Message boards :
Number crunching :
Dual E5-2670 vs. Threadripper 16 core ??
Message board moderation
Author | Message |
---|---|
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
I have 2 dual E5-2670 machines, each with 16 cores running a total of 32 threads each, same as the Threadripper 1950x. Ignoring the cost factor, I would like to know if anyone is running the latter, and can offer some comparisons between the 2 setups, in terms of processing SETI WUs. Since I am running dual GTX 1080s on each machine, 3 WUs/GPU, I am running 25 threads of CPU WUs on each. Both machines are running 80K+ RAC, currently #31 and 33 on the computer list. Just curious, as I would have to kill an older relative to get the $$$$ for a Threadripper machine... |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Comparing some of the Ryzen 7 1800X systems to my E5-2670 I have been a bit disappointed in the performance of of the Ryzen systems. With a about 1GHz higher clock and DDR4 memory I would expected much greater performance. Maybe the Threadripper version will be better. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I took a look at your E5-2670. It is not all that impressive in regard to your APR or number of tasks per day validated. I don't know which Ryzen7 1800X systems you are comparing it too. Could you link the hosts of your comparison. I'm curious. I also wonder if those systems are stock or have been overclocked to what they are capable of. Your Host 8130144 Astropulse (anonymous platform, CPU) Number of tasks completed 269 Max tasks per day 573 Number of tasks today 0 Consecutive valid tasks 540 Average processing rate 58.15 GFLOPS Average turnaround time 0.59 days SETI@home v8 (anonymous platform, CPU) Number of tasks completed 92853 Max tasks per day 32391 Number of tasks today 64 Consecutive valid tasks 32359 Average processing rate 20.85 GFLOPS Average turnaround time 0.30 days My Ryzen7 1700X Host 8030022 Astropulse (anonymous platform, CPU) Number of tasks completed 62 Max tasks per day 156 Number of tasks today 0 Consecutive valid tasks 123 Average processing rate 75.89 GFLOPS Average turnaround time 1.75 days SETI@home v8 (anonymous platform, CPU) Number of tasks completed 52827 Max tasks per day 5165 Number of tasks today 153 Consecutive valid tasks 5133 Average processing rate 79.59 GFLOPS Average turnaround time 0.45 days I don't know of anyone running BOINC on Threadripper yet. As expected it is running into some of the same early issues of the Ryzen AM4 platform with CPU and memory overclocks being difficult to stabilize. But progress is being made and much faster compared to Ryzen since Ryzen cleared the path already on a lot of issues. After all TR is barely a month old and Ryzen is over 6 months old. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I have 2 dual E5-2670 machines, each with 16 cores running a total of 32 threads each, same as the Threadripper 1950x. Ignoring the cost factor, I would like to know if anyone is running the latter, and can offer some comparisons between the 2 setups, in terms of processing SETI WUs. Since I am running dual GTX 1080s on each machine, 3 WUs/GPU, I am running 25 threads of CPU WUs on each. Both machines are running 80K+ RAC, currently #31 and 33 on the computer list. No, they ain't cheap, that's fer sure. Some of the build cost totals for TR systems have been $3K on the bare-bones side to $5K on the decked out builds. A lot of the cost seems to be everyone building TR is putting in 128GB of memory. That adds to the cost fast. For a BOINC cruncher you certainly don't need that much memory. TR builds seem to be targeted for content creation. 4K video rendering, massive Linux compiles and some heavy NAS storage configurations from the threads I read. I guess you need lots of memory for that work. My Host 8030022 is currently fluctuating between #24 and #25 on the Top 100 hosts list. Two GTX 1070s and one GTX 1060 in the configuration. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Bill G Send message Joined: 1 Jun 01 Posts: 1282 Credit: 187,688,550 RAC: 182 |
I have one, almost assembled sitting on the kitchen floor. I bit the bullet and decided not to refit one of my computers that was in desperate need. I hope to apply some power tomorrow after I put in the optical drive from the older system. I do not expect to get super numbers from it as I will be only using 750Ti to start with. I have no money left for better GPUs. When I get it running on the 750 running I am going to switch in my 1050Ti's. Oh, only 32K of memory as this is a dedicated cruncher. SETI@home classic workunits 4,019 SETI@home classic CPU time 34,348 hours |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Congrats, Bill. Hope to see the new cruncher online soon. I have been lurking on this Threadripper thread at Overclock.net. Lots of good info on this board. asus-rog-zenith-extreme-x399-threadripper-overclocking-support [Edit] I think you meant 32GB of memory. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I took a look at your E5-2670. It is not all that impressive in regard to your APR or number of tasks per day validated. I don't know which Ryzen7 1800X systems you are comparing it too. Could you link the hosts of your comparison. I'm curious. I also wonder if those systems are stock or have been overclocked to what they are capable of. Comparing APR values is not very useful. It is even less useful for a hosts with a lot of CPU tasks rescheduled to a GPU. As it causes the APR values to be artificially inflated. At the moment your host is displaying SETI@home v8 (anonymous platform, CPU) Average processing rate 90.64 GFLOPS but it doesn't seem to be completing CPU tasks 4.5 times faster than my E5-2670 host. That is just to point out how useless APR is for comparison. A somewhat more accurate way to compare hosts is by run time of similar tasks. Arecibo tasks with a normal AR or GBT VLAR tasks are normally good to compare. It looks like your host is completing CPU tasks about 30% faster that the current #28 host 8295687. Which is also a Ryzen 7 1700X running the AVX app. Based on the run times I would guess they are running closer to stock clocks and you are a bit higher? Host 8295687 is completing similar CPU about 20% faster than my E5-2670 @ 3.0GHz. Given the Ryzen 7 1700X has a base clock of 3.4GHz, which is about 13% higher that my host, most of the performance gain looks to be from a higher clock. Given Ryzen is 5 years newer this is the source of my disappointment. The number of concurrent tasks can also have an effect on task run times. I have all 32 CPU threads running tasks since I'm not currently feeding any GPUs on my host. Once I put a pair of 1050 ti or1060 3GB GPUs in the system I expect the CPU run times to be slightly reduced. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Fair enough because some of the CPU tasks have been rescheduled. I noticed immediately that the AVX app runtimes on the Ryzen are 45-60 minutes faster than my FX processors running the same AVX app. The Ryzen runs GBT tasks faster than Arecibo tasks. Something to do with GBT tasks running faster on the AVX app than the Arecibo tasks. So I don't mind rescheduling twice a day the GBT tasks originally assigned as GPU to CPU. And the Ryzen system is running at a 500 Mhz deficit to the FX processors FYI. I have the 1700X currently running at 3.9Ghz all cores which is the normal single core boost frequency. I am not running Core Performance Boost. Just all cores clocked at 3.9 Ghz. [EDIT] Thought of something else that is probably pertinent. I have had my memory at 3200 Mhz with fast timings from almost day one. I got it up to 3333 Mhz and fast timings about a week ago. Ryzen responds best to faster memory because the memory clock speed is the same speed that the cross CCX bus communicates with the modules. That Host 829687 is interesting when I saw how consistent his GPU times are. It looks like he processes nothing but Arecibo tasks. How is he excluding GBT tasks without aborting them? There is no task granularity choice of telescope origin with the MB tasks. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Bill G Send message Joined: 1 Jun 01 Posts: 1282 Credit: 187,688,550 RAC: 182 |
Congrats, Bill. Hope to see the new cruncher online soon. I have been lurking on this Threadripper thread at Overclock.net. Lots of good info on this board. Indeed that is what I meant. Thanks for that link, I have been looking for a good support link without success. SETI@home classic workunits 4,019 SETI@home classic CPU time 34,348 hours |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13732 Credit: 208,696,464 RAC: 304 |
