Message boards :
Number crunching :
Building a 32 thread xeon system doesn't need to cost a lot
Message board moderation
Author | Message |
---|---|
Dr Grey Send message Joined: 27 May 99 Posts: 154 Credit: 104,147,344 RAC: 21 |
This article says 8 core xeons are going pretty cheap on ebay right now I've got to say it's a tempting build if you don't mind the power bills. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
There is a thread over on Einsten@home where someone wanted to make a CPU only cruncher. They were initially looking to use older 5000 series Xeons. Some users talked them into a system with s pair of E5-2670 or E5-2660's given how cheap they are. If you are lucky you can even manage to get them for under $50. Paired with a workstation MB instead of a server one & you can probably spend even less for 16c/32t of crunching fun. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Sidewinder Send message Joined: 15 Nov 09 Posts: 100 Credit: 79,432,465 RAC: 0 |
It's pretty much what I did with one of my crunchers. I actually bought (2) Xeon E5-2670s (Sandy Bridge-EP) and am looking to replace my old Q8300 with the 2nd Xeon. The most expensive piece was, and usually is, the motherboard. One word of caution if you're building with one of these is that many motherboards for this socket (especially the server-grade mobos) come with Narrow ILM sockets (e.g. this one) and thus require a Narrow ILM CPU cooler. There are very few (and fewer good) Narrow ILM coolers. So I actually spent a little more money and bought a AIO water cooler for it. If you buy water coolers that use the Asetek plate assembly, you can grab a cheap Narrow ILM retention kit from their eBay store (here). |
Gamboleer Send message Joined: 3 Jun 06 Posts: 29 Credit: 12,391,598 RAC: 0 |
I recently built two of these, one with dual 2660's and one with dual 2670's. I mostly do Einstein@Home and wanted to put more CPU time on Gravitational Wave searches. All processors ran me about $60 each on eBay. Although I'm sure the vendors are all about the same, I bought from "gfsi". He accepted Best Offer on his CPUs when I offered about 90% of his asking price. The 2670 system is on an ATX board, ASUS Z9PA-D8. It goes in and out of stock on Amazon and Newegg. You do need the narrow CPU coolers on this one; I'm running dual Cooler Master Hyper 212 EVOs. Fans go on the outside of each radiator, and there's enough room to put a third inbetween. One CPU runs slightly hotter than the other because it's sucking the first one's exhaust, but both are under 60c running 100%. The 2660 system is on an ASRock EP2C602-4L/D16 board. This one is SSI/EEB form factor, which requires a special oversized case if you're doing a tower. The CPUs on this one are far enough apart that you have a wide choice of CPU coolers. I picked these boards because they were dual E5 boards that were upgradeable to the V2 series, which I hope means I have an inexpensive upgrade path when the V2 CPUs get replaced and show up cheaply on eBay. Both boards are running fine with normal DDR3-1600 non-ECC gamer RAM (the fastest you can use with a V1 series E5; V2's can handle 1866). Both boards can also handle dual PCI-E 2.0 16x full length graphics cards for some GPU computing, another factor in my decision. I'm running Windows 10 Pro for the OS. There is very little performance gain going from 2660 to 2670, but the 2670 has a 115w TDP versus 95w for the 2660. I would recommend the 2660 for a build. |
AMDave Send message Joined: 9 Mar 01 Posts: 234 Credit: 11,671,730 RAC: 0 |
For those multi-core maniacs out there, I would recommend you view the May 2016 edition of PC PRO magazine, pgs 84-85, "i7 or Xeon." It mentions Haswell, Broadwell, and Skylake chip families. It lists specs on a new chip --> Broadwell-EP Xeon-E5 2602 v4. Also, it touches upon chips with 4, 8, 10, 12, 22, and wait for it...wait for it...26 cores. The latter is an upcoming skylake Xeon, which is rumored to have 26 cores, and 65MB L3 cache. It is slated for release in Q1 2017. Imagine dropping two of these HTT beasts on a mobo with 4 graphics cards. That's 100< SETI WUs crunching concurrently on a single box. Of course, you'd need to win the lottery first. |
Dr Grey Send message Joined: 27 May 99 Posts: 154 Credit: 104,147,344 RAC: 21 |
Of course, you'd need to win the lottery first. Well my money's on the Grand National for tomorrow. Fingers crossed. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
For those multi-core maniacs out there, I would recommend you view the May 2016 edition of PC PRO magazine, pgs 84-85, "i7 or Xeon." It mentions Haswell, Broadwell, and Skylake chip families. It lists specs on a new chip --> Broadwell-EP Xeon-E5 2602 v4. I hope they didn't bite on the E5-2602v4 5.1GHz rumor that has been going around for months. To fit in line below the E5-2603v4 it will likely be a 4c or 6c at 1.5GHz. I would estimate the E5-2699v5 will be 26c/52t @ 2.1GHz. Based on how the previous generations has progressed thus far. With the E5-2699v4 having 22c/44t @2.2GHz and the E5-2699v3 having 18c/36t @ 2.3GHz. Running 100 SETI@home tasks at once with 0 cache would not be ideal. My 12c/24t server often would switch to backup projects when there were only minor SETI@home server hiccups. However back the E5-2660/E5-2670 that weer flooded onto the market and can be had for cheap. Now if someone would just flood dual LGA2011 boards into the market so they price on them would drop below $200 that would be nice. In specing out the parts for these CPUs the MB is the single most expensive part. Unless you go nuts on memory. Then you can easily spend over $1,000 just on memory. 32GB & 64GB simms are not cheap. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
