Gigabyte R113-C10-AA02 Topology
A major complaint I saw about the Lenovo ST45 V3 was its lackluster PCIe connectivity, and by comparison, the R113-C0 does much better.
Compared to servers with more PCIe lanes like EPYC 8000/9000 the little R113-C0 may look quite simple, but generally speaking, there is plenty of high-speed PCIe lanes to go around for such a small and inexpensive system.
The topology output supports the published block diagram, while simultaneously demonstrating how simple the PCIe configuration is compared to a larger server like the Dell R770 that was recently reviewed.
Gigabyte R113-C10-AA02 Management
One big complaint I saw in the coverage for the Lenovo ST45 V3 was the lack of a BMC or IPMI out-of-band management interface. Personally, I have always looked at the presence of a BMC as the dividing line between a workstation product and a server product, so I am very glad to find one here.
The ASPEED AST2600 drives the MegaRAC SP-X interface on this board, providing access to logging, sensor information, fan profiles, and a full iKVM with remote media.
I was able to use the iKVM to perform my OS installations for testing without issue, and the interface should be very familiar to anyone who has used a Supermicro or ASRock Rack IPMI interface recently.
Gigabyte R113-C10-AA02 Test Configuration
- Chassis: Gigabyte R113-C0-AA02
- Motherboard: Gigabyte MC13-LE0
- CPU: AMD EPYC 4464P (12-core, 24-thread)
- Memory: 2x V-Color 48GB DDR5-5600 ECC UDIMMs
- OS SSD: Samsung MZ-VL21T0B 1TB
- Data SSD: 4x 3.84TB Micron 5300 PRO
- Hypervisors: Windows Server 2025, VMware ESXi 8.0
Next, let us get to the performance.
Gigabyte R113-C10-AA02 Performance
We have actually used this server in a number of pieces recently. Something we have been discussing is just how far ahead the EPYC 4004 line is versus the Xeon E-2400 and Xeon 6300 series. Here is an example using SPEC CPU2017:
Taking those numbers and looking at the performance per core, we get something that looks a bit like this:
The AMD EPYC 4464P is very interesting since it is a 65W TDP part that with 12 cores is very close on a per-core performance basis versus the Intel Xeon E-2488. The advantage, of course, is 50% more cores. Taking a comparison to common dual socket installed base SKUs, the Xeon Gold 6252, the performance per core delta is substantial.
We just did the EPYC 4464P versus the Intel Xeon E-2488 in the Lenovo ST45 V3 review, and this is within +/- 2% of those results, so it feels a bit duplicative to go though those tests again. We have been using the EPYC 4004 for about a year now, and this performs excatly like we have seen other systems perform with the EPYC 4464P. At the same time, we tested not only 96GB in 2x 48GB and 192GB in 4x 48GB ECC DDR5 configurations and wanted to go through that quickly.
A Note on 1DPC versus 2DPC
When you populate both channels in the AMD EPYC 4004 series platform, the memory speed decreases from up to DDR5-5200 to DDR5-3600. That is why we see many use 2x 32GB DIMMs instead of 4x 16GB. Putting that into context, here is a Geekbench 5 run of the same system with four DIMMs (2DPC) as the baseline and two DIMMs (1DPC) as the comparison.
Geekbench 5 and Geekbench 6 behave differently, so here is the Geekbench 6 comparison as well:
2-10% is a reasonable range for most applications, but if there are extremely memory bandwidth-sensitive applications, then the deltas can be much higher.
Next, let us get to testing hypervisors and some of the common deployment scenarios.
Small typo:
“As mentioned previously, the Proxmox VE 8 worked fine in this configuration and we tested it with up to 196GB of memory.”
I’m guessing you either meant 96GB or 192GB. Not sure what kind of configuration would get you 196GB.
@James, thanks! I’ve fixed it.
@James 6x32GB sticks and a 1x 4GB or 2x 2GB would give you 196GB :)