DLVR at Arrow Lake is a real place. Its purpose and existence are clear.
DLVR at Arrow Lake is a real place. Its purpose and existence are clear.
Based on my view, DLVR enables using a typical input voltage and fine-tuning it to suit core needs. Cores that process faster receive more voltage, while slower ones get less—contrary to earlier designs that gave all cores identical power. In theory, this should enhance energy use. However, those who watched Der8auer’s video understand it leads to significant efficiency drops, as excess power is wasted. At baseline, the system draws around 240 watts in Cinebench R23. This highlights how impressive Arrow Lake’s power efficiency truly is. The 14900KS could exceed 390 watts, and it runs at higher frequencies, though thermal throttling brings P cores down to about 5.5 GHz and E cores to 4.4 GHz, making only a 100 MHz difference versus 285K. Direct die cooling helped keep it steady at 5.8 and 4.4 GHz, while 70-80 MHz on P and 60 MHz on E cores reduced power consumption. If it ran cooler, it might draw even less. This demonstrates TSMC’s node superiority over Intel’s older 10nm (Intel 7). The improved efficiency likely contributed to the 50% cache increase and better gaming results. Of course, architectural tweaks also play a role—Arrow Lake outperforms Raptor Lake in raw latency and single-threaded tasks despite fewer threads. Lion Cove’s enhancements, such as an 8-bit decoder, boost performance more for multi-threaded workloads. I think Raptor Cove still edges out in latency and lighter single-thread scenarios. Pat Gelsinger’s push for rapid growth and foundry investment makes sense. Intel’s 18A is expected to succeed, with 20A already targeting 3nm. They’re also introducing GAA transistors and Power Via. At 1.5V input with a 1.1V core, 220A current, power loss could reach 88 watts—an enormous figure. For gaming, enabling Power Gate Mode cut draw to 180 watts in Cinebench, a 60% reduction. Even with this improvement, the real question remains: why does DLVR exist and why did Intel disable bypass options? The ring bus appears slower on Arrow Lake, contributing to higher latency and memory delays. New CUDIMM memory helps lower latency and boost performance. With better thermal control and voltage regulation, P cores can overclock safely for gaming or E cores for multi-threaded tasks. Arrow Lake is surprisingly efficient when optimized; its cost should drop accordingly.
The schedule of participants on this forum isn't clear. I'm here to help move the discussion forward.