Given the Ryzen 7 1700X has a base clock of 3.4GHz, which is about 13% higher that my host, most of the performance gain looks to be from a higher clock. Given Ryzen is 5 years newer this is the source of my disappointment. You need to look at Ryzen in comparison to what it replaced- it is a massive improvement over the previous architecture. No, Ryzen doesn't match the current Intel CPUs for IPC (Instructions Per Clock), but it comes pretty close- which is a monumental improvement. With a year or 2 of data of real world usage, I personally expect Ryzen2 to come very close to matching the Intel CPUs at that time. Improved Ryzen hardware, and improved compiler options. As it stands, you can buy a high end Intel system, with it's much higher IPC, or you can buy a Ryzen system for half the price (or less) that is only slightly behind the Intel CPU in performance, or actually beats it for multi-threaded applications. The very latest release of Intel CPUs takes back the performance crown for even mutli-threaded workloads, but you will pay at least 2.5 times the price just for the CPU compared to a Ryzen based CPU. And that doesn't include the much higher pricing for the motherboard suitable to run those CPUs compared to the motherboard required for a Ryzen based system. Grant Darwin NT |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Given the Ryzen 7 1700X has a base clock of 3.4GHz, which is about 13% higher that my host, most of the performance gain looks to be from a higher clock. Given Ryzen is 5 years newer this is the source of my disappointment. I have the feeling OP is looking for a reason to justify upgrading one of their dual E5-2670 systems with a Threadripper. So I don't think comparing the performance of AMDs previous offing is relevant. For someone with an AMD system Ryzen/Threadripper are obvious choices to upgrade. Personally my plan is to upgrade my E5-2670 systems with v2 CPUs before I look to replace them. I don't see the massive cost difference some seem to be making up between the two. The only two components that would be different in building a new system are the MB & CPU for a system. If I compare two similar MBs with nearly the same features they have the same list and retail price of $160. GIGABYTE AORUS GA-AX370-Gaming K5 vs GIGABYTE AORUS GA-Z270X-Gaming K5 I wasn't able to find Z370 MBs on retail sites yet, but they have the same list price as their Z270 version. 6c/12t Ryzen 5 1600X 3.6GHz $249 vs 6c/12t Core i7 8700K 3.7GHz $359 4c/4t Ryzen 3 1300X 3.5GHz $129 vs 4c/4t Core i3 8350K 4.0GHz $168 Similar CPUs do have a gap but really isn't that much for the total cost of the system. However it makes for a reasonable cost to performance comparison. Getting back to the thread topic. BOINCStats says we have 39 hosts with Threadripper CPUs. I did find one 8322358 with a Threadripper 1950X. It was running stock but gives some point of reference. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
But since we are talking about the Threadripper ecosystem, we shouldn't be comparing the costs between Ryzen motherboards/Ryzen CPUs and Intel motherboards/i7 CPUs. We should be comparing the costs of Threadripper motherboards/Threadripper CPUs and dual Xeon motherboards/Xeon CPUs. The cost comparison between AMD Threadripper motherboards/dual Xeon motherboards is a wash. Same price range. The bigger difference comes with the CPUs. With dual Xeons, the core count is going to be higher than a single Threadripper. Question then comes to performance gains (if any) of the single TR vs the performance of the dual Xeons with lower clock speed. The few website reviews pitting the performance of the AMD 1950X for example against common dual Xeon had the TR system beating out the Silver range of Xeon. Only when the dual Xeon system had the Gold processors did its performance simply outgun the TR system. But it was at 5X the cost for the CPUs. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13732 Credit: 208,696,464 RAC: 304 |
Really, the comparison should be between Xeon CPUs and Epyc CPUs. Threadripper CPUs are just AMDs equivalent of Intel's Extreme Edition offerings. In the SPEC suite of tests, Single thread performance, the Epyc 7601@ 3.2 GHz roughly matches the E5-2690@ 3.8GHz (ahead on some, behind on others, about the same for most). Running 2 threads on the one core, the Epyc CPU does much better than the Xeon (roughly 30%). Running a (unrealistic) multithreaded benchmark, the Epyc system outperforms the Intel systems by over 40% For some performance figures that are more realistic, Source- Annandtech Grant Darwin NT |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I know it would have made more sense to compare Epyc vs Dual Xeon, but the OP asked about TR. I question why you would want to spend $1000 on a TR 1950X when you can buy the Epyc 7401P for $1100 which gets you 24cores/48 threads. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