AMDave Send message Joined: 9 Mar 01 Posts: 234 Credit: 11,671,730 RAC: 0 |
I hope they didn't bite on the E5-2602v4 5.1GHz rumor that has been going around for months. Y-u-u-p. The author did not indicate that it was a rumor. However, he did write that the yet-to-be-releaesed Skylake Xeon was rumored to have 26 cores. Imagine dropping two of these HTT beasts on a mobo with 4 graphics cards. That's 100< SETI WUs crunching concurrently on a single box. In hindsight, that would be wasteful. It would complete 100 WUs in what, a couple of hours, then it wouldn't be able to download any more until the next day. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13755 Credit: 208,696,464 RAC: 304 |
It would complete 100 WUs in what, a couple of hours, then it wouldn't be able to download any more until the next day. Why not? The 100 WU limit is per device so you would have 100 WUs per GPU and 100 WUs per CPU. As WUs are returned, you would download more to fill the cache to the limit of the cache or the server side limit; whichever comes first (which for most current hardware is the server side limit). Of course if you returned nothing but errors, then you would eventually be limited to 1 WU per day until you stared returning valid work. Grant Darwin NT |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Yes but it's 100 WUs for all cores, not 100 WUs per core. I get exactly 300 WUs for 2 GPUs and 4 cores, that's 100 per GPU and only 100 for my quad core... [edit] So it seems useless to have more than 100 cores, because they will stop processing at the slightest server hiccup. Aloha, Uli |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13755 Credit: 208,696,464 RAC: 304 |
Yes but it's 100 WUs for all cores, not 100 WUs per core. I still don't see the issue. If you had a second CPU, you would get another 100WU. Once you finish a WU, you can then download another. The 100WU limit per device is not a per day limit. EDIT- So it seems useless to have more than 100 cores, because they will stop processing at the slightest server hiccup. If it were 100 cores on a single CPU, yes. But as it's on multiple CPUs then it's not an issue (if my understanding of the limits is correct). Anyone with a multi-CPU system able to advise whether the CPU limit is 100WUs per system, or per CPU? Grant Darwin NT |
AMDave Send message Joined: 9 Mar 01 Posts: 234 Credit: 11,671,730 RAC: 0 |
It would complete 100 WUs in what, a couple of hours, then it wouldn't be able to download any more until the next day. I didn't know it was so granular. I thought device = computer. So, with 4 GPUs, each one utilizing a logical cpu, such a system would complete @10,400 WUs/day, which in turn would be @1,040,000 credits/day. Rough guesstimates, of course. Although it would depend on the GPUs installed, I shudder to think of the amount of electricity needed/day. EDIT: Scratch my guesstimating. I composed during the posting of the previous two messages. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13755 Credit: 208,696,464 RAC: 304 |
I didn't know it what so granular. I thought device = computer. To me device = CPU and GPU. It certainly is that way for GPUs, but now I'm not so sure about CPUs. Hoping someone with a multi socket system will inform us. Is the CPU limit per CPU or per system? Grant Darwin NT |
AMDave Send message Joined: 9 Mar 01 Posts: 234 Credit: 11,671,730 RAC: 0 |
I didn't know it what so granular. I thought device = computer. Ditto |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Is the CPU limit per CPU or per system? Indeed, a good question! I didn't recognized it's 2 CPUs. Aloha, Uli |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Yes but it's 100 WUs for all cores, not 100 WUs per core. Actually it's a hard limit of 100 CPU tasks. Not per CPU socket. When I was running my dual 6c/12t server and we transitioned from 100 GPU tasks to 100 GPU tasks per device per vendor I asked if the CPU limit could be treated the same. However I'm unsure that BOINC has a detection method to determine the number of CPUs. I believe BOINC only detects the number of cores/threads present in the system. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
AMDave Send message Joined: 9 Mar 01 Posts: 234 Credit: 11,671,730 RAC: 0 |
Yes but it's 100 WUs for all cores, not 100 WUs per core. I run 1 WU at a time on my GPU and, according to SETIspirit, the GPU completes (100 < WUs)/day. ...we transitioned from 100 GPU tasks to 100 GPU tasks per device per vendor... Not quite following you. Does this mean with 4 cards installed, the limit would be 400 (GPU) WUs/day? |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Yes but it's 100 WUs for all cores, not 100 WUs per core. If you look at the Application details on one of your hosts you can see the number of tasks it has completed for each app in a given day. Number of tasks today SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
AMDave Send message Joined: 9 Mar 01 Posts: 234 Credit: 11,671,730 RAC: 0 |
Yes but it's 100 WUs for all cores, not 100 WUs per core. 970 WUs today?!? I've never felt so productive in my life :) |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Yes but it's 100 WUs for all cores, not 100 WUs per core. Well, that's obviously not one of your systems. I just grabbed one form the top hosts list as you have your systems hidden. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.