Question The FX 8350 revisited. Good time to talk about it because reasons.

Page 15 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hans Gruber

Platinum Member
Dec 23, 2006
2,092
1,065
136
Just curious. How much RAM was involved? In my experience having "too little" memory can have pretty much the same effect.
It's not the amount of memory. 16GB of memory for the computer and in the case of the 7700K it was a 1080 which has 8GB of vram. With my 3570k it was 16GB of computer memory and a GTX970 which is 4GB of vram. When I switched to a 3600 the problems went away. Another problem was the FPS. They were good but frames were being dropped and the game was choppy. The FPS counters do not account for this. It was a CPU issue in BF5. I should note BF1 did not have any effect close to what BF5 had on older CPU's.

The guy with the 7700K went to a 5600x and said the problem went away completely using the 1080 with it.
 

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
28,274
19,921
146
It's not the amount of memory. 16GB of memory for the computer and in the case of the 7700K it was a 1080 which has 8GB of vram. With my 3570k it was 16GB of computer memory and a GTX970 which is 4GB of vram. When I switched to a 3600 the problems went away. Another problem was the FPS. They were good but frames were being dropped and the game was choppy. The FPS counters do not account for this. It was a CPU issue in BF5. I should note BF1 did not have any effect close to what BF5 had on older CPU's.

The guy with the 7700K went to a 5600x and said the problem went away completely using the 1080 with it.
Good stuff! Exactly what I have been ranting about for months and months. You have to play the games to understand. Testing games without playing them, for timely clicks, needs to stop being venerated and spammed in forums. None of them can tell you if the scene or NPCs were not rendering properly, if the audio was borked, if the CPU could handle the game while talking with friends on discord. Or if in the middle of MP action, the FPS looked okay even though the experience wasn't.
 
  • Like
Reactions: scineram

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
28,274
19,921
146
I have been using the FX8350 @4,6GHz 32GB 1866MHz, with a RX6600, for some 1440p gaming.

Which causes me to recall, that one of the biggest talking points against it, was the platform itself. Turns out all the pearl clutching and bashing concerning PCIE 2.0 was pointless. 4K results are also irrelevant, as the low budget hardware scene has never targeted that res. And assuredly not when this CPU was relevant.

TPU's testing methodology reflects my experience. That being, you lose very little performance - https://www.techpowerup.com/review/amd-radeon-rx-6600-xt-pci-express-scaling/28.html

For games like GTA5, I do feel it pairs better with a GTX1080.
 

burninatortech4

Senior member
Jan 29, 2014
661
369
136
I have been using the FX8350 @4,6GHz 32GB 1866MHz, with a RX6600, for some 1440p gaming.

Which causes me to recall, that one of the biggest talking points against it, was the platform itself. Turns out all the pearl clutching and bashing concerning PCIE 2.0 was pointless. 4K results are also irrelevant, as the low budget hardware scene has never targeted that res. And assuredly not when this CPU was relevant.

TPU's testing methodology reflects my experience. That being, you lose very little performance - https://www.techpowerup.com/review/amd-radeon-rx-6600-xt-pci-express-scaling/28.html

For games like GTA5, I do feel it pairs better with a GTX1080.
16x PCIe 2.0 is equivalent to 4x PCIe 4.0. (Navi 23) Being a laptop chip to begin with, it makes sense that it wouldn't be bottlenecked by lanes. I agree the PCIe 2.0 bashing was unwarranted.
 
Last edited:

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
28,274
19,921
146
16x PCIe 2.0 is equivalent to 4x PCIe 4.0. Being a laptop chip to begin with it makes sense that it wouldn't be bottlenecked by lanes. I agree the PCIe 2.0 bashing was unwarranted.
The 6600 series was also laptop based like the 6400 and the 6500? If so, to quote the immortal Johnny Carson - I did not know that.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
No, its not "laptop based". It was dual use, yes, but it wasn't laptop first.

As for the PCIe lanes, it has 8 lanes that are PCIe 4 capable. Those same 8 lanes are what's used when its in a dual gpu configuration. Everyone accepted a long time ago that 8 lanes was enough for anything but the absolute highest end, which the rx6600 certainly is not.

It is also an 8 GB card. What did we all learn from the 5500xt? The larger memory cards were MUCH less affected by the smaller PCIe link because they do FAR less texture swapping over the PCIe link. This is the beef with the 6500xt. It has a narrow PCIe link AND low vram (4GB). It absolutely shows in benchmarks.

On top of all that, the rx6600 is the lower clocked narrower of the two 6600s, meaning that it is less capable of saturating the PCIe bus than the XT.

I'm glad the gpu works for you.
 

