AMD FX CPUs are underrated and their proper overlcocking

[Gameborn]

So, to start of , i'll explain a bit of how fx cpu is is designed from inside so you can better understand what i'll say later.

As you may or may not already know, fx cpus are modular. For example, eight core have 4 modules. Each module have two main processing cores in it, one FPU(floating processing unit) block and one l2 cache block, and each core have it's own l1 cache. The reason of bad single-thread fx performance is that of only one core in that module is used, only half of fpu is being used. Also there is l3 cache and memory controller on the chip alongside modules.

One of the weakest points of fx cpus is memory controller. l3 cache (which works at speed of memory controller) and slow ram just can't continue to feed all eight cores with data, so faster ram makes much bigger difference on fx (and ryzen too, but no as much) than on intel.  And by the way, recommended ram for fx is 1866 mhz. When fx cpu's was released (2011) most games used only two threads so they weren't very demanding in memory speed unlike single thread performance, this is why they performed so badly at launch.

Now about overclocking. Most people by overclocking mean just increasing clockspeed (because this is how you usually overclock intel) just by raising cpu multiplier. But, as described earlier, the bottleneck in fx is memory controller, it's inability to feed cores with data. Hyper-transport is module that controls how fast data from ram will move to CPU And this is where more advanced overclocking come in. You might seen "FSB or Base Clock" and "CPU-NB" aswell as "HT"(hyper-transport) options in bios. Cpu-NB is integrated north bridge with memory controller in it. Cpu cores, cpu-nb and hypertransport have their own multiplier which multiplies by base clock (FSB). This is why easiest method of properly overclocking fx is by raising base clock, and if needed cpu-nb and ht by themselves, but not all motherboards work correctly with it. So, when Base clock or all multipliers are being raised, now data will move faster from ram to l3 cache and from l3 cache to cores so they won't stand idle waiting for data. I've seen benchmark (can't remember link) where 4.0 ghz fx 8300 with overclocked cpu-nb and ht leaded 10-15 fps higher in battlefield 1 (85 instead of 70) than 4.4 ghz with stock cpu-nb and ht.

I'm not talking that i'm expert in cpu's, but this is what I know about them. I feel bad for amd that they released fx's too soon because in new games that can make good use of all 8 threads and l3 cache overclocked 8-core amd fx cpu's can compare with haswell i5 and sometimes even further. Good luck to amd with ryzen. Write your thoughts about this.

Sorry for messy explanation, I do my best at English.

[tom_w141]

No. Bulldozer was shit.

 

http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/5

 

EDIT: These CPUs have been out and pretty much unchanged since 2012. Frequency is a poor performance metric. The main reason bulldozer was so bad was the fact that they weren't "true" cores and their IPC was poor compared to Intel. This is why AMD have been able to improve Bulldozer's IPC by over 50% with Zen.

[Gameborn]

Just now, tom_w141 said:

No. Bulldozer was shit.

 

http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/5

have you read whole post? What are in this test? Skyrim? Dragon age origins? Diablo 3? I said that games when bulldozer was released was only effectively working with two threads and haven't near got full use of cpu. Thanks two new consoles with 8-core chips now games are optimized for multi-threading. 

[GDRRiley]

bulldozer sucked becasue of software not following it and por IPC even at 5ghz an i5 was beating it in most tasks

[Himommies]

There's quite clearly a large gap in performance.

[GDRRiley]

Just now, Gameborn said:

have you read whole post? What are in this test? Skyrim? Dragon age origins? Diablo 3? I said that games when bulldozer was released was only effectively working with two threads and haven't near got full use of cpu. Thanks two new consoles with 8-core chips now games are optimized for multi-threading. 

still they are not that much, dx12 may help the cards but it will not stop them from just being generally bad 

[Himommies]

2 minutes ago, Gameborn said:

have you read whole post? What are in this test? Skyrim? Dragon age origins? Diablo 3? I said that games when bulldozer was released was only effectively working with two threads and haven't near got full use of cpu. Thanks two new consoles with 8-core chips now games are optimized for multi-threading. 

You know know matter how much you say the newer 8 core CPU will alwasy beat the 5 year old 8 core cpu

[tom_w141]

2 minutes ago, Gameborn said:

have you read whole post? What are in this test? Skyrim? Dragon age origins? Diablo 3? I said that games when bulldozer was released was only effectively working with two threads and haven't near got full use of cpu. Thanks two new consoles with 8-core chips now games are optimized for multi-threading. 

Ok well lets look at bulldozer today:

 

Spoiler Alert: Gen 2 i5 still beats it 5 years later.

 

 

[Princess Luna]

I would say English is not your issue but more how you organized your paragraphs, it is just one huge block of text, I looked at it and it demotivated me to read straight off, consider using more punctuation and more clear sentences. 

[Gameborn]

1 minute ago, tom_w141 said:

Ok well lets look at bulldozer today:

 

Spoiler Alert: Gen 2 i5 still beats it 5 years later.

 

 

haha testing cups in 1440p. It's more of gpu test

[tom_w141]

Just now, Princess Cadence said:

I would say English is not your issue but more how you organized your paragraphs, it is just one huge block of text, I looked at it and it demotivated me to read straight off, consider using more punctuation and more clear sentences. 

tldr: The FX would have been good if games at the time used it to its full potential and a sort of explanation about how its modules work.

[tom_w141]

1 minute ago, Gameborn said:

haha testing cups in 1440p. It's more of gpu test

You are clutching at straws here... He also tests at 1080p, he is just being thorough.. It sucked then and it sucks now  Ask literally anyone.

[Gameborn]

1 minute ago, Princess Cadence said:

I would say English is not your issue but more how you organized your paragraphs, it is just one huge block of text, I looked at it and it demotivated me to read straight off, consider using more punctuation and more clear sentences. 

thanks, I just not write big texts like this often and usually I just have number of unorgonized thoughts, I just write them straight off, and it's hard to manage it in language that you use only in internet

[Princess Luna]

1 minute ago, tom_w141 said:

tldr: The FX would have been good if games at the time used it to its full potential and a sort of explanation about how its modules work.

AMD invested too hard on core count before core count made as much sense as it does now for desktop and gaming usage, not to mention they were so focused on beating Intel on raw frequency they end up with a much inferior architecture, to make a great processor you must improve balance everything like Intel does, otherwise that's what you get, a high core and frequency beast no doubt but under power in the end that will set your house on fire due to bad thermal solutions.

[Nena Trinity]

6 minutes ago, tom_w141 said:

Ok well lets look at bulldozer today:

 

Spoiler Alert: Gen 2 i5 still beats it 5 years later.

 

 

 

[Princess Luna]

2 minutes ago, Gameborn said:

thanks, I just not write big texts like this often and usually I just have number of unorgonized thoughts, I just write them straight off, and it's hard to manage it in language that you use only in internet

Where are you from?

[Gameborn]

Just now, Princess Cadence said:

Where are you from?

Ukraine 

[tom_w141]

3 minutes ago, Princess Cadence said:

AMD invested too hard on core count before core count made as much sense as it does now for desktop and gaming usage, not to mention they were so focused on beating Intel on raw frequency they end up with a much inferior architecture, to make a great processor you must improve balance everything like Intel does, otherwise that's what you get, a high core and frequency beast no doubt but under power in the end that will set your house on fire due to bad thermal solutions.

I'm not OP I know why it failed.

[SCHISCHKA]

15 minutes ago, Gameborn said:

I'm not talking that i'm expert in cpu's, but this is what I know about them. I feel bad for amd that they released fx's too soon because in new games that can make good use of all 8 threads and l3 cache overclocked 8-core amd fx cpu's can compare with haswell i5 and sometimes even further. Good luck to amd with ryzen. Write your thoughts about this.

What they failed to do was provide software compilers that optimised for their new architecture. in 2011/2012 if you were developing software or running VMs then FX was a bargain and you got 8 cores. If you were gaming you only got useful 4 cores, FX playing a game doesnt go over 50% because they failed to provide software optimizations to game publishers.

[tom_w141]

7 minutes ago, Nena360 said:

 

Adored's (don't know real name) is a very smart and insightful guy whose videos I enjoy, but Steve @ HWUnboxed has very thorough testing methodology which doesn't leave much room for argument.

