• Happy holidays, folks! Thanks to each and every one of you for being part of the Tom's Hardware community!

Question Bottlenecked 3900x?

ps2cho

Distinguished
Oct 19, 2008
68
1
18,535
I just upgraded from a 2080 super to a 3080 FE and I am strangely seeing issues.

  1. FPS in PUBG (my primary game) is almost no different.
  2. If I try to stream now, my FPS is seeing a really large decrease.
  3. I see CPU usage is 80-90% in PUBG.

Is this an indication that my 3080 is now bottlenecked by my processor? I already reformatted and reinstalled a fresh Windows 10 copy and it made no difference.

CPU: 3900x cooled with H115i (max temps 55C)
GPU: 3080 FE
RAM: 32GB G Skill 4000MHz (set at 3600MHz to keep infinity fabric 1:1)
Mobo: ASUS Crosshair Impact XIII
PSU: Corsair SF750

Here is what it shows. Any ideas?
CPUuse-X2.jpg
 
There is no such thing as a "fresh Windows 10 "COPY"", so what did you ACTUALLY do? Did you do a CLEAN INSTALL, or did you use a refresh, reset or restore method of some kind? A factory reset, or refresh, and any form of restore, is worlds away from a clean install where all prior partitions on the drive are removed and Windows is directed to install to the unpartitioned space, where it will create all necessary partitions and perform any necessary formatting automatically.

And if you DID do a clean install, as outlined at the link below, did you THEN go to the product support page for your motherboard, and download, and install, the manufacturer drivers for the chipset, network adapters and audio? And did you THEN go to the Nvidia website and download the MOST recent driver for your card model, and install it? And, does your motherboard have the MOST recent BIOS version installed, or an older one?

Furthermore, since you're using a SF power supply, is this a small case, with potential cooling issues? If so, have you seriously monitored the CPU and graphics card for potential cooling issues using HWinfo, Ryzen master or another acceptable monitoring utility?

I assure you, that 3900x cannot POSSIBLY be a bottleneck for that graphics card.

What kind of memory configuration do you have? Is it two DIMMs or four? If it's two, which DIMM slots are they installed in?





If there are any steps listed here that you have not already done, it would be advisable to do so if for no other reason than to be able to say you've already done it and eliminate that possibility.



First,

Make sure your system has the MOST recent BIOS version installed. If it does not, then update. This solves a high number of issues even in cases where the release that is newer than yours makes no mention of improving graphics card or other hardware compatibility. They do not list every change they have made when they post a new BIOS release. In cases where you DO already have the latest BIOS version, simply resetting the BIOS as follows has a fairly high percentage chance of effecting a positive change in some cases so it is ALWAYS worth TRYING, at the very least.


BIOS Hard Reset procedure

Power off the unit, switch the PSU off and unplug the PSU cord from either the wall or the power supply.

Remove the motherboard CMOS battery for about three to five minutes. In some cases it may be necessary to remove the graphics card to access the CMOS battery.

During that five minutes while the CMOS battery is out of the motherboard, press the power button on the case, continuously, for 15-30 seconds, in order to deplete any residual charge that might be present in the CMOS circuit. After the five minutes is up, reinstall the CMOS battery making sure to insert it with the correct side up just as it came out.

If you had to remove the graphics card you can now reinstall it, but remember to reconnect your power cables if there were any attached to it as well as your display cable.

Now, plug the power supply cable back in, switch the PSU back on and power up the system. It should display the POST screen and the options to enter CMOS/BIOS setup. Enter the bios setup program and reconfigure the boot settings for either the Windows boot manager or for legacy systems, the drive your OS is installed on if necessary.

Save settings and exit. If the system will POST and boot then you can move forward from there including going back into the bios and configuring any other custom settings you may need to configure such as Memory XMP, A-XMP or D.O.C.P profile settings, custom fan profile settings or other specific settings you may have previously had configured that were wiped out by resetting the CMOS.

In some cases it may be necessary when you go into the BIOS after a reset, to load the Optimal default or Default values and then save settings, to actually get the hardware tables to reset in the boot manager.