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
28,274
19,921
146
No, its not "laptop based". It was dual use, yes, but it wasn't laptop first.

As for the PCIe lanes, it has 8 lanes that are PCIe 4 capable. Those same 8 lanes are what's used when its in a dual gpu configuration. Everyone accepted a long time ago that 8 lanes was enough for anything but the absolute highest end, which the rx6600 certainly is not.

It is also an 8 GB card. What did we all learn from the 5500xt? The larger memory cards were MUCH less affected by the smaller PCIe link because they do FAR less texture swapping over the PCIe link. This is the beef with the 6500xt. It has a narrow PCIe link AND low vram (4GB). It absolutely shows in benchmarks.

On top of all that, the rx6600 is the lower clocked narrower of the two 6600s, meaning that it is less capable of saturating the PCIe bus than the XT.

I'm glad the gpu works for you.
Thanks, that's more in keeping with what I've seen and read.
 

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
28,274
19,921
146
Yea 6500 and 6400 are primarily intended for laptop use. I assumed the lane count of 6600 meant it had laptop DNA. I guess I misspoke.
It's all good. I learned long ago, Cunningham's law is a superior method of acquiring new knowledge in forum discussions. Peoples' urge to tell you how wrong you are is irresistible.

Now back to me white knighting Vishera :p Having done testing with Witcher 3 lately, I was disappointed homie's FX board went belly up, and he could not test it here -


He changes both the i5 and amount of ram, and incorrectly concludes the ram was responsible for the frame pacing issues. Having tested a better quad core in the 3200G with 16GB of much faster DDR4, it was the i5 and ram both having nothing left to give. It got a chuckle out of me, because the i5 is pegged at 100% and homie is like "I guess it needs more ram since it's maxed out." If he had tried the i5 with 8GB, a fast GPU, ultra settings, and crowd density maxed, frame pacing would have been ugly around the docks and market where he was testing.

MY FX 8350 performs similar to the 3770K he uses. But because I was using a GTX 1080 with everything including hairworks maxed, I could see 100% spikes briefly in that same area. Both on Roach or sprinting around. The 3200G would suffer more in those spots.

I expressed my surprise about how intensive the game is for its age, and how many cores it can hit. I was stoked to see he had a similar reaction.
 

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
28,274
19,921
146
Overclocking the ram, swapping the cooler, and a 5-600MHz CPU overclock, would give it a nice boost. Not overclocking FX is a PCMR felony offense. UK Steve is really a console gamer, so he gets off with a warning this time. ;) And that is a crap cooler.

I haven't played Elden Ring yet, not really into S&M. :p But I have read ram speed is important too. I will test it with 2133MHz and 4.6GHz when I buy it. I will use a GTX 1080 though. Not bothering with a weak GPU for a game like that. Hopefully I can get over 30fps that way; time will tell.
 

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
28,274
19,921
146
OC? With that cooler? We don't want him to melt anything... important... ;)
I did cover that in my post. But I still like the joke. He has AMD compatible coolers laying around. If it fits AM4 it will work with AM3+. I have used everything from a Wraith prism to the current Noctua D15 with FX.

I wasn't going to overclock mine since that has been done to death for the last decade. But I was like

free.jpg
 

rvborgh

Member
Apr 16, 2014
195
94
101
interesting thread.

i just replaced the four of the Opteron 61xx ES chips (48 cores at 3.0 GHz) in my Quad Socket H8QGi-F with a set of Opteron 63xx ES chips (32 cores).

16 cores in a G34 package got rather toasty, so i ended up doing a "Compute Unit" downcore to 4 Piledriver cores per die (8 cores per G34 chip), and total of 16 cores. This clock gates one of the cores in each compute unit, allowing that core to not have to share anything with the other core (effectively disabling CMT). This helps a bit with IPC.

I just finished my overclocking effort with these. i was able to up the ref clock to 205, and run pb1 (all core turbo) at 4 GHz, and pb0 (single core turbo) at 4.5 GHz with complete stability. Quite a few of the cores can run 4.6, and some 4.7 with sane voltages, but i'm trying to keep my VRMs alive and down in the 1.2v range.

The G34 package seems to work well for this combo as there is more heat sink area for 8 cores.

The Piledrivers are decent and mild upgrade over the K10. The main advantage to these Piledriver Opterons isn't really the cores themselves, but the newer memory controller. With the K10s i could run the processors at 3.6-3.8 GHz every 3rd core to keep heat down, but the memory controllers just could not keep up.

Anyways, its been fun spending time getting these running at their optimum speed.

Final results 2.jpg
 

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
28,274
19,921
146
interesting thread.

i just replaced the four of the Opteron 61xx ES chips (48 cores at 3.0 GHz) in my Quad Socket H8QGi-F with a set of Opteron 63xx ES chips (32 cores).

16 cores in a G34 package got rather toasty, so i ended up doing a "Compute Unit" downcore to 4 Piledriver cores per die (8 cores per G34 chip), and total of 16 cores. This clock gates one of the cores in each compute unit, allowing that core to not have to share anything with the other core (effectively disabling CMT). This helps a bit with IPC.

I just finished my overclocking effort with these. i was able to up the ref clock to 205, and run pb1 (all core turbo) at 4 GHz, and pb0 (single core turbo) at 4.5 GHz with complete stability. Quite a few of the cores can run 4.6, and some 4.7 with sane voltages, but i'm trying to keep my VRMs alive and down in the 1.2v range.

The G34 package seems to work well for this combo as there is more heat sink area for 8 cores.

The Piledrivers are decent and mild upgrade over the K10. The main advantage to these Piledriver Opterons isn't really the cores themselves, but the newer memory controller. With the K10s i could run the processors at 3.6-3.8 GHz every 3rd core to keep heat down, but the memory controllers just could not keep up.

Anyways, its been fun spending time getting these running at their optimum speed.

View attachment 70534
Fun is the name of the game.

That is the coolest thing I will see today. Thanks for posting it.
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
interesting thread.

i just replaced the four of the Opteron 61xx ES chips (48 cores at 3.0 GHz) in my Quad Socket H8QGi-F with a set of Opteron 63xx ES chips (32 cores).

16 cores in a G34 package got rather toasty, so i ended up doing a "Compute Unit" downcore to 4 Piledriver cores per die (8 cores per G34 chip), and total of 16 cores. This clock gates one of the cores in each compute unit, allowing that core to not have to share anything with the other core (effectively disabling CMT). This helps a bit with IPC.

I just finished my overclocking effort with these. i was able to up the ref clock to 205, and run pb1 (all core turbo) at 4 GHz, and pb0 (single core turbo) at 4.5 GHz with complete stability. Quite a few of the cores can run 4.6, and some 4.7 with sane voltages, but i'm trying to keep my VRMs alive and down in the 1.2v range.

The G34 package seems to work well for this combo as there is more heat sink area for 8 cores.

The Piledrivers are decent and mild upgrade over the K10. The main advantage to these Piledriver Opterons isn't really the cores themselves, but the newer memory controller. With the K10s i could run the processors at 3.6-3.8 GHz every 3rd core to keep heat down, but the memory controllers just could not keep up.

Anyways, its been fun spending time getting these running at their optimum speed.

View attachment 70534

With 16 active cores it should score about 1550 pts in CB R15, and something like 2500 pts with all 32 cores activated, in term or CB R23 this is 9200 pts and 14800 pts respectively.

Dunno what is the OS but with the FX8350 Windows scheduler at the time did schedule one thread per module untill it s exhausted and then rely on CMT for more than 4 threads.
 

Hotrod2go

Senior member
Nov 17, 2021
298
168
86
I remember arguments back in the day when Intel was accused of making their compilers favour Intel chips first, then others later (not that there were many "others"). So this of course in turn downgraded the FX line of cpus when benchmarking. I use to have links to those articles about this, bookmarked but have lost the backups of those bookmarks... more than a decade ago now all that was.
 

rvborgh

Member
Apr 16, 2014
195
94
101
With 16 active cores it should score about 1550 pts in CB R15, and something like 2500 pts with all 32 cores activated, in term or CB R23 this is 9200 pts and 14800 pts respectively.

Dunno what is the OS but with the FX8350 Windows scheduler at the time did schedule one thread per module untill it s exhausted and then rely on CMT for more than 4 threads.

i think my system is a bit short. i just did a Cinebench R15 run and the score came out to 1420cb. All 16 cores running at 4 GHz. Ref clock of 205. Temps were 43C on processor 2 and 39C on processor 1. Have not tried running it with all 32 cores yet since i actually do not know if all 32 cores can run at 4 GHz without burning up the sockets :) Perhaps that will be the next experiment. The problem is the Noctua coolers. They were good for keeping the temps of my former 61xx ES processors at around 55C when doing IntelburnTest runs for hours on end, but i have no idea how they would fare with another 100w of power hitting them per socket.

CB R15 when i had the 61xx ES procs installed (all 48 cores at 3.0 GHz at 1.2v) scored around 3225cb. These are doing 1420cb with just 16 cores. i'd say that is a pretty good showing for the Piledriver cores compared to the K10.

