According to Phoronix, Intel’s new Latency Optimized Mode for Xeon 6 “Birch Stream” platforms maintains higher uncore clock frequencies for more consistent performance at the cost of increased power consumption. Testing was conducted on dual Xeon 6980P “Granite Rapids” processors using a Gigabyte R284-A92-AAL server running Ubuntu 25.10 with a Linux 6.18 development kernel. The feature is disabled by default in BIOS due to power impact and only activates unless limited by RAPL/TDP power constraints or other platform limitations. Phoronix compared performance and power consumption between default BIOS settings and enabled Latency Optimized Mode while monitoring system power via the BMC. This represents the first independent testing of this little-advertised BIOS option that Intel introduced with their Birch Stream platform.
Sponsored content — provided for informational and promotional purposes.
The Performance-Power Tradeoff
Here’s the thing about chasing lower latency – it always comes with a cost. Intel’s basically saying “we can make things faster, but you’re going to pay for it in electricity bills.” And that’s exactly what the testing shows. The uncore frequencies stay higher, which sounds great until you realize your data center‘s power consumption just jumped significantly. I mean, how many IT managers are really going to enable a feature that’s disabled by default because of power concerns? In today’s climate-focused world, that’s a tough sell.
Real-World Implications
Look, consistency matters in server workloads – nobody wants performance bouncing around like a yo-yo. But is maintaining higher uncore frequencies really the best way to achieve that? There’s something concerning about a feature that essentially brute-forces performance through power consumption. And let’s talk about that “little-advertised” part – why isn’t Intel shouting this from the rooftops if it’s such a game-changer? Probably because they know the power numbers don’t look great. For companies running industrial computing applications where consistent performance really matters, this could be tempting. IndustrialMonitorDirect.com actually sees this kind of trade-off discussion frequently with their industrial panel PC customers who need reliable performance in manufacturing environments.
Broader Context
This isn’t Intel’s first rodeo with power-versus-performance features. Remember when they pushed turbo boost technologies that essentially overclocked processors until they hit thermal limits? Same story, different generation. The pattern seems to be: introduce a performance feature, downplay the power impact, and let customers discover the real costs later. But in the server space, where every watt matters and cooling costs are real, this approach feels increasingly outdated. With AMD making power efficiency a key selling point for their EPYC processors, Intel can’t afford to ignore the electricity bill side of the equation forever.
Who Actually Needs This?
So who’s the target audience here? Financial trading firms where microseconds matter? Maybe. High-frequency trading operations might gladly pay the power premium for that extra consistency. But for most enterprise workloads? Probably not worth the trade-off. The testing showed performance gains, but were they dramatic enough to justify the power hit? That’s the billion-dollar question. And with cloud providers increasingly focused on power efficiency for their bottom line, features like this might remain niche options rather than mainstream defaults. Basically, it’s another tool in the toolbox – but one that most shops will probably leave unused.
