- Server Model: G292-Z20-rev-100
- always the POSITIVE first: overall it’s an excellent machine but the user needs to know the details
- it has 1x BMC (LAN) and 2x slots for SPF fiber modules for example the SFP1G-SX-85
- so have SPF fiber modules ready for server + a SPF module capable switch
- it would have been nice to have had another 1000MBit or 10.000MBit copper based LAN port but well install an PCIe NIC then
- it can hold up to 8 GPUs and 2x PCIe cards
- 2x internal NVMe
- 2x SAS attached NVMes (the orange slots)
- was able to meassure 800MByte/s sequential on both with those KINGSTON SKC3000D2048G directly connected on mobo
- and the front panel connected KINGSTON DC 1500m (Mini) SAS connected NVMe
- the confusing naming of those drives is “U.2 PCIe NVMe” or (Mini)SAS attached NVMes, no kidding, search for em on ebay.
- a lot of BANG for the BUCK
- a bit too loud
- a bit too power hungry (uses 250W in idle with no GPUs installed and 350W with all cores on 100%)
- unfortunately those front drive panels while having nice status and activity leds, feel pretty cheaply “plastic” made
having owned a AMD workstation (so far pretty happy with it) for a while GIGABYTE is a Taiwanese brand that is associated with:
- high quality consumer hardware
- the software quality of their firmwares and drivers is “okayish” (with a bit of tinkering it usually can made to work)
so really wanted to give this thing a test drive.
what was kind of unexpected:
- the massive loudness of that thing for maximum cooling but with no proper throtteling of the FANs depending on temp?
- the (imho) pretty high power usage of 250W in idle
- for comaprison: the Intel Board based AQUADO servers 5 years old with loads of SATA drives only used 150W (100W less!)
- the 20 year old Dual Xeon SuperMicro servers used 300W (50W more)
- other gigabyte mobos:
Intel Xeon vs AMD EPYC vs Altra ARM CPU benchmark battle! 😀
PS: some Geekbench v6 comparisons:
- Xeon based HP HP Z4 G4 Workstation vs AMD x86 EPYC based GIGABYTE G292-Z20-00
- price wise the HP HP Z4 G4 Workstation wins
- performance wise they seem to be pretty “equal” but of course the 24x AMD EPYC cores outrun the only 6x Xeon cores in multi core tests
- WHY does everyone bench performance but NEVER the associated wattage? as if that was not important?
- luckily it can be said that the G292-Z20-rev-100 does not use much (350W) more when all CPUs Cores are at 100%
- ARM based ADLINK Ampere Altra Developer Platform (Ampere Altra Max @ 2.80 GHz 1 Processor, 96 Cores) vs GIGABYTE G292-Z20-00 (AMD EPYC 7402P @ 2.80 GHz 1 Processor, 24 Cores)
- ADLINK Ampere Altra ARM RISC Server Workstation Developer Platform tested by geerlingguy
- power wise: “70W to 90W running idle” so let’s say 80W in idle <- this is good!!! 🙂
- “With four sticks of (EEC) RAM” “was seeing about 200W of power draw under load“
- “That’s a 35-watt difference! A stick of DDR ECC RAM uses maybe 5 or 6 watts, so two should just add 10-15W max. Where’s the other 15W coming from?”
- src: https://www.jeffgeerling.com/blog/2023/ampere-altra-max-windows-gpus-and-gaming
- x86 ARM gaming? not yet: the poor gaming performance “CRYSIS” jeffgeerling is talking about is (probably) due to the fact the the ARM CPU has to “emulate” some how a 86x CPU with creates A LOT of overhead
- emulating ARM Android on x86 is the same story, LOADS OF OVERHEAD = VERY VERY SLOW X-D
- so for gaming this the GIGABYTE G292 Z20 Server is probably the WAY BETTER machine (have still to test that)
- as expected singlecore AMD x86 are (still) faster than ARM but the ARM has 96x cores using ~half the energy of a 24x Core AMD EPYC x86?
- thanks for testing geerlingguy fastest (yet) CPU wise tested system of geerlingguy:
- price wise the Ampere Altra is unfortunately pretty expensive, as there is no mass production yet?