Ah... yes. That would give you 4 nonshared cores. Having excess cores allows the Piledrivers to run as they should on all 8 cores.

PS: These 16 cores are pulling about 540w at the wall during the CB R15 run. So the power draw of having all 4 sockets populated with these 6380ES and running all 64 cores at 4 GHz would be something else. i've got dual redundant 1400w power supplies so that would be okay, but i don't think the motherboard could take that.

PPS: i finally did do a CB 15 run with all cores at 3.5 GHz. Around 2150cb. Sure enough, per core around 15% slower per core due to CMT/sharing.
 
Last edited:
  • Wow
Reactions: DAPUNISHER

rvborgh

Member
Apr 16, 2014
195
94
101
Fun is the name of the game.

That is the coolest thing I will see today. Thanks for posting it.

Thanks :) It has been fun :) Albeit a bit of strange fun pushing these. The scary part of it being... wondering whether one of the 6 phase VRMs on this SuperMicro server board will go up in smoke or not :) I've got the SC748 fans blowing across the ones on the front for processor 2, but the ones for processor 1 are at the rear of the board with a lot less cool air being blow across their heat sink.
 

burninatortech4

Senior member
Jan 29, 2014
661
369
136
interesting thread.

i just replaced the four of the Opteron 61xx ES chips (48 cores at 3.0 GHz) in my Quad Socket H8QGi-F with a set of Opteron 63xx ES chips (32 cores).

16 cores in a G34 package got rather toasty, so i ended up doing a "Compute Unit" downcore to 4 Piledriver cores per die (8 cores per G34 chip), and total of 16 cores. This clock gates one of the cores in each compute unit, allowing that core to not have to share anything with the other core (effectively disabling CMT). This helps a bit with IPC.

I just finished my overclocking effort with these. i was able to up the ref clock to 205, and run pb1 (all core turbo) at 4 GHz, and pb0 (single core turbo) at 4.5 GHz with complete stability. Quite a few of the cores can run 4.6, and some 4.7 with sane voltages, but i'm trying to keep my VRMs alive and down in the 1.2v range.

The G34 package seems to work well for this combo as there is more heat sink area for 8 cores.

The Piledrivers are decent and mild upgrade over the K10. The main advantage to these Piledriver Opterons isn't really the cores themselves, but the newer memory controller. With the K10s i could run the processors at 3.6-3.8 GHz every 3rd core to keep heat down, but the memory controllers just could not keep up.

Anyways, its been fun spending time getting these running at their optimum speed.

View attachment 70534
Really neat. Any experience with the 6366 HE? How would 4 of those compare to this?
 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
I remember arguments back in the day when Intel was accused of making their compilers favour Intel chips first, then others later (not that there were many "others"). So this of course in turn downgraded the FX line of cpus when benchmarking. I use to have links to those articles about this, bookmarked but have lost the backups of those bookmarks... more than a decade ago now all that was.

The best example I can recall was the BapCo issue where they were checking the vendor ID rather than CPUID for SSE support when it was the Athlon XP vs P4 days. There may be others but that's the one that comes to mind first. That said, The FX series was pretty much a turd that aged gracefully.

The SSE thing I mentioned is covered here.

 
Last edited:
  • Like
Reactions: AnitaPeterson

UsandThem

Elite Member
May 4, 2000
16,068
7,380
146
The best example I can recall was the BapCo issue where they were checking the vendor ID rather than CPUID for SSE support when it was the Athlon XP vs P4 days. There may be others but that's the one that comes to mind first. That said, The FX series was pretty much a turd that aged gracefully.

The SSE thing I mentioned is covered here.

Somehow I remember that article from all those years ago. Aanand Lal Shimpi was such a great writer on stories like this,
One of the most honest quotes we ever received from an Intel employee (although not an official quote for obvious reasons) was in response to the question "Will Intel ever make a return to the level of dominance that existed a few years ago?" His response was on the order of, 'not as long as AMD continues to stay on the ball'. With the Athlon processor AMD finally had a powerful part on their hands and unless they seriously screw things up, you can expect to see this sort of heated competition between AMD and Intel for some time to come

That outlook was spot-on for close to 10 years...........when Bulldozer arrived, and Intel all but took the crown on having much better CPUs compared to AMD for the next decade......all until the arrival of Zen 1000 series in Released in 2017.
 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
Somehow I remember that article from all those years ago. Aanand Lal Shimpi was such a great writer on stories like this,


That outlook was spot-on for close to 10 years...........when Bulldozer arrived, and Intel all but took the crown on having much better CPUs compared to AMD for the next decade......all until the arrival of Zen 1000 series in Released in 2017.

