Tachyum has upgraded the specs of its Prodigy HPC CPU by 50% which was made possible with the help of new EDA tools.
Tachyum's Prodigy CPU Now Better Than Ever (On Paper), Upgraded To 192 Cores
In its latest presser, Tachyum confirms that by utilizing the latest EDA tools, the company achieved a major spec bump for its Prodigy CPU. The Universal Processor, which was originally supposed to feature 128 cores, now gets 64 additional cores which is an increase of 50% over the original design and rounds up to a total of 192 cores.
The company states that with the new EDAA tools, the number of SERDES also went up from 64 to 96 per chip. This increased the die size from 500mm2 to 600mm2 but that's a marginal increase of 20%. The Prodigy CPU can also house additional cores to the Prodigy chip that would still fit within the reticle limit of 858mm2 but then the chip would become bandwidth limit. The chip already runs a 16-channel DDR5 memory interface with speeds of 7200 MT/s and beyond so additional cores will definitely need a much better memory standard.
Maybe on-die HBM3 memory may become the next logical step for future Prodigy designs. But there are also other enhancements that have come from the latest tools which aren't just limited to the number of cores. The L2 and L3 cache has also been increased from 128 MB to 192 MB and the other changes include:
- 50% Increase in core count from 128 to 192
- Increase of the chip L2/L3 cache from 128MB to 192MB
- Support of DDR5 7200 memory in addition to DDR5 6400
- More speed with 1 DIMM per channel
- Larger package accommodates additional 32 serial links and as many as 32 DIMMs connected to a single Prodigy chip
Prodigy is a Universal CPU in the sense that it can switch from traditional HPC workloads to AI/ML workloads which will be a key area that Tachyum will be targeting. The company already claims some really big numbers such as Prodigy CPUs delivering 4x the performance of today's highest performing x86 CPUs (for cloud workloads) and 3x the performance of the highest performing GPUs in HPC and 6x in AI workloads. The first chips are said to be on track for delivery by the end of 2023 so we will know soon enough if Tacyum can hit its performance goals.
WccftechContinue reading/original-link]
