FX8350 low performance on Linux need help

[Phantom329]

Hi LTT,

I have a Linux gaming server that is using my old desktop hardware. I recently got a FX8350 because its about the best thing that works in my mobo. I tried running a vanilla Minecraft server today and the performance was awful, it was over 100% usage of CPU with 1 person on the server. I don't know Linux that well and I'm not sure how to diagnose the issue. I don't think I updated my BIOS for the new chip and pcpp said it might need it. I just checked on pcpp again and now it says incompatible. What utilities should I use to monitor performance? I don't have access to the GUI of the server right now, only cmd.

PCPartPicker Part List

Type	Item	Price
CPU	AMD FX-8350 4 GHz 8-Core Processor	$175.99 @ Best Buy
Motherboard	ASRock 970 Extreme3 ATX AM3+ Motherboard
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Case	Rosewill CHALLENGER ATX Mid Tower Case	$60.99 @ Amazon
Power Supply	Corsair Builder 500 W 80+ Bronze Certified ATX Power Supply
	Prices include shipping, taxes, rebates, and discounts
	Total	$345.66
	Generated by PCPartPicker 2019-08-03 18:23 EDT-0400

 

[xhackforeverxdx]

4 minutes ago, Phantom329 said:

Hi LTT,

I have a Linux gaming server that is using my old desktop hardware. I recently got a FX8350 because its about the best thing that works in my mobo. I tried running a vanilla Minecraft server today and the performance was awful, it was over 100% usage of CPU with 1 person on the server. I don't know Linux that well and I'm not sure how to diagnose the issue. I don't think I updated my BIOS for the new chip and pcpp said it might need it. I just checked on pcpp again and now it says incompatible. What utilities should I use to monitor performance? I don't have access to the GUI of the server right now, only cmd.

PCPartPicker Part List

Type	Item	Price
CPU	AMD FX-8350 4 GHz 8-Core Processor	$175.99 @ Best Buy
Motherboard	ASRock 970 Extreme3 ATX AM3+ Motherboard
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Case	Rosewill CHALLENGER ATX Mid Tower Case	$60.99 @ Amazon
Power Supply	Corsair Builder 500 W 80+ Bronze Certified ATX Power Supply
	Prices include shipping, taxes, rebates, and discounts
	Total	$345.66
	Generated by PCPartPicker 2019-08-03 18:23 EDT-0400

 

Pretty sure a basic minecraft server setup only can use one core (back in my mincraft days atleast 4 years ago) and well the single core performance of the fx-8350 is ehm ye hahahah

[flibberdipper]

Just now, xhackforeverxdx said:

Pretty sure a basic minecraft server setup only can use one core (back in my mincraft days atleast 4 years ago) and well the single core performance of the fx-8350 is ehm ye hahahah

It'll use both cores of the G3258 I have in my clusterfuck server so at the very least it uses two.

[xhackforeverxdx]

Just now, flibberdipper said:

It'll use both cores of the G3258 I have in my clusterfuck server so at the very least it uses two.

Well 2 is better then one but still using 2 fx cores is like using 1 normal cpu core so idk ^^ 

[KarathKasun]

The server itself is single threaded.  The other cores being used are just doing Java VM housekeeping.

 

FX has the single core performance of Core 2 Duo/Quad, and your motherboard is likely throttling the chip to hell because FX 8350 eats tons of power and requires a relatively high end board to not throttle.  You can reduce clocks and undervolt to bypass this, and maybe increase performance under heavy load.

 

Also, in Linux with an 8 core CPU, 800% is maximum CPU load.  100% just means it is using a single core.

 

And in Minecraft servers, terrain generation will make the server run horribly for the first couple of minutes or hours as you are using it.  Once most of the terrain in a decent sized area is generated, it should settle down considerably.

[t4ils]

A Minecraft server is heavily single-threaded; I believe only the networking threads are multi-threaded. Though, if you genuinely think it's an issue with Linux, try it on Windows. I have a G3258 (2 cores, no HT) OC'd to 4.2GHz on my Arch Linux server and I'm able to run a vanilla server and a SkyFactory 4 server at the same time, each one with about 1-3 players each.

