 0703 depending on what you use the rig for - mine is a pure folding rig and for that either BIOS works great.
 
Nathan_P said:
You are being brave, I've got a pair of B stepping chips and its not always a pleasant experience working with them.

On a Z9PE-D8 the latest BIOS is a complete no no, anything beyond 3506 will not support QAxx cpu's. Your best bet is to try 0503 or 0703 and then see if the board is stable enough for your purposes. If not you will need to flash to 3109 and then 3506. 3302 is a buggy mess and I could not get mine to boot. If you aren't running Maxwell gpu's you should be OK on either 0503 or 0703 depending on what you use the rig for - mine is a pure folding rig and for that either BIOS works great.
Click to expand...

Does the Z9PE have the upgrade bios without a CPU from usb stick feature or will I have to find a cheap x79 chip to boot with as I flip BIOS version?

Not brave, foolish. I already have the cpus in hand so it is worth a shot. If I have to sell them and shop for later stepping then I will. I also have 2 - Xeon Phi cards with water blocks from alpha cool and likely to pick up 2 more. Video card is a fermi (580gtx)... May upgrade to a 7970 or 90 when a deal comes around but this is intended to be a workstation for a Machine Learning sandbox. No gaming here. Esssentially a single node of the Tianhe-2 supercomputer - 2 8 core e5's, 4 xeon phi's and plenty of ram. (32gb ecc so far but I would really prefer 128 (my wallet not so much - another 8x4gb to make 64 is most likely till I feel a memory pinch). The xeon phis are the 3120 server versions so 270 watts per card to dissapate. Will likely need a second power supply yet. I only have a kilowatt ;)

I've got a ThermalTake core x9 box which should give me more than enough room for 2 -480mm rads, pumps and reservoir. I will probably have to rig the cards on pci extension cables to fit them all in ( 5 double width cards). There's plenty of room if I use 2.5" ssds and pull the most of the drive bays.

At this point I would just like to see the GD thing post and get linux installed.
 
What do you plan to use the Phis for? Can't say I ever got one of those buggy POSs to do anything.

As for B0 chips if this is a new board you already have too new of a BIOS. I am not sure Asus lets you downgrade. Has anyone gotten that to work? Also I have seen newer BIOSes work on my Asus 2P board, but you had to upgrade from an older BIOS (or perhaps have an older board).

In my B0 adventure I had too new of a board, so I eventually returned the B0s and got C0s or higher. I do have a B0 running perfectly in an x79 1P Asrock boards. In 2P the boards are really picky and I donot know of new board that works well with them.
 
As far as downgrading asus boards there is a hack thank goodness...
http://smarttechtips.blogspot.ca/2012/08/how-to-donwngrade-asus-bios.html.

It may still not work but the z9pe and P9X79 definitely support the memory model that the phi's require to run. Hopefully they got the bugs out of that in the early bios releases and the stars will align.

I have seen a couple of ebay seller's claiming working status of this stepping on the W9PE so I will send them email for the BIOS version. I would really like to have the pci-e lanes that come with the dual cpu motherboard. The single cpu just doesn't have enough real lanes to be efficient in passing data back and forth between the phi's and the main cpus.


I have several projects in mind for my box. I'm going to do a couple of machine learning and data mining courses. Then I will use the box for 1) Kaggle Contests.. As a learning tool 2) See If I can make GnuBG smarter by porting the models to a scaleable architecture 3) I am fascinated by technical trading and wonder if there are not still opportunities for small traders with big engines.

The fp64 is about a teraflop per card when it is scaling efficiently. What is more appealing is the the phi's are running uos. micro-linux kernel and busybox. intel's MPSS sits on top of this or OpenCL for parellelizing the code. I have two of the 31S1P's and will probably add a couiple of the 5110P's The phi's have 6GB of shared DDR. The ones I have, have 57 cpu cores (essentially pentium 5's with the crap stripped out and some nice wide pipes and fp64 instructions added in) with 6GB of shared DDR5 memory. It provides a lot more flexibility than a GPU for distributed development than leveraging a GPU.

NVidia has been pushing the benefit of fp32 but that is mostly marketing hype for their cripple chip. There is a reason there wont' be any new Teslas till Pascal comes out and for general compute a gpu doesn't make a lot of sense. Great solution for pure matrix vector math but not everything is matrix math. Sometimes being able to run 200 versions of a general program and sharing memory is what you want to do or in the case of the Tianhe-2 200 versions of a general program * 10,000 or so nodes.
 
Ah let me clarify - I got my Phi to work on a Z9PE-D16/2L board just fine. After that though the Intel compilers and drivers were a bit of a disappointment. Maybe they have improved since.

Unfortunately the BIOS information isn't necessarily enough. I made that mistake. The seller for my B0s had a screenshot of the chips running on a very recent bios (at the time). My board with the same BIOS was a no go. Judging by the good support he provided I do not think he was lying, but it is possible. If he was telling the truth there is more to it than the BIOS. Something that perhaps isn't updated with the BIOS, or even hardware? At any rate I would try one of the really old BIOSes and see what happens.
 
Not sure if the Z9PE will boot without a CPU.

you can downgrade some Bios, I think you can drop from 3506 to 3109, etc, what you can't do is drop from 3xxx to 0xxx, they have changed the format of the bios.
 
You must log in or register to reply here.
Back
Top