Intel Xeon E5-2609 v4 Broadwell-EP Linux Benchmarks

[phoronix]

Phoronix: Intel Xeon E5-2609 v4 Broadwell-EP Linux Benchmarks

Recently I purchased a Xeon E5-2609 v4 Broadwell-EP processor as a $300 Xeon with eight physical cores but clocked at just 1.7GHz and without any Turbo Boost while the TDP is 85 Watts. Here are some benchmarks compared to other LGA-2011 v3 CPUs in my possession under Linux along with an AMD FX reference point too and followed by some Skylake Xeon benchmarks.

http://www.phoronix.com/vr.php?view=23503

[chuckula]

Some power consumption and performance per watt numbers would be extremely interesting there too since obviously this chip is obviously not being hot rodded in any way.

Overall that shows not just how impressive the IPC for a Broadwell core is, but how well the entire chip is architectured to take advantage of parallelized workloads. This particular Xeon doesn't even have hyperthreading so we aren't even getting full Broadwell core performance to boot. The shared L3 cache architecture makes for a true 8-core processor and AMD would do well in the future to not just copy Intel's core designs from 2011 like that are trying to do with Zen copying Sandy Bridge, but also go and copy Intel's ingenious core communication fabric. Right now AMD still seems to be stuck in 2009 copying Nehalem in that regard. Maybe by 2020 AMD can have a good competitor to this Broadwell part.

[Michael]

Some power consumption and performance per watt numbers would be extremely interesting there too since obviously this chip is obviously not being hot rodded in any way.

There are perf-per-watt and power consumption numbers.

[chuckula]

There are perf-per-watt and power consumption numbers.

Oh yeah, caught those at the end of the article. Good stuff for a server chip that runs parallelized workloads although a 1.8GHz processor is obviously not the best single-core chip.

[schmidtbag]

Overall, the E5-2609 v4 isn't very exciting at all. It's only about great if you want an eight-core, low-power system and don't care too much about raw performance but want something cool and quiet. I regret spending $300 on this CPU as for my purposes would have been better off spending a bit more to get a more compelling Broadwell-EP model.

Not sure why you expected any different... The results didn't surprise me at all.

[chuckula]

I think the conclusion is a bit too negative. You are running an 8 core chip that is intentionally not exceeding 1.7 GHz in clockspeed and that still manages to perform quite strongly in the exact types of benchmarks that you want a chip like this to handle well, especially in that postgres benchmark where it clearly has a strong performance/watt lead over Intel chips with the same number of cores, hyperthreading, and higher clocks that can churn through work faster. On top of that, a consumer workstation platform is clearly not what this Xeon is really designed for, but it still performed quite well.

As for the FX 8370, the main thing to remember is that if you doubled the clockspeed of this Xeon the FX 8370 would still have a 900MHz boost speed advantage. I think it's more of an indictment of the failed Piledriver architecture that an 8 core AMD part running north of 4 GHz has to get lucky to beat a vastly lower clocked and lower power Xeon in a few benchmarks. It should literally be winning all benchmarks over the Xeon even with lower IPC. Furthermore, I noticed that the performance/watt numbers left out the FX 8370 and I assure you that even in the benchmarks where a supposedly 8-core 4.3GHz chip "wins" over a 1.7GHz chip, the performance per watt numbers would paint a radically different picture,.

[schmidtbag]

As for the FX 8370, the main thing to remember is that if you doubled the clockspeed of this Xeon the FX 8370 would still have a 900MHz boost speed advantage. I think it's more of an indictment of the failed Piledriver architecture that an 8 core AMD part running north of 4 GHz has to get lucky to beat a vastly lower clocked and lower power Xeon in a few benchmarks. It should literally be winning all benchmarks over the Xeon even with lower IPC. Furthermore, I noticed that the performance/watt numbers left out the FX 8370 and I assure you that even in the benchmarks where a supposedly 8-core 4.3GHz chip "wins" over a 1.7GHz chip, the performance per watt numbers would paint a radically different picture,.

The only similarity between the FX-8370 and the E5-2609 v4 is they're both x86-64 compatible. The FX-8370 hardly even qualifies as an 8-core, when you consider they're paired together via "modules". They don't even share the same target audience. Both processors are so vastly different that you may as well be ranting about how an actual real-world 20-year-old piledriver is slower and less fuel efficient than a modern v8 pickup truck.

[c117152]

Not sure why you expected any different... The results didn't surprise me at all.

It fell exactly on hertz to hertz divided by node ~ number of transistors per inch. Any other result would have been a miracle \ disaster.

Fact is, Xeon\Cores's performance been tracking frequencies and node size within a 10% margin for the last 10 years. Some special processing like crypto and decoding\encoding multimedia may give the appearance of performance leaps. But that's merely the result of the limited test cover. Not the actual hardware getting better beyond the node size in the general purpose compute.

[Michael_S]

The only similarity between the FX-8370 and the E5-2609 v4 is they're both x86-64 compatible. The FX-8370 hardly even qualifies as an 8-core, when you consider they're paired together via "modules". They don't even share the same target audience. Both processors are so vastly different that you may as well be ranting about how an actual real-world 20-year-old piledriver is slower and less fuel efficient than a modern v8 pickup truck.

I think you take it a bit too far. If you're setting up a multi-threaded web server with a specific budget in mind you could use either one. And in that case, the E5-2609 v4 is clearly a better buy than the FX-8370. The Xeon gives you a small edge in raw performance and a huge edge in performance per watt.

So really you're just comparing a new, efficient work truck with a five year older work truck.

[Snaipersky]

One of the big selling points of E5 chips is RDIMM support. When you want to stack 128GB/+ of RAM on a single board, you go to the E5 or E7 range.
That chip would be wonderful for managing large databases and dozens of VM's