Microsoft has officially unveiled all the in-depth specs of the upcoming Xbox Series X, after months of rumors, speculations, and leaks. The console is definitely looking to pack a punch but at the same time isn’t quite as impressive as initially anticipated. Right off the bat, let’s have a look at the specs:
| CPU | 8x Zen 2 Cores @ 3.8 GHz (3.6 GHz w/SMT) |
| GPU | 12 TFLOPS, 52 CUs @ 1.825 GHz RDNA 2 GPU |
| Die Size | 360.45 mm2 |
| Process | 7nm Enhanced |
| Memory | 16 GB GDDR6 w/320mb |
| Memory Bandwidth | 10GB @ 560 GB/s, 6GB @ 336 GB/s |
| Internal Storage | 1 TB Custom NVME SSD |
| I/O Throughput | 2.4 GB/s (Raw), 4.8 GB/s (Compressed) |
| Expandable Storage | 1 TB Expansion Card |
| External Storage | USB 3.2 External, HDD Support |
| Optical Drive | 4K UHD Blu-Ray Drive |
| Performance Target | 4K @ 60 FPS, Up to 120 FPS |
Eight Zen 2 Cores @ 3.8 GHz (3.6 GHz with SMT)
This is basically a semi-custom Zen 2 APU with eight cores on par with the Ryzen 5 3600. While the core count is similar to the Ryzen 7 parts, the lack of Turbo Boost and reduced SMT speed will have a notable impact on gaming performance.
7nm AMD RDNA 2 GPU with 52 CUs (12 TFLOPs)
The CPU is a big step up from the preceding Jaguar core, but in light of recent rumors, the GPU is a bit disappointing. The Xbox Series X will feature a 52 Compute Unit GPU based on the enhanced 7nm (not EUV) RDNA 2 architecture. It will have a single-precision compute rating of 12 TFLOPs. In comparison, the present Navi based RX 5700 XT with 40 Compute Units offers nearly 10 (9.75) TFLOPs of performance.
Both the CU count and the compute performance ratings aren’t as impressive as expected. However, at the end of the day, it’s the first-party optimizations that have the final say in terms of performance. Whatever the case though, don’t expect 4K 120 FPS (1080p/1440p 120 FPS is likely) in any demanding modern title. In fact, you’ll be limited to near-60 FPS (at 4K) in the majority of games with the medium to high preset. Furthermore, you can forget about ray-tracing at 4K. With reasonable quality levels, it’ll be limited to 1080p at best.
Memory, Storage & I/O
As for the memory, we’re getting 16GB of GDDR6 @ 14Gbps. The part about the bandwidth is a bit misleading. The memory is actually partitioned into two blocks with the first 10GB over a 320-bit bus and a resulting memory bandwidth of 560GB/s. However, the remaining 6GB uses a reduced 192-bit bus resulting in a much lower 336GB/s bandwidth when the system has to use the entire 16GB memory. I’m not certain why the memory is partitioned this way, but it could be to separate the CPU and GPU memory pools. That’s just a guess though and till the console actually launches we won’t know for sure.
The memory is paired with a 1TB NVMe SSD that most likely runs using the PCIe 4.0 standard. As for I/O, you’re looking at 2.4GB/s raw throughput and as much as 4.8GB/s compressed data.
As for the remaining terms on the Xbox One X glossary, they mostly appear to be a marketing gimmick. The important ones are Mesh Shading, Ray-Tracing, GPU Work Creation, Variable Rate Shading and Variable Refresh Rate (FreeSync). To get a good idea about what they mean and how core DX12 technologies work, have a look at this post:
What is the Difference Between DirectX 11 vs DirectX 12: In-depth Analysis
The interesting part about the blog post was with respect to DirectML. It’s basically a machine learning API that uses neural networks to improve performance. I’m guessing something similar to NVIDIA’s DLSS. Using mixed-precision compute, you’re looking at 24 TFLOPs of FP16 compute performance and 97 TOPs of 4-bit INT performance on the new Xbox.
I wouldn’t pay much heed to the other big numbers that Microsoft is throwing at the press. I’m not saying that they’re invalid, it just sounds like a big PR stunt, especially the DigitalFoundry interview. The core numbers are: a 12 TFLOP RDNA 2 GPU with 52 CUs. That won’t be beating the GeForce RTX 2080 Ti anytime soon. In fact, it’ll perform closer to the GeForce RTX 2080 Super, and considering that NVIDIA’s next-gen Ampere GPUs will come with at least a 50% performance uplift over the existing Turing parts, that’s not very impressive.