[5x5]

32 minutes ago, KarathKasun said:

The server itself is single threaded.  The other cores being used are just doing Java VM housekeeping.

 

FX has the single core performance of Core 2 Duo/Quad, and your motherboard is likely throttling the chip to hell because FX 8350 eats tons of power and requires a relatively high end board to not throttle.  You can reduce clocks and undervolt to bypass this, and maybe increase performance under heavy load.

 

Also, in Linux with an 8 core CPU, 800% is maximum CPU load.  100% just means it is using a single core.

 

And in Minecraft servers, terrain generation will make the server run horribly for the first couple of minutes or hours as you are using it.  Once most of the terrain in a decent sized area is generated, it should settle down considerably.

What are you talking about? Throttling only occurs when you overclock past the boards limits. At stock it has no issue driving the FX8 chips. In addition, C2Q? Really, you're obviously clueless as to FX but let me.wxplain it. FX*3** models are not the original Bulldozer. Vishera fixed a lot of the basic issues with FX and IPC was on the level of Nahalem/Westmere. Not Sandy or Ivy levels by any means but Core2? Please. 

 

 

The issue here is software. I've run Minecraft servers on Athlon II dual cores. For anyone who seriously believes the problem is the CPU, I'd suggest reading more and posting less.

[5x5]

49 minutes ago, Phantom329 said:

Hi LTT,

I have a Linux gaming server that is using my old desktop hardware. I recently got a FX8350 because its about the best thing that works in my mobo. I tried running a vanilla Minecraft server today and the performance was awful, it was over 100% usage of CPU with 1 person on the server. I don't know Linux that well and I'm not sure how to diagnose the issue. I don't think I updated my BIOS for the new chip and pcpp said it might need it. I just checked on pcpp again and now it says incompatible. What utilities should I use to monitor performance? I don't have access to the GUI of the server right now, only cmd.

PCPartPicker Part List

Type	Item	Price
CPU	AMD FX-8350 4 GHz 8-Core Processor	$175.99 @ Best Buy
Motherboard	ASRock 970 Extreme3 ATX AM3+ Motherboard
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Memory	G.Skill Ripjaws Series 4 GB (1 x 4 GB) DDR3-1333 Memory	$27.17 @ Amazon
Case	Rosewill CHALLENGER ATX Mid Tower Case	$60.99 @ Amazon
Power Supply	Corsair Builder 500 W 80+ Bronze Certified ATX Power Supply
	Prices include shipping, taxes, rebates, and discounts
	Total	$345.66
	Generated by PCPartPicker 2019-08-03 18:23 EDT-0400

 

Run under Windows and update the firmware and software. Linux will require more manual testing Ning whereas Windows will automate a lot of it for Vishera.

[Phantom329]

I did a bit of research and Minecraft is multithreaded but its not an even distribution. It has the main loop and puts other loops (like chunk loading or mobs) on different threads. Since I'm seeing around 100 to 130% usage could there be an issue with the server not putting loops on different threads? I would rather not switch to Windows.

[KarathKasun]

11 hours ago, 5x5 said:

What are you talking about? Throttling only occurs when you overclock past the boards limits. At stock it has no issue driving the FX8 chips. In addition, C2Q? Really, you're obviously clueless as to FX but let me.wxplain it. FX*3** models are not the original Bulldozer. Vishera fixed a lot of the basic issues with FX and IPC was on the level of Nahalem/Westmere. Not Sandy or Ivy levels by any means but Core2? Please. 

 

 

The issue here is software. I've run Minecraft servers on Athlon II dual cores. For anyone who seriously believes the problem is the CPU, I'd suggest reading more and posting less.

Guy, Man, I have 4 FX systems that are mothballed.  Per core performance is Core 2 level.  A dual Core 2 Quad workstation with 8 cores @ 3.4ghz is about the same speed.  FX is 5% slower per core than Phenom II at the same clocks, even the 8350.

