WYP
News Guru
This board can handle 544W of power output!
Read more about ASUS' Prime X299 Edition 30 motherboard.
Read more about ASUS' Prime X299 Edition 30 motherboard.
Last edited:
ASUS Celebrates 30 Years of Motherboard Manufacturing with ASUS Prime X299 Edition 30
- Thread starter WYP
- Start date
looz
New member
HEDT is such a pointless segment. Workloads which need such processing power are better handled by multi-socket servers with better power efficiency and possibly much more processing power, not to mention supporting features for RAID and ECC memory.
Anything benefiting for higher clock speeds is better handled by desktop segment where power draw stays reasonable. There's an added benefit of often better memory speeds since dual channel memory controllers generally clock higher.
Anything benefiting for higher clock speeds is better handled by desktop segment where power draw stays reasonable. There's an added benefit of often better memory speeds since dual channel memory controllers generally clock higher.
Agreed Looz. HEDT was Intel's attempt to target enthusiasts running multi-GPU systems. They did it by limiting PCIE lanes on their 'standard' desktop parts, thereby creating a new market segment they figured enthusiasts would buy into, with the higher lane parts costing more money. They failed rather miserably, and combine that with a well-timed kick to their gonads in the form of Ryzen, Intel's strategy fell flat on it's face.
I don't think it's pointless, while some workloads that are nearly completely serial do just need higher clocks, and some that are completely parallel do just benefit from huge numbers of cores, many workloads need you to find a sweet spot with where current technology is at between the balance of cores vs clock speeds, and whether it's more worthwhile spending the energy on running many cores at a low clock speed or fewer cores at higher speeds, worth considering Amdahls law here:
For even a 75% parallel workload, you'll likely harm efficiency by going beyond 16 cores, as the best speedup you could possibly ever get is x4, even with an infinite number of cores. A 90% parallel workload raises this to a possible x10 speed up at infinity(Say 256 cores), but x8 at 32 cores, so you're just wasting energy by throwing more than say 128 at it.
Many workloads, like financial/spreadsheets, or say code compilation, sits somewhere between 75% and 90% parallel, so for many workloads getting the highest clocked ~12-16 cores is actually the best route you can take for maximum speed when you also take into account that the clock speed possible generally regresses past a point with core count. This is why the financial sector is dominated with HEDT builds.
For even a 75% parallel workload, you'll likely harm efficiency by going beyond 16 cores, as the best speedup you could possibly ever get is x4, even with an infinite number of cores. A 90% parallel workload raises this to a possible x10 speed up at infinity(Say 256 cores), but x8 at 32 cores, so you're just wasting energy by throwing more than say 128 at it.
Many workloads, like financial/spreadsheets, or say code compilation, sits somewhere between 75% and 90% parallel, so for many workloads getting the highest clocked ~12-16 cores is actually the best route you can take for maximum speed when you also take into account that the clock speed possible generally regresses past a point with core count. This is why the financial sector is dominated with HEDT builds.
Last edited:
looz
New member
I mean there's barely any extra energy being used since the cores are mostly sitting idle at that point. But provisioning is way off whack then, true.
In situations where this matters overclocked HEDT's aren't feasible due to reliability concerns. In addition such workloads can often be accelerated with specific hardware.
I don't have numbers to back this claim up, but it seems that the majority of HEDT is sold to people who simply want "the best".
In situations where this matters overclocked HEDT's aren't feasible due to reliability concerns. In addition such workloads can often be accelerated with specific hardware.
I don't have numbers to back this claim up, but it seems that the majority of HEDT is sold to people who simply want "the best".
NeverBackDown
AMD Enthusiast
Yeah looz. Most people buy it because it's the best rather than actually needing it. More money than sense.
There's no doubt there are a lot of people who buy it for that reason, doesn't change the fact there's a lot of people who buy it because nothing would do the job better, hence why it's so widely used in industry. Doesn't matter how idle the cores are, their cache is one of the most energy consuming parts, and if you're popping data in and out of them, that's getting used regardless of how much computation you do on that data, you have to constantly refresh it. And ofc you still spent money on those cores.
Last edited:
