Need a bit of advice concering GPU's

[KFXE 13]

Hello,

I have two separate machines running FSX.  They both seem to run the sim about the same.  But that's the problem.  I don't think they should.  I don't even think it should be close.

Here are the two machines.  In another thread I got around to pointing out the two machines specs and asked why they would seem to run the sim the same but didn't really get a clear answer on that.  Probably because it was about an OOM error and not running the Sim per se?   Anyway, here are the two machines......

MACHINE 1

OS - Windows 10 Pro, Processor    AMD A10-9700 3094Mhz  10 Cores 6C + 4G,   RAM 16GB, GPU - PNY 1060 GTX 6GB, MAINBOARD - AsRock AB350M Pro4 (Release Date February 2017)

MACHINE 2

OS - Windows 7 Ultimate, Processor    AMD Athlon 64 X2 Dual Core 5200+ 2600Mhz,   RAM 2GB, GPU - Nvidia Quadro 2000, MAINBOARD - FoxxConn A78AX (release Date - January 10, 2011 )

In that other thread I asked.......Why on Earth aren't these two machines showing a massively different FSX experience?  Again, the sim looks and feels nearly identical on the two.  Now, weeks later, I see more and more that the older machine might actually be providing the better simulation in terms of graphics, clarity, speed