It is probably also worth mentioning that for anything that might require an attempt to DO a hard reset in the first place, IF the problem is related to a lack of video signal, it is a GOOD IDEA to try a different type of display as many systems will not work properly for some reason with displayport configurations. It is worth trying HDMI if you are having no display or lack of visual ability to enter the BIOS, or no signal messages.

Trying a different monitor as well, if possible, is also a good idea if there is a lack of display. It happens.


Second,

Go to the product page for your motherboard or exact laptop model on the device manufacturer's website. Download and install the latest driver versions for the chipset, storage controllers, audio and network adapters. Do not skip installing a newer driver just because you think it is not relevant to the problem you are having. The drivers for one device can often affect ALL other devices and a questionable driver release can cause instability in the OS itself. They don't release new drivers just for fun. If there is a new driver release for a component, there is a good reason for it. The same goes for BIOS updates. When it comes to the chipset drivers, if your motherboard manufacturer lists a chipset driver that is newer than what the chipset developer (Intel or AMD, for our purposes) lists, then use that one. If Intel (Or AMD) shows a chipset driver version that is newer than what is available from the motherboard product page, then use that one. Always use the newest chipset driver that you can get and always use ONLY the chipset drivers available from either the motherboard manufacturer, AMD or Intel.


IF you have other hardware installed or attached to the system that are not a part of the systems covered by the motherboard drivers, then go to the support page for THAT component and check to see if there are newer drivers available for that as well. If there are, install them.


Third,

Make sure your memory is running at the correct advertised speed in the BIOS. This may require that you set the memory to run at the XMP profile settings. Also, make sure you have the memory installed in the correct slots and that they are running in dual channel which you can check by installing CPU-Z and checking the Memory and SPD tabs. For all modern motherboards that are dual channel memory architectures, from the last ten years at least, if you have two sticks installed they should be in the A2 (Called DDR4_1 on some boards) or B2 (Called DDR4_2 on some boards) which are ALWAYS the SECOND and FOURTH slots over from the CPU socket, counting TOWARDS the edge of the motherboard EXCEPT on boards that only have two memory slots total. In that case, if you have two modules it's not rocket science, but if you have only one, then install it in the A1 or DDR4_1 slot.



Fourth (And often tied for most important along with an up-to-date motherboard BIOS),

A clean install of the graphics card drivers. Regardless of whether you "already installed the newest drivers" for your graphics card or not, it is OFTEN a good idea to do a CLEAN install of the graphics card drivers. Just installing over the old drivers OR trying to use what Nvidia and AMD consider a clean install is not good enough and does not usually give the same result as using the Display Driver Uninstaller utility. This has a very high success rate and is always worth a shot.


If you have had both Nvidia and AMD cards installed at any point on that operating system then you will want to run the DDU twice. Once for the old card drivers (ie, Nvidia or AMD) and again for the currently installed graphics card drivers (ie, AMD or Nvidia). So if you had an Nvidia card at some point in the past, run it first for Nvidia and then after that is complete, run it again for AMD if you currently have an AMD card installed.



And lAnd last, but not least, if you have never done a CLEAN install of Windows, or have upgraded from an older version to Windows 10, or have been through several spring or fall major Windows updates (OR if you have installed ANY, EVEN REMOTELY QUESTIONABLE SOFTWARE, ESPECIALLY ANYTHING THAT'S BEEN PIRATED), it might be a very good idea to consider doing a clean install of Windows (AND THEN NOT PUT THAT QUESTIONABLE OR PIRATED SOFTWARE BACK ON) if none of these other solutions has helped. IF you are using a Windows installation from a previous system and you didn't do a clean install of Windows after building the new system, then it's 99.99% likely that you NEED to do a CLEAN install before trying any other solutions.


How to do a CLEAN installation of Windows 10, the RIGHT way
 
  • Like
Reactions: Kona45primo
Thanks for the response!