 

Also.  That board throttles at stock with the FX 8000 series.  I had one.  I can make it throttle with an FX 4100.

[5x5]

13 minutes ago, KarathKasun said:

Guy, Man, I have 4 FX systems that are mothballed.  Per core performance is Core 2 level.  A dual Core 2 Quad workstation with 8 cores @ 3.4ghz is about the same speed.  FX is 5% slower per core than Phenom II at the same clocks, even the 8350.

 

Also.  That board throttles at stock with the FX 8000 series.  I had one.  I can make it throttle with an FX 4100.

 Mind posting any proof cause you're making a lot of stuff up atm. Core 2 IPC is much lower than both Phenom II and FX. The direct competitor to Phenom II was Nahalem and the core i* first gen processors. Seriously, you're young, I get that, but read more and post less otherwise you look silly

[KarathKasun]

1 hour ago, 5x5 said:

 Mind posting any proof cause you're making a lot of stuff up atm. Core 2 IPC is much lower than both Phenom II and FX. The direct competitor to Phenom II was Nahalem and the core i* first gen processors. Seriously, you're young, I get that, but read more and post less otherwise you look silly

https://www.cpubenchmark.net/compare/Intel-Core2-Quad-Q9650-vs-AMD-FX-4100-Quad-Core/1050vs255

 

With a 20% clock advantage, FX loses to Core 2 Quad.

 

I have the hardware mothballed, in storage, and a 2x Xeon E5450 workstation performs on the same level as the FX 8350.

[5x5]

3 minutes ago, KarathKasun said:

https://www.cpubenchmark.net/compare/Intel-Core2-Quad-Q9650-vs-AMD-FX-4100-Quad-Core/1050vs255

 

With a 20% clock advantage, FX loses to Core 2 Quad.

Passmark? Really? That's laughable.... And comparing an extreme edition CPU to an entry-level one? Even Userbench, which rates single core performance as 60% of the CPU's value, still has the FX 4350 ahead of the Q9550 by near 50% in single-core and ~25% in multi-core

https://cpu.userbenchmark.com/Compare/Intel-Core2-Quad-Q9550-vs-AMD-FX-4350/m334vs2880

 

At this point, I'm fairly certain you're trolling.

[KarathKasun]

2 minutes ago, 5x5 said:

Passmark? Really? That's laughable.... And comparing an extreme edition CPU to an entry-level one? Even Userbench, which rates single core performance as 60% of the CPU's value, still has the FX 4350 ahead of the Q9550 by near 50% in single-core and ~25% in multi-core

https://cpu.userbenchmark.com/Compare/Intel-Core2-Quad-Q9550-vs-AMD-FX-4350/m334vs2880

 

At this point, I'm fairly certain you're trolling.

It does not change the fact that FX with 20% higher clocks performs WORSE than Core 2.  Which was my statement.

 

FX with nearly 50% higher clocks gets less than 25% higher multi core, think about that for a second.

[5x5]

Just now, KarathKasun said:

It does not change the fact that FX with 20% higher clocks performs WORSE than Core 2.  Which was my statement.

That's not correct, at all. I've already linked you an example, yet you still deny the results. Instead, posting a well-known unreliable site and claiming moon logic. I will reporting you for misinformation soon.

[KarathKasun]

Just now, 5x5 said:

That's not correct, at all. I've already linked you an example, yet you still deny the results. Instead, posting a well-known unreliable site and claiming moon logic. I will reporting you for misinformation soon.

Go ahead.  IPC is Instructions Per Clock, FX is slower PER CLOCK than Core 2... this is a fact.

[5x5]

Just now, KarathKasun said:

Go ahead.  IPC is Instructions Per Clock, FX is slower PER CLOCK than Core 2... this is a fact.

Yes, you got one thing right. Good job!

 

Now remember that the Q9650 actually runs at pretty much the same clockspeed as the FX 4350 yet is a good 40% slower in single-threaded?

[KarathKasun]

3 minutes ago, 5x5 said:

Yes, you got one thing right. Good job!

 

Now remember that the Q9650 actually runs at pretty much the same clockspeed as the FX 4350 yet is a good 40% slower in single-threaded?

Umm, when is 4ghz = 3ghz?  That is a 33% clock advantage for the FX 4350... what are you smoking?

 

FX 8350 will throttle in the OPs motherboard, quite badly.  Down to the 3ghz range.  FX 8350e is that boards limit before you hit VRM issues, giving you a 3.3ghz clock, in which case FX is slower.  Not sure what you are trying to prove or defend here.

[5x5]

1 minute ago, KarathKasun said:

Umm, when is 4ghz = 3ghz?  That is a 33% clock advantage for the FX 4350... what are you smoking?

33% more clockspeed for 47% more performance - so the IPC on FX is about 15% higher or the same as Nahalem. Thank you for proving yourself wrong

[KarathKasun]

Just now, 5x5 said:

33% more clockspeed for 47% more performance - so the IPC on FX is about 15% higher or the same as Nahalem. Thank you for proving yourself wrong

And what of MT performance?  What about actual workloads?

 

Sorry, I have the hardware, FX performs at Core 2 levels in pretty much everything.  Oh, it does better in a few benchmarks... that is not real application performance.  There is a reason FX is known as a horrible CPU, it can do Nehalem levels of performance in a few benchmarks but fell far short in reality.

[5x5]

6 minutes ago, KarathKasun said:

And what of MT performance?  What about actual workloads?

 

Sorry, I have the hardware, FX performs at Core 2 levels in pretty much everything.  Oh, it does better in a few benchmarks... that is not real application performance.  There is a reason FX is known as a horrible CPU, it can do Nehalem levels of performance in a few benchmarks but fell far short in reality.

Oh you are hilarious mate. You mathed yourself into oblivion and still refuse to accept the fact that Core 2 is just slower than both Piledriver and Vishera. Seriously, if Core 2 was as good as your hallucinations, Intel would not have released Westmere at all.  But here, a few benchmarks where Core 2 Quads get beated into submission

[KarathKasun]

2 minutes ago, 5x5 said:

Oh you are hilarious mate. You mathed yourself into oblivion and still refuse to accept the fact that Core 2 is just slower than both Piledriver and Vishera. Seriously, if Core 2 was as good as your hallucinations, Intel would not have released Westmere at all.  But here, a few benchmarks where Core 2 Quads get beated into submission

And you are still comparing a 3ghz (sub 3ghz in some of those graphs) part to a 4ghz part.  Proving my point.

Oh, would you look at that, in those benchmarks, the performance delta is almost EXACTLY the same as the clock delta.  And there are faster C2Q chips you can pick up from the Xeon line of chips.  Up to 3.4ghz.  And they OC to the 4.4ghz range, resulting in a total performance difference of ~10% in ST OC to OC testing.  In MT, 2x E5450's keep up with the 8350 without overclocking.

 

Memory bandwidth is a pure synthetic, and the other numbers prove it.

 

Westmere was not much faster than Core 2, it was primarily brought about to integrate the cores into a single die and address memory bandwidth.  It didnt need to be significantly faster per core or per clock.

[Phantom329]

13 hours ago, KarathKasun said:

And what of MT performance?  What about actual workloads?

 

Sorry, I have the hardware, FX performs at Core 2 levels in pretty much everything.  Oh, it does better in a few benchmarks... that is not real application performance.  There is a reason FX is known as a horrible CPU, it can do Nehalem levels of performance in a few benchmarks but fell far short in reality.

What would you recommend that I do for my server? Should I try to get a different AM3 CPU or switch to a different socket. I don't want to spend a ton so common things on eBay or Craigslist would be preferred. I would like to be able to run multiple servers on different threads if possible so I don't want to get just a dual core, quad at a minimum.