Anand was great. Ian Cutress was great. Ryan, eh not so much. He would probably be fine reviewing video cards, if Anandtech still got them. As the head of it though, I have not been impressed. I am glad Ian went his own way.

I like the site but they have fallen of a cliff in quality over the past few years.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
That outlook was spot-on for close to 10 years...........
Technically, the outlook flopped when Hector Ruiz took the reigns.

The Longmont/Mile High CPU Cores were killed off by him. This is after he gave them a go ahead too. It is basically Intel seeing Pentium M at Israel. Then, having the CEO go ballistic and shuttering that CPU design center down.

2002-2008:
-> Every next-generation(non-K7 related) High Performance core never launched.
-> CEO in a fit shut down the Low Power cores for next-gen Personal Internet Communicator.

Now everyone and their family has at least one "MiniPC" somewhere....
Would have been nice to have gotten the Longmont "Bulldozer" core in 2009 on 45nm in a minipc(>10W target). Rather than wait for the Sunnyvale "Bulldozer" core in 2011 with no miniPC(>45W target).

Longmont "Bulldozer" core => APU First, Server Second
A single 45nm Bulldozer processor is smaller than 45nm Greyhound processor.
10W first, then scale up many cores up.
2009 Server CPU target => When finessing it out of AMD, 8-core 45nm LM "Bulldozer" would be smaller than 4-core "Greyhound" 65nm. Cost-effectiveness(small die)/power efficiency(small core(SINGULAR core)) was the goal under the 2005-2007 Longmont-Mile High design.
1st Bulldozer Design => Premium "Internet Box"/PIC product with Fusion.
2nd Bulldozer Design => Many-core HPC product standard market. 8/16 HEDT/Server first, then 4 client product with a really really small die.

Sunnyvale "Bulldozer" module => Server First, APU Last
A single 32nm Bulldozer processor is larger than 32nm Greyhound processor.
140W first, then lower.

finalcut1.jpeg
finalcut.png

Orignal Bulldozer by PH/CRM/BB was from scratch off of Bobcat, while its HR/DM/MB remake was based off of Andy Glew's K10. The current "Bulldozer" stole the name from the original core.

Mobile first codename:
"Bulldozer is the code name for the Fusion chip that will be designed for everything from handhelds to servers."
Mobile first platform:
"Bulldozer will be part of the "Falcon" PC platform that also includes an integrated memory controller, a graphics processor, cache memory and a PCI Express controller."

[In 2007 BD was going to be a 45nm product. When we decided to make the change, do it right and move it to 32[n]m everything changed.

Anything that was written about BD in 2007 is probably not accurate.]
45nm processor != 32nm processor

Initial patent w/ Bobcat(unchanged from early cancelation and revival):
bobcattobulldozer.png
Should be noted just to iterate on this again:
Execution core => Scheduler + Function Units => Bobcat has 3 schedulers tied to specific functions thus has three execution cores.

Initial architecture details of Bulldozer:
Screenshot 2022-11-08 at 02-08-48 GitHub - open64-compiler_open64 open64 compiler.png
The original Bulldozer followed the same example as Bobcat: Three separate instances of; Integer Execution, Address Execution, and Floating Point Execution. However, the Integer and Floating Point executions were double instanced, and Address Execution was widen for the higher memory IPC. It was not duplicated so it can centrally address all memory<->reg operations for all clusters(2x Int/2x FPU). As the Original Bulldozer had a single-threaded focus, most new workloads Media/HPC(SSE5) were expected to run in single-threaded mode. While, existing code optimized for K8/Greyhound(<SSE4) was expected to run on Lo-/Hi- clusters. Thus, getting the highest single-threaded and multi-threaded core performance.
New workloads => Use both for single-threaded workloads // More ILP for Integer and more DLP for SIMD.
Old workloads => Use both as multi-threaded workloads // Instead of L2<->L2 like that of K10-BD/K8H-GH, the original LM-Bulldozer core did sliding L1d<->L1d transfers for multithreaded, via central address execution.

Taking the above(+post-mortem investigation of 2005-2007 interviews post-Moore's CMT slides and post-Hester's Bulldozer slides):
FU126A => Two instances of 2 ALUs
FU126B => Two instances of 2 FPUs
FU126C => One instance of 3 AGUs : 1x L1D
50% area overhead was not against Greyhound, but rather Bobcat. As Bulldozer, the original design started from Bobcat.

Actual Bulldozer: Synthesized macros via Longmont/Mile High (Low power/High Density) -> K10 Bulldozer: Template-based full custom macros via Sunnyvale (High performance/No density)
 
Last edited:
  • Like
Reactions: AnitaPeterson