  1. Yes 100% fresh copy. Burned the ISO, deleted the partitions and installed brand new.
  2. I used AMD chipset drivers from AMD's website rather than Asus', so I can give that a try as its one thing I have not tried yet.
  3. Nvidia drivers are latest.
  4. Motherboard BIOS is latest.
  5. No cooling issues - CPU load is 55C, GPU max load is 70C. No throttling issues. All measured via HWmonitor.
  6. Two DIMS for the RAM since the ITX mobo is only two slots. I have selected the DOCP/XMS settings at 3600MHz and CPUz verifies its enabled.
RAMspec-S.jpg
 
If you are seeing 80%-90% CPU utilization on a game, it's an indication that the CPU is being hosed up by the game and GPU upgrades aren't going to improve performance. There's really nothing you can do about that aside from getting a better CPU.

that’s what it feels like to me…but I wanted to ask in case it could be anything else. The Ryzen 3D cache processors that hopefully will be announced tomorrow at CES would be the right upgrade for me if it indeed is the culprit.
 
Pretty hard to believe that a 3900x is going to bottleneck ANY modern game, to an appreciable degree. Regardless of single or multicore performance, this is a recent architecture that is still very much relevant and should be near the top 10-15% of CPUs out there in gaming performance.

Even when compared to the 11th and 12th Gen Intel i7 and i9 models, and the Zen 3 models, there isn't a HUGE difference in terms of single or multicore performance, especially since most games are only going to utilize a portion of the total cores, and the single core performance while it exists, isn't enough that I'd point to it and say "there it is". Especially since like two years ago that 3900x was just about the best gaming CPU you could get, give or take one or two models.

If you're using task manager to measure utilization, I would recommend you don't. Windows is notoriously poor at monitoring or reporting accurate information in this regard. HWmonitor, also notoriously bad, not just for that, but for reporting the WRONG sensor readings as something other than what they are in a lot of configurations, especially and moreso with some chipsets than others.

Try HWinfo, install it, choose "sensors only" and uncheck the "summary option", scroll down to the utilization data and run your game in the same way. See what you come up with there. Might be the same as task manager, but I sincerely doubt it.
 
...
RAM: 32GB G Skill 4000MHz (set at 3600MHz to keep infinity fabric 1:1)
...
So I understand...how have you set the timings for the memory? by using the XMP profile and then moving the DRAM clocks back to 1800Mhz (for 3600 memory)? If so, timings are most likely way looser than needed, so latency is way higher than it should be and contributing to degraded performance in high FPS gaming. It's hard to say what would be worse: de-syncing IF at 4000 or 3600 with sloppy timing but best would clearly be 3600 with tight timings. Running memory off-spec like that needs some work with DRAM Calculator to tighten up timings while maintaining stability.

How have you set up your processor? Run it in stock or PBO so that individual cores can freely boost to maximum clock speed as needed (you won't see none of that in task manager). Also be sure to run the latest AMD chipset driver and the Ryzen Balanced power plan (since it's Ryzen 3000). DO NOT lower voltage, leave it in AUTO. It can be expected peak as high as 1.5V if needed when boosting while gaming. It won't do that in heavier workloads, though, as the algorithm pulls back as it gets hotter.
 
Last edited:
I just ran CPU-Z single thread and its showing 100% to the reference (530 score).
Cinebench R23 scored 1,279 which looks about "right".
XMP at 3600MHz has a profile of 18-22-22-42 per CPU-Z memory tab.
Thus far from this, it does not appear my processor is showing any performance issues - however PUBG individually clearly is. So its either sloppy coding, or a bottleneck if its a very CPU-intensive game?

I turned on PBO. Later tonight once the kids are in bed I will update on the HWInfo utilization %'s for comparison sakes.
 
...
XMP at 3600MHz has a profile of 18-22-22-42 per CPU-Z memory tab.
...
It might be adequate at 4000 but that's very loose timing at 3600 where most memory is 16-18-18-18 and 'fast' is 14-15-15-15. Too, secondary timings can be just as important to get dialed back.

I'm not into FPS gaming but I'm pretty sure low latency memory is considered pretty important for performance. Also, it's not uncommon to lower memory clocks just to tighten up timings for other latency dependent applications. Through-put memory performance for applications like Cinebench and video transcoding won't be affected much if any at all.

