The codenamed Intel Xe-HP GPU, HD Graphics Neo, was spotted in the Geekbench database along with its benchmarks. The chip based on Intel’s Gen 12 GPU architecture will target the Data Center and AI market segments and launch around next year. The benchmarks and listings were discovered by TUM_APISAK (via Videocardz) and showed some interesting results.
Intel Xe-HP architecture supports Neo GPUs for datacenter Specifications & benchmarks Details
Intel Xe-HP “Neo” GPUs were tested on Intel Whitley platforms with next-generation Ice Lake-SP processors. The processor itself has 2 cores with 24 cores and 48 threads, each with a total of 48 cores and 96 threads. The CPU base clock is reported to be 3.50 GHz along with 30 MB of L2 cache & 36 MB of L3 cache. The system also has 128 GB of DDR4 memory (ECC) and is tested on Ubuntu 20.04.1 operating system.
GPU specifications include 512 Compute Units or 4096 cores. Each Compute Unit on a Gen 12 GPU consists of 8 cores. The GPU is based on 1-Tile configuration, which is the lowest-level SKU of the Intel Xe-HP family. This product line has up to 2 and 4 Tiles providing a total of 1024 CU (8192 cores) and 2048 EU (16,384 cores). The Xe-HP “Neo” GPU is running at a clock speed of 1.15 GHz and has 6.36 GB of memory although it can be rounded to 6 or 8 GB but we don’t know for sure.
Neo GPUs based on Intel’s 12th generation Xe-HP architecture were also formally introduced during Architecture Day. This chip runs at a higher clock speed of 1.3GHz and achieves 10.5 TFLOPs of FP32 compute performance in the extended demo shown. The 2-layer GPU scored 21.1 TFLOP in the FP32 calculation while the 4-layer GPU achieved 42.0 TFLOP for the highest FP32 calculation.
Intel Vehicle HP GPUs
|GPU name||Intel Neo 1-Tile||Intel Neo 2-Tile||Intel Neo 4-Tile|
|GPU family||Intel Xe-HP||Intel Xe-HP||Intel Xe-HP|
|Process GPU||10nm SuperFin||10nm SuperFin||10nm SuperFin|
|Computer unit||512 EU||1024 EU||2048 EU|
|Number of cores||4096 cores||8192 cores||16,384 cores|
|Core clock||1300 MHz (Demo)||1300 MHz (Demo)||1300 MHz (Demo)|
|Calculator FP32||10.5 TFLOPs (Demo)||21.1 TFLOPs (Demo)||42.0 TFLOPs (Demo)|
|TDP||~ 150W||~ 300W||~ 400-500W|
Intel showed off a near-perfect ratio for its Xe-HP GPUs during the show. Also from what I can see here, Intel may be planning to name its Xe-HP GPU series “Neo” line for data centers.
In terms of performance, the Intel Xe-HP “Neo” GPU scored 25,475 in the Geekbench OpenCL benchmark. The graphics card’s performance is inconsistent even with one-generation old graphics cards, which shows that the GPU has yet to be optimized in terms of driver support as we are looking at an early model. clocked at just 1.15GHz lower than the ones even Intel showed running at 1.3GHz in their test lab.
For now, it is interesting to see the chips in testing and will be ready to ship in 2021. For gamers, Intel has the Xe-HPG lineup that we’ve covered here.
|GPU family||Intel Xe-LP (1st Gen)||Intel Xe-HPG (1st Gen)||Intel Xe-HP (1st Gen)||Intel Xe-HP (2nd Gen)||Intel Xe-HPC (1st Gen)|
|GPU segmentation||Input level (Integrated + Discrete)||Regular / advanced gaming (discrete)||Data Centers & Workstations||Data Centers & Workstations||High performance computer|
|GPU generation||Generation 12||Generation 12||Generation 12||Generation 13||Generation 12|
|Handle button||Intel 10nm SuperFin||Outsourcing workshop||Intel 10nm SuperFin||TBA||Intel 10nm SuperFin
|GPU products||Tiger Lake
DG1 / SG1 card
|GPU DG2||Arctic sound||Jupiter Sound||Old bridge|
|Specifications / Design||96 EU / 1 Tile / 1 GPU||512 EU / 1 Tile / 1 GPU||2048 EU / 4 cells per GPU||TBA||8192 EU / 16 cells per GPU|
|Memory subsystem||LPDDR4 / GDDR6||GDDR6||HBM2e||TBA||HBM2e|
|Launch||Year 2020||Year 2021||Year 2021||2022?||2021-2022|