Dell PowerEdge R670 Rack Server Review

The Dell PowerEdge R670 Rack Server is part of the 17^th generation line of servers. It is a 1U dual socket system hosting Intel Xeon 6 processors. Plus, some high-density storage with up to 20x EDSFF drive bays in certain configurations, loads of memory and optional liquid cooling. The tradition of an overwhelming number of options continues with this PowerEdge R670. What’s it good for you may be asking? Pretty much everything. Dell says it’s good for hyperscale workloads, scale-out database, virtualization, cloud-native applications, and most likely a lot more.

Let’s start with the front of the chassis. There is the option of a security bezel. Also, the option of a front or rear I/O configuration but that’s just the beginning. There is only one front I/O configuration for storage, and it comes with 8x EDSFF E3.S NVMe drive bays. From there we return to the rear I/O configuration and there are several storage configurations using EDSFF E3.S NVMe Drives, except for that first one. A no backplane configuration. Then, there’s an 8-bay EDSFF, 16-bay EDSFF, 20–bay EDSFF and option to install 2x more EDSFF in the rear but we’ll get to that. Next, chassis configurations with 2.5-inch drive bays including an 8-bay for SAS/SATA/NVMe drive types, another 8-bay version with a universal backplane where you can load whatever drive type in any of the slots. Lastly a 10-bay version for SAS and SATA drive types.

There is a small control panel on the left with a USB port and a mini-display port. The KVM module is optional, meaning no USB or Mini-display ports. You also have another option that incorporates Quick Sync 2 for at-chassis management of the system using a smart phone or tablet. On the right-hand side, you get a system health LED, system ID button LED, power button, Type-C USB port, and the host status LED, which is that little wrench icon. That would be consistent with the front I/O configuration as well.

You still get the same control panel options as mentioned but you also get a Dedicated iDRAC Ethernet port and a serial COM port. Adjacent to that, you have the option to install a Boot Optimized Storage Device or BOSS. Next, a storage cage with 8x EDSFF E3.S drive bays, then the Primary OCP NIC with the secondary OCP card slot above. The primary OCP card slot shares a NIC port with iDRAC but both are OCP 3.0 compatible. 

The other 8-bay configuration looks quite similar in front but the PCIe slots and OCP card slot have cover panels. The other drive configurations with 16 and 20-bays have a 5x drive cage layout with 4x cages occupied on the 16x bay and all 5x drive cages used for the 20x drive bay setup. Lastly, the no-backplane setup with no drives in the front drive bays and all 5 Drive cage locations outfitted with a perforated cover for air flow. Still gotta keep it cool!

The back of the front I/O configuration still houses dual Power Supply Units and a few PCIe slots at the back of the chassis. There are several options for the PSUs depending on the configuration. Still looking at the front I/O version, it has a small I/O panel with VGA port, dual USB 3.0 ports and a dedicated iDRAC port, then the System identification LED. In this case the OCP card slots are covered as the OCP connectors have been moved to the front of the chassis. 

That dedicated iDRAC port on the back of the chassis enables remote management of the system. With this server line, Dell has introduced iDRAC 10. Additional management utilities supported on this platform include some new applications including NativeEdge Endpoint, and NativeEdge Orchestrator, which run a special secure operating system. Those last two are for edge applications but also provide a compute foundation for distributed data centers. Dell OpenManage is also supported plus a Command Line Interface. Open manage also provides integration with Microsoft Windows Admin Center, VMware vCenter plus connections with third-party management utilities. 

The back of the other configurations, that do not have front I/O, include that same I/O panel on the back but those OCP 3.0 mezzanine card slots are now active. One of them is specifically for Dell’s Boot Optimized Storage Subsystem, AKA a BOSS, or an internal M.2 interposer board for internal system boot. Is the interposer board the same as the Dell BOSS? No. It does not have access from the rear of the system to swap drives easily but does still have 2x M.2 drives. Also, there is no RAID supported like on the BOSS, which is a small controller that offers RAID 0 or 1. There is a small cover panel next to the right-hand side PSU, where optional liquid cooling tubes would be located providing for both an input tube and output tube.     

Once we remove the cover, you can see 2x risers supporting expansion devices standard PCIe and CXL 2.0. Again, a bunch of riser options with either 3x x16 or 2 x8 Gen 5 slots and other combinations between.