And yeah, of course a more potent processor will help. Even a 5600X should improve things with gaming. But anything will be better off with lower memory latency.

I'm waiting to find out just how much the new 3D cache processors shake things up.
 
Last edited:
It might be adequate at 4000 but that's very loose timing at 3600 where most memory is 16-18-18-18 and 'fast' is 14-15-15-15. Too, secondary timings can be just as important to get dialed back.

I'm not into FPS gaming but I'm pretty sure low latency memory is considered pretty important for performance. Also, it's not uncommon to lower memory clocks just to tighten up timings for other latency dependent applications. Through-put memory performance for applications like Cinebench and video transcoding won't be affected much if any at all.

And yeah, of course a more potent processor will help. Even a 5600X should improve things with gaming. But anything will be better off with lower memory latency.

I'm waiting to find out just how much the new 3D cache processors shake things up.
I’ve always considered and read that RAM timing is a 5-10% gain sort of thing. Since I’m short 20-40% in FPS compared to my peers I just don’t feel like this is a RAM settings issue

I definitely will play with the RAM and see if I can get better timings. It was on sale at 32GB and I probably jumped on it too fast without researching timings much. I just saw the speed and figured I couldn’t go wrong.
 
I’ve always considered and read that RAM timing is a 5-10% gain sort of thing. Since I’m short 20-40% in FPS compared to my peers I just don’t feel like this is a RAM settings issue

I definitely will play with the RAM and see if I can get better timings. It was on sale at 32GB and I probably jumped on it too fast without researching timings much. I just saw the speed and figured I couldn’t go wrong.
I don't mean to say it's the only thing holding your system back, but it seems a low hanging fruit if you know what I mean. Well, low hanging but surrounded by a bunch of thorns since it's not as simple as just lowering timing. Some things have to be kept in ratio and the secondary timing values are pretty important too. Can be a lot of trial and error optimizing timing and the number of variables makes it astronomical if you don't use something like DRAM Calculator to at least get a start on it. This is one reason why it's better to invest in low-latency 3600 memory rather than average latency 4000; they may actually be the same but you have to figure out the timings for low latency 3600 operation if your CPU won't let you keep the IF sync'd at 4000.

Also really important for Ryzen 3000 is cooling as they overclock themselves if cooled well enough. Radiator placement figures in that as a hot GPU exhausting through the radiator doesn't help it much.
 
Last edited:
Pretty hard to believe that a 3900x is going to bottleneck ANY modern game, to an appreciable degree. Regardless of single or multicore performance, this is a recent architecture that is still very much relevant and should be near the top 10-15% of CPUs out there in gaming performance.

Even when compared to the 11th and 12th Gen Intel i7 and i9 models, and the Zen 3 models, there isn't a HUGE difference in terms of single or multicore performance, especially since most games are only going to utilize a portion of the total cores, and the single core performance while it exists, isn't enough that I'd point to it and say "there it is". Especially since like two years ago that 3900x was just about the best gaming CPU you could get, give or take one or two models.

If you're using task manager to measure utilization, I would recommend you don't. Windows is notoriously poor at monitoring or reporting accurate information in this regard. HWmonitor, also notoriously bad, not just for that, but for reporting the WRONG sensor readings as something other than what they are in a lot of configurations, especially and moreso with some chipsets than others.

Try HWinfo, install it, choose "sensors only" and uncheck the "summary option", scroll down to the utilization data and run your game in the same way. See what you come up with there. Might be the same as task manager, but I sincerely doubt it.


IDK why people say this, the numbers are right there on many review sites, Zen 2 was already beginning to show its limits in gaming on an RTX 2070 Super and that 3080 you have is faster by a good margin than an RTX 2080 Ti. A 3900X was roughly able to keep up with a 10400. Zen 2 really can't consistently keep up with anything over a 3060 Ti now.

You are CPU limited.