- power wise: “70W to 90W running idle” so let’s say 80W in idle <- this is good!!! 🙂
- ADLINK Ampere Altra ARM RISC Server Workstation Developer Platform tested by geerlingguy
- Apple Mac Studio 10-core M1 Max @ 3.22 GHz 1 Processor, 10 Cores
- GIGABYTE G292-Z20-00 vs Mac Studio – M1 Max 10-core
- ADLINK Ampere Altra Developer Platform vs Mac Studio – M1 Max 10-core
some benchmarks: PLEASE TAKE THOSE GEEKBENCH RESULTS WITH MASSIVE HEAPS OF SALT AND ONLY AS ROUGH REFERENCE!
#prepare mkdir -p /home/user/software/benchmark cd /home/user/software/benchmark # install wget https://dwaves.de/scripts/benchmark_md5sum.sh.txt mv benchmark_md5sum.sh.txt benchmark_md5sum.sh # check what's in it before running it vim benchmark_md5sum.sh # if all good chmod +x benchmark_md5sum.sh # run time ./benchmark_md5sum.sh title: === dwaves.de v1.2 === dates: date-creation: long-time-ago date-last-test: 2024-12-24 what: simple sequential harddisk write and single core cpu benchmark requirements: apt install coreutils usage: in bash or sh run it like: time ./benchmark_md5sum.sh === installed cpu: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Address sizes: 43 bits physical, 48 bits virtual Byte Order: Little Endian CPU(s): 24 On-line CPU(s) list: 0-23 Vendor ID: AuthenticAMD Model name: AMD EPYC 7402P 24-Core Processor CPU family: 23 Model: 49 processor : 0 vendor_id : AuthenticAMD cpu family : 23 model : 49 model name : AMD EPYC 7402P 24-Core Processor stepping : 0 microcode : 0x830107a cpu MHz : 1500.000 cache size : 512 KB physical id : 0 === number of cores: 24 === generate 1GB test file 1048576+0 records in 1048576+0 records out 1073741824 bytes (1.1 GB, 1.0 GiB) copied, 2.28527 s, 470 MB/s === display filesize of generated testfile 1.0G md5sumTestFile === running md5sum on testfile (single core) cd573cfaace07e7949bc0c46028904ff md5sumTestFile real 0m4.724s
head to https://www.geekbench.com/download/ to get latest version
as it is closed source there is no possibility of checking whats in it 🙁
Geekbench is a proprietary and freemium cross-platform utility for benchmarking the central processing unit (CPU) and graphics processing unit (GPU) of computers, laptops, tablets, and phones.
https://www.primatelabs.com/legal/eula-v5.html
https://www.primatelabs.com/legal/eula-v6.html
“TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, IN NO EVENT SHALL PRIMATE LABS OR ITS LICENSORS BE LIABLE FOR”… (per default or only in case they get hacked???)… “VIRUSES, TROJAN HORSES, MALICIOUS SOFTWARE, FRAUDULENT ACTIVITIES, MALWARE OR THE LIKE WHICH MAY BE TRANSMITTED TO OR THROUGH OUR ONLINE SERVICES, THE APP OR BY ANY THIRD PARTY”
wget https://cdn.geekbench.com/Geekbench-5.5.1-Linux.tar.gz
tar fxvz Geekbench-5.5.1-Linux.tar.gz
time ./Geekbench-5.5.1-Linux/geekbench_x86_64
real 2m31.201s
results: https://browser.geekbench.com/v5/cpu/23309422
wget https://cdn.geekbench.com/Geekbench-6.3.0-Linux.tar.gz
tar fxvz Geekbench-6.3.0-Linux.tar.gz
time ./Geekbench-6.3.0-Linux/geekbench_x86_64
real 6m13.743s
results: https://browser.geekbench.com/v6/cpu/10237810
those Geekbench results are … strange to say the least… because more RAM “suddently” gives better single core performance? (THE SERVER WITH NEWER BIOS VERSION + LESS RAM WAS SLOWER THAN THE SAME SERVER WITH OLDER BIOS AND MORE RAM?)
how is that of any logic?
OF COURSE they now also have an AI Benchmark: https://www.geekbench.com/ai/
fun fact:
liked this article?
- only together we can create a truly free world
- plz support dwaves to keep it up & running!
- (yes the info on the internet is (mostly) free but beer is still not free (still have to work on that))
- really really hate advertisement
- contribute: whenever a solution was found, blog about it for others to find!
- talk about, recommend & link to this blog and articles
- thanks to all who contribute!