ML1 Send message Joined: 25 Nov 01 Posts: 20267 Credit: 7,508,002 RAC: 20 |
I know it would have made more sense to compare Epyc vs Dual Xeon, but the OP asked about TR. I question why you would want to spend $1000 on a TR 1950X when you can buy the Epyc 7401P for $1100 which gets you 24cores/48 threads. All a question of which motherboard and which form factor you prefer?... Unfortunately, there also seems to be an obfuscating irrational game of Marketing and Hype to bamboozle the Marketing junkies! Regardless, it's very good to see some refreshingly new and very capable silicon rather than the very obvious expensive drip-feed of old we've seen for too long from the too long ('effective' / 'defective') monopolist... Happy fast crunchin'! Martin Note that this post contains puns of a personal opinion only... See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Bill G Send message Joined: 1 Jun 01 Posts: 1282 Credit: 187,688,550 RAC: 182 |
Keith Congrats, Bill. Hope to see the new cruncher online soon. . Well it is online.....believe it or not I took the HD out of the old system and put in in the TR and it booted right up. (there were a few problems after that) but it is at least running on 32 cores right now and doing SETI just fine. I had aborted all the WUs before I took the other computer off line about 2 weeks ago so when I started BOINC I got a full 300 WUs in one request I will not play with timing for a bit, I want it to settle in an get a feel for it. SETI@home classic workunits 4,019 SETI@home classic CPU time 34,348 hours |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Congrats on getting the TR up and running painlessly. I see you are running the Anonymous platform, I assume via the Lunatics installer. You will do yourself a BIG favor by running back through the installer and choosing the AVX app. Ryzen and TR REALLY love the AVX pathways. Much faster than the SSE3 app. Almost 50% faster.. And with that many cores available, you should be able to do some serious improvement to the system APR. Once you get comfortable with the stock speeds, try for getting the memory clocks up as that really improves the system speed. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Bill G Send message Joined: 1 Jun 01 Posts: 1282 Credit: 187,688,550 RAC: 182 |
Congrats on getting the TR up and running painlessly. I see you are running the Anonymous platform, I assume via the Lunatics installer. You will do yourself a BIG favor by running back through the installer and choosing the AVX app. Ryzen and TR REALLY love the AVX pathways. Much faster than the SSE3 app. Almost 50% faster.. And with that many cores available, you should be able to do some serious improvement to the system APR. Once you get comfortable with the stock speeds, try for getting the memory clocks up as that really improves the system speed. Thanks, I have done that. I remember that being said before but I did not remember. One thing I notice is that I do not think my HD is going to last long as it is constant !00% useage according to Task Manager....... SETI@home classic workunits 4,019 SETI@home classic CPU time 34,348 hours |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Is that from the BOINC checkpoints on so many concurrent tasks? If so, increase the timer to at least 120 seconds. I do notice quite a bit of disk activity on my Ryzen. But is on a NVMe SSD drive so it is very fast. I have the BOINC checkpoint at 120 seconds for a good compromise. I don't lose much on task restarts. That covers my Linux special app machine too since it completes task so quickly. If on the other hand, the disk activity is because of SATA contentions, there are different settings in the BIOS that reduce that. Look and search through the TR thread I linked for posts on SATA, SSD and NVMe BIOS settings. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Bill G Send message Joined: 1 Jun 01 Posts: 1282 Credit: 187,688,550 RAC: 182 |
It was being caused by BOINC. I changed the checkpoint timer to 120 and the disk has quieted down. Thanks again. SETI@home classic workunits 4,019 SETI@home classic CPU time 34,348 hours |
©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.