Assuming your motherboard allows it, if you want to game and have forked up a chunk of change for that 3080, you'd be well advised to get a 5800X. You are talking about a 20%+ difference in FPS with a 3080 vs any Zen 2.


SotTR.png
 
To each their own. I personally don't see the value in spending ~350.00 to gain 20-25fps, but I guess some do. Might be worth it if you're that serious about it.
..
Good points but he must already have laid out something north of $2000 for an RTX3080FE to replace a perfectly capable 2080TI, ostensibly in pursuit of the all important FPS.

To paraphrase a bit: in for a dollar, in for a dime.
 
Last edited:
Good points but he must already have laid out something north of $2000 for an RTX3080FE to replace a perfectly capable 2080TI, ostensibly in pursuit of the all important FPS.

To paraphrase a bit: in for a dollar, in for a dime.
I got the FE at BB for MSRP. I have no problems upgrading the CPU, but with 3d cache release info tomorrow, it makes sense for me to hold tight, then the jump from a 3900x to 5900-3d or whatever they will call it should be a better choice since its supposidely going to see a 15% IPC improvement, so going from my 3900x to it should see a decent 30%+ improvement.
 
IDK why people say this, the numbers are right there on many review sites, Zen 2 was already beginning to show its limits in gaming on an RTX 2070 Super and that 3080 you have is faster by a good margin than an RTX 2080 Ti. A 3900X was roughly able to keep up with a 10400. Zen 2 really can't consistently keep up with anything over a 3060 Ti now.

You are CPU limited.

Assuming your motherboard allows it, if you want to game and have forked up a chunk of change for that 3080, you'd be well advised to get a 5800X. You are talking about a 20%+ difference in FPS with a 3080 vs any Zen 2.


SotTR.png
Explain how upgrading the GPU lead to reduced framerates.
 
You're a lucky one for that...maybe we all are if the dam is breaking and stock improves all over!

And yah, holding tight at this point for Ryzen 3D seems the best idea. We'll know more soon what we're getting.
I've tried every thursday all year and scored it on the 23rd. I "heard" it was the largest drop to date. I failed on the first stock, said OOS within 250mi, then they did a 2nd round drop 15mins later and it went straight to checkout. Considered myself very lucky!
 
.... Considered myself very lucky!
BTW, one thing to be aware of is this could be a vapor launch. It's entirely possible considering how supply constrained AMD is for 7nm wafers.

This close to Zen 4 the PR win would be big enough just launching it and getting it into reviewers hands to show a clean lead and recapture the gaming crown. But there may never be huge numbers in the market and really doesn't need to be; just enough to get word of mouth traction in the enthusiast forums.

And launching isn't shipping so it may still be several months before supply hits the shelves even if it's not a vapor launch.
 
Last edited:
I agree with waiting for the 3d cache models, if you decide to upgrade anytime soon. From what I read the early indications are that just the 3d cache alone might account for as much as a 25% increase in performance on some games and as low as 8% in others, with a 15% gain from generational changes could mean a moderately higher overall gain versus generational alone. As drea.drechsler said, dropping the reduction in memory speed or at least (preferably) tightening up the timings is probably a good idea as well.
 
Explain how upgrading the GPU lead to reduced framerates.

That's not exactly what he said. What he said was: "If I try to stream now, my FPS is seeing a really large decrease."

He probably wasn't entirely CPU limited before the upgrade. It's not either on or off, you can be 60% CPU limited and go to being 95% CPU limited with a GPU upgrade while only seeing very mild FPS increases, which is probably what has happened. Once you're at that point, adding anything on top (like streaming) results in a disproportionately large loss of FPS.

Also Zen 2's (well documented) weakness has to do with how it's cache deals with the CCXs, and guess what streaming stresses?

Zen 3 fixes a lot of that, and would be the cheapest most painless upgrade. The best platform right now for streaming+gaming though is Alder Lake with DDR5. I've seen benchmarks showing it 40% faster than either AL + DDR4 or Zen 3 + DDR4. It's clearly a memory constrained use case with the more powerful CPUs like Zen 3 / Alder Lake paired to a high end GPU. OFC, good luck finding DDR5 right now.