An EPYC Battle. AMD vs. Intel in the Server Space

So Anandtech got to play with Epyc and compare it to Broadwell and Skylake Xeons.

"With the exception of database software and vectorizable HPC code, AMD's EPYC 7601 ($4200) offers slightly less or slightly better performance than Intel's Xeon 8176 ($8000+). However the real competitor is probably the Xeon 8160, which has 4 (-14%) fewer cores and slightly lower turbo clocks (-100 or -200 MHz). We expect that this CPU will likely offer 15% lower performance, and yet it still costs about $500 more ($4700) than the best EPYC. Of course, everything will depend on the final server system price, but it looks like AMD's new EPYC will put some serious performance-per-dollar pressure on the Intel line."

"Meanwhile, although we have yet to test it, AMD's single socket offering looks even more attractive. We estimate that a single EPYC 7551P would indeed outperform many of the dual Silver Xeon solutions. Overall the single-socket EPYC gives you about 8 cores more at similar clockspeeds, while AMD loses the disadvantage of cross socket communication, and the server board gets simpler and thus cheaper. For price conscious server buyers, this is an excellent option."

Things are about to get really interesting in the server space me thinks...


Was going to make a thread until I saw this one, I'll drop the PcPer link here:

Holy **** there are a LOT of skews... No 32 core option like we saw in all the leaks (coming later maybe? fingers crossed). Also, damn the prices for the top end skews are wayyy higher than the old ones. Intel is segmenting this out to a whole new level compared to what they had done with prior generations.

1 Like

Hmm wow. I really don't have time to read it right now. Will later. But not 32 cores? I thought that was def gonna be a thing. Odd. Leaves Intel vulnerable. Yeah their cores are faster for sure but if you have more ya can still win.

It's interesting to see the power consumption too. In some loads with less cores and similar/less performance Intel in pulling a lot more power.

Seems they are really betting hard on AVX512 performance and the ability to scale to more sockets but I honestly don't know how appealing, especially the scaling, that is. Most systems are two sockets.

I think it said in the review that they attributed this to the FPU being twice as wide as AMD's (128 bit vs 64 bit). Does not bode well at this stage.

This does NOT look good for intel.

Multi-core performance comparisons depend on which skews & test your looking at, but in general at the same price it looks to favor AMD in general.

Single-core performance comparisons depend on which skews your looking at & test, but in general probably favor Intel due to ipc and turbo advantages.

Platform comparisons depend on what factors your looking at, but I'd say AMD's pcie lane count per cpu trumps Intel's faster onboard LAN so I'd favor AMD.

1 Like

Anyone else notice the irony of intels numbering scheme? Xeon 41xx, 61xx, 81xx(yeah I'm conveniently ignoring the others)...

1 Like

Intel saw the reviews then let a 5 year old make their slide deck


I think we'll end up with what I have been laughing at for a couple years now. AMD adopted a RISC mindset with the K5 being an actual RISC processor. They kind of kept the same design from then on, just moved to X86 as the instruction set, which always interested me. Whats better, they found a way to get around the terminal life of getting to the smallest nm-size by having a multi-CCX plane CPU. They tried the same thing with the FX cores but they didn't have the design quite finished. Though, now, I think we'll see the market completely flip. Its going to be hilarious.

What was once a company trying to feed off the back of intel in the 80's is now kicking them down. What a weird world to live in.

1 Like
1 Like

I wouldn't think AVX512 is really important, you get significantly better performance from a GPU on that level. We do compute with GPU and have no intention of leaving it, because at server-scale, when you need processing power for a specific instruction set, you need a lot of it. A CPU just can't rival the scalability of a GPU.

More sockets is going to be the only real winner there.

That said, I haven't really seen many systems in the real world that use more than 2 sockets. All our compute nodes are 2 socket because we don't need more. It puts the cluster in a situation where a node failure will bring down a lot of VMs. We don't like that.

I like that EPYC 32 cores can all-core boost to 2.7GHz. That's impressive. Do we know if that 3.8GHz on the 8180 is single-core turbo or all-core? That, I think, is going to be a major determining factor.

1 Like

That is most definitely a single core. If you think it pulls a lot of power as is, you can't even imagine how much it would draw if you ran all cores at 3.8ghz...

1 Like

That's true. I wish they had more detailed info. Intel seems to be trying to hide a lot lately.

With Broadwell-EP they released a turbo chart showing what frequency the processors would run at depending on how many cores were being utilized, fingers crossed they do that again.

Found this in the anandtech article:

looks like AVX512 hamstrings the CPU.

AnandTech has a blog going through the entire intel press livestream that can be found here:

Bunch of marketing stuff, nothing new but if anyone wants it's there