[SpaceGhostC2C]

1 hour ago, tom_w141 said:

The main reason bulldozer was so bad was the fact that they weren't "true" cores and their IPC was poor compared to Intel.

Those are three different issues. 

 

1) "not being true cores" have nothing to do with it. Intel has been releasing 4-cores,and then 33% more expensive 4-cores for a while now. In the limit, a 4-module FX CPU could only have advantages against an hyperthreaded quad-core... if that was the only difference. 

 

2) IPC, on the other hand, was a problem on its own. It wasn't really that IPC was bad, as proven by his capable they still are today. It's just that AMD doesn't have any substantial improvement in that dimension at the time, only clocks increases. Just like KabyLake brought no IPC gains, and only manages to beat skylake on the basis of higher clocks. It doesn't mean KabyLake has bad IPC. Which takes us to 

 

3) Bulldozer wasn't "shit". In fact, it was one of the best platforms I've seen: AMD have us all unlocked CPUs, all the PCIe lanes (2.0 at the time), no restriction. It was almost suicidal, and both Intel and Ryzen would loom very different if most consumers had gotten used to it. 

But it was doom: AMD had its "KabyLake moment" right when Intel got its biggest IPC boost in a while, and ever since. Even worse, AMD had no lead at that time that could absorb the leap made by Intel (by contrast, Intel had such a lead at Ryzen's launch  that, despite getting a hit from stagnating while its competitor improves, will prevent any "sandy bridge moment" for AMD). 

 

 

If Intel had a similar  IPC boost then as it does now, FX would have looked great and praised by reviewers. If Intel had today the kind of IPC gains that Sandy bridge brought back in the day, people would be saying Ryzen is "shit". 

Sometimes we just get carried away by comparisons. 

[Jumper118]

Did I travel back to 2012 again -. - 

[-rascal-]

I did some testing with my Phenom II X6 1090T and FX-8350 a few years ago...unfortunately, I never finished the "User Review"...and the remaining data I had collected was lost when my Corsair ForceGT SSD died.

 

The 'Vishera' FX-8350 did not benefit much with a higher CPU-NB compared to the 'Thuban' Phenom II X6 1090T.

 

Stock HT frequency for the FX-8350 is already 2600 MHz. You can push this higher, but will cause overclock stability problems are you push it further and further out of spec (e.g. 3000 MHz). The performance benefit for higher HT vs higher stable CPU overclock was not worth the hassle, IMO.

 

From the...incomplete review that I did...

FX-8350 at stock settings, comparing 2200 MHz CPU-NB (stock), and 2600 MHz (overclocked).

Quote

FX-8350 @ 4.0GHz (stock), 2200MHz CPU-NB (stock), 2600MHz HT (stock) DDR3 RAM @ 1600MHz

AIDA64 Memory Benchmark: Read: 13775 MB/s, Write: 10897 MB/s, Copy: 16748 MB/s, Latency: 53 ns

SiSoftware Sandra 2014 Memory BW: Aggregated Memory performance: 15.63 GB/s, Integer Memory bandwidth: 15.65 GB/s, Float Memory bandwidth: 15.6 GB/s

Cinebench R15 - Single-Threaded, Turbo core OFF: CPU (cb): 97

Cinebench R15 - Multi-Threaded: CPU (cb): 644

Unigine Valley - Extreme HD Preset: FPS: 38.8, Score: 1622, Minimum FPS: 19.7, Maximum FPS: 75.8

Unigine Valley - Basic Preset: FPS: 76.2, Score: 3187, Minimum FPS: 23.1, Maximum FPS: 117.8

3D Mark Fire Strike 1.1: Overall Score: 6159, Graphics Score: 7083, Physics Score: 7828, Combined: 2680

Bioshock Infinite - Ultra DX11 with DDOF, 1920x1080: Average: 63.15 FPS, Minimum: 12.33 FPS

Bioshock Infinite - Lowest Settings, 1280x720: Average: 186.58 FPS, Minimum: 30.01 FPS

WinRAR 5.01 - Time for compression to complete: 282 seconds

 

 

FX-8350 @ 4.0GHz (stock), 2600MHz CPU-NB (overclocked), 2600MHz HT (stock) DDR3 RAM @ 1600MHz