Beneath those two risers are the BOSS module slot and the OCP NIC card slot. Then the motherboard with associated DIMM slots, a row of fans and then the backplane for the upfront storage devices. If this was the front I/O model you would see a front riser on the front left and BOSS module slot in front on the right separated by the storage backplane in the middle. With the non-front I/O version, there is the option to install a front PERC storage controller. As mentioned before, the PCIe 5.0 slots also support CXL 2.0 devices.  

The CXL protocol runs across the standard PCIe physical layer. It can support both standard PCIe devices like NICs, Storage Controllers, and GPUs, as well as CXL devices on the same link. There are many different forms of CXL devices including the more recognizable options like NICs, Storage controllers, and GPUs. One of the unique things you can do with CXL 2.0 is install additional memory through a CXL Memory Expansion Add-In-Card. It fits into one of the PCIe 5.0 expansion slots which is also CXL 2.0 compatible and uses the PCIe 5.0 interface. 

This additional memory provides advantages for memory pooling and resource sharing, plus improved ROI for data center and high-performance computing environments. The memory pools can be accessed by multiple hosts, significantly increasing overall memory utilization. Form factors for these memory expansion devices include a PCIe form factor as well as EDSFF from factors. With that last one, you can scale memory on the fly without powering down. 

The module is installed in a CEM slot with a x16 link width outfitted with 4x DIMM modules for an additional 384GB of memory. By the way, in this respect CEM is referring to a Card Electromechanical specification. In this case the PCIe slot and card interface. This memory can be allocated a number of ways but in a more basic context either Conventional Memory Mode, aka General Purpose Memory Mode, or, Special Purpose Memory Mode. With Conventional Memory Mode the CXL memory is directly mapped and added to the system’s physical memory and is visible in UEFI. The CXL memory appears with the standard DDR5 memory as a unified resource. With Special Purpose Memory, the CXL memory can be allocated separately from physical memory to support specific applications.

The Dell CXL memory expansion card looks like a standard PCIe card but is outfitted with up to 384 GB of memory using 4x DRAM modules. Select PowerEdge systems, like this one, can be managed using iDRAC with the Lifecycle controller. Per CPU you can install up to 4x of these CXL Add-in-cards but with the R670 we only have a maximum of 3x x16 slots. Therefore, at maximum only 3x of these cards can be installed and then you would be depending on that lone OCP 3.0 mezzanine card slot for additional network communications. 

GPUs would also be considered a CXL device. This system can be outfitted with up to 3x 75W GPUs. That power limit would include the NVIDIA L4 24GB, but also the NVIDIA Tesla T4 and the NVIDIA A2. No auxiliary power is required for these cards as the PCIe slot provides up to 75W of power. 

The system is powered by Dual Intel Xeon Scalable processors from the 6500 or 6700-series with either P-cores or E-cores at least at this point in January of 2025 or thereabouts. P-core and E-cores do not mix so you have to choose one or the other. We will mention the following comparisons are with top-of-the-line CPUs in each category. With E-cores you do get more of them with up to 144 cores compared to CPUs with P-cores that can provide up to 86 cores. However, CPUs with P-cores have Intel Hyperthreading technology for up to 172 virtual threads. E-cores on the other hand do not have Intel’s hyperthreading technology but still up to 144 physical cores and better energy efficiency. If you want more information on Intel Xeon 6 processors, there’s a short video and you can see here. 

In total, there are 32x DIMM slots divided between the 2x processors for 16x DIMM slots per CPU. The processors support an 8-channel memory architecture so up to 2x DIMMs per Channel at capacity. At full capacity, which usually never happens anyway… the system can be outfitted with up to 8TB of DDR5 Registered DIMMs.

There are 4x dual fan modules installed in this platform, either high-performance silver or standard how-swapable cooling fans. Those fans suck in air through the front panel and blow over the CPUs, memory modules, and finally out the back of the chassis.

Hardly a full account of all of the various options on the Dell PowerEdge R670 server but it will have to suffice. There are a ton of options on this system and those CXL memory expansion devices bring another layer of possibilities to system performance. If you want to learn more about this system, or any other systems, contact us today! We are 99.9% sure we have what you need, all at the best price.