AIDA64 Memory Benchmark: Read: 14375 MB/s, Write: 11716 MB/s, Copy: 17319 MB/s, Latency: 50.3 ns

SiSoftware Sandra 2014 Memory BW: Aggregated Memory performance: 17.53 GB/s, Integer Memory bandwidth: 17.52 GB/s, Float Memory bandwidth: 17.53 GB/s

Cinebench R15 - Single-Threaded, Turbo core OFF: CPU (cb): 97

Cinebench R15 - Multi-Threaded: CPU (cb): 645

Unigine Valley - Extreme HD Preset: FPS: 38.9, Score: 1626, Minimum FPS: 19.6, Maximum FPS: 75.7

Unigine Valley - Basic Preset: FPS: 78.6, Score: 3289, Minimum FPS: 23, Maximum FPS: 121.2

3D Mark Fire Strike 1.1: Overall Score: 6184, Graphics Score: 7108, Physics Score: 7912, Combined: 2687

Bioshock Infinite - Ultra DX11 with DDOF, 1920x1080: Average: 63.78 FPS, Minimum: 12.51 FPS

Bioshock Infinite - Lowest Settings, 1280x720: Average: 193.29 FPS, Minimum: 29.99 FPS

WinRAR 5.01 - Time for compression to complete: 271 seconds

 

Now compare the results for the Phenom II X6 1090T.

Quote

Phenom II X6 1090T @ 3.2GHz (stock), 2000MHz CPU-NB (stock), DDR3 RAM @ 1600MHz

AIDA64 Memory Benchmark: Read: 8879 MB/s, Write: 6859 MB/s, Copy: 10305 MB/s, Latency: 48.6 ns

SiSoftware Sandra 2014 Memory BW: Aggregated Memory performance: 12.58 GB/s, Integer Memory bandwidth: 12.58 GB/s, Float Memory bandwidth: 12.58 GB/s

Cinebench R15 - Single-Threaded, Turbo core ON: CPU (cb): 94

Cinebench R15 - Single-Threaded, Turbo core OFF: CPU (cb): 86

Cinebench R15 - Multi-Threaded: CPU (cb): 489

Unigine Valley - Extreme HD Preset: FPS: 37.3, Score: 1561, Minimum FPS: 17.7, Maximum FPS: 73.8

Unigine Valley - Basic Preset: FPS: 57.1, Score: 2388, Minimum FPS: 18.1, Maximum FPS: 91.4

3D Mark Fire Strike 1.1: Overall Score: 6092, Graphics Score: 6989, Physics Score: 6927, Combined: 2844

Bioshock Infinite - Ultra DX11 with DDOF, 1920x1080: Average: 59.37 FPS, Minimum: 10.66 FPS

Bioshock Infinite - Lowest Settings, 1280x720: Average: 163.44 FPS, Minimum: 30.09 FPS

WinRAR 5.01 - Time for compression to complete: 497 seconds

 

Phenom II X6 1090T @ 3.2GHz (stock), 3000MHz CPU-NB (overclocked), DDR3 RAM @ 1600MHz

AIDA64 Memory Benchmark: Read: 10481 MB/s, Write: 9562 MB/s, Copy: 11474 MB/s, Latency: 44 ns

SiSoftware Sandra 2014 Memory BW: Aggregated Memory performance: 16.1 GB/s, Integer Memory bandwidth: 16.09 GB/s, Float Memory bandwidth: 16.1 GB/s

Cinebench R15 - Single-Threaded, Turbo Core OFF: CPU (cb): 86

Cinebench R15 - Multi-Threaded: CPU (cb): 499

Unigine Valley - Extreme HD Preset: FPS: 37.7, Score: 1577, Minimum FPS: 17.8, Maximum FPS: 73.7

Unigine Valley - Basic Preset: FPS: 59.1, Score: 2474, Minimum FPS: 18.6, Maximum FPS: 95.9

3D Mark Fire Strike 1.1: Overall Score: 6118, Graphics Score: 7016, Physics Score: 6959, Combined Score: 2859

Bioshock Infinite - Ultra DX11 with DDOF, 1920x1080: Average: 62.95 FPS, Minimum: 10.6 FPS

Bioshock Infinite - Lowest Settings, 1280x720: Average: 178.76 FPS, Minimum: 30.17 FPS

WinRAR 5.01 - Time for compression to complete: 460 seconds

 

I still see a lot of questions about 'Bulldozer' and 'Vishera' even though it has already come and gone with 'Ryzen' in the market.

Heck, EVEN I have moved on from the FX-8350.

That said, since it is the summer (for me), and I have a bit more free time in the evening these days, I can do a "revisited" test...but this time, actually FINISH.

I still have my Phenom II X6 1090T, FX-8350, and ASUS Crosshair V Formula 990FX motherboard, and various DDR3 1600 MHz / 1866 MHz / 2133 MHz RAM.

 

On how it is set-up, my FX-8350 runs at 4.8 GHz,  2600 MHz HT, and 2600 MHz CPU-NB as a daily driver.

 

Things to note: 

• These tests were done on Windows 7 Ultimate 64-bit at the time
• Phenom II X6 1090T could only be overclocked to 4.1 GHz with 1.45V on air
• FX-8350 can pull 4.8 GHz with 1.4V, 4.9 GHz with ~1.45V, and only semi-stable with 5.0 GHz 1.5V+ ...on air cooling.
• However...I did order a EKwb P240 kit not too long ago, and it supports both AMD and Intel out of the box...but will take some time to set it up

Spoiler

 

[Gameborn]

7 hours ago, -rascal- said:

I did some testing with my Phenom II X6 1090T and FX-8350 a few years ago...unfortunately, I never finished the "User Review"...and the remaining data I had collected was lost when my Corsair ForceGT SSD died.

 

The 'Vishera' FX-8350 did not benefit much with a higher CPU-NB compared to the 'Thuban' Phenom II X6 1090T.

 

Stock HT frequency for the FX-8350 is already 2600 MHz. You can push this higher, but will cause overclock stability problems are you push it further and further out of spec (e.g. 3000 MHz). The performance benefit for higher HT vs higher stable CPU overclock was not worth the hassle, IMO.

 

From the...incomplete review that I did...

FX-8350 at stock settings, comparing 2200 MHz CPU-NB (stock), and 2600 MHz (overclocked).

 

Now compare the results for the Phenom II X6 1090T.

 

I still see a lot of questions about 'Bulldozer' and 'Vishera' even though it has already come and gone with 'Ryzen' in the market.

Heck, EVEN I have moved on from the FX-8350.

That said, since it is the summer (for me), and I have a bit more free time in the evening these days, I can do a "revisited" test...but this time, actually FINISH.

I still have my Phenom II X6 1090T, FX-8350, and ASUS Crosshair V Formula 990FX motherboard, and various DDR3 1600 MHz / 1866 MHz / 2133 MHz RAM.

 

On how it is set-up, my FX-8350 runs at 4.8 GHz,  2600 MHz HT, and 2600 MHz CPU-NB as a daily driver.

 

Things to note: 

• These tests were done on Windows 7 Ultimate 64-bit at the time
• Phenom II X6 1090T could only be overclocked to 4.1 GHz with 1.45V on air
• FX-8350 can pull 4.8 GHz with 1.4V, 4.9 GHz with ~1.45V, and only semi-stable with 5.0 GHz 1.5V+ ...on air cooling.
• However...I did order a EKwb P240 kit not too long ago, and it supports both AMD and Intel out of the box...but will take some time to set it up

  Reveal hidden contents

 

if you want, you can test it bit in modern games like the witcher 3 and battlefield 1, because only cintetic benchmarks and some older games get straight improvement by increasing clockspeed. I don't know much about older phenoms, and most likely they had even worser memory controller so overclocking cpu-nb helped on them too.

And yeah, even with good cooler, I wouldn't recommend running it as daily driver with higher than 1.4v.

[Nena Trinity]

On 27.5.2017 at 11:35 PM, tom_w141 said:

Adored's (don't know real name) is a very smart and insightful guy whose videos I enjoy, but Steve @ HWUnboxed has very thorough testing methodology which doesn't leave much room for argument.

They are both right, it only depends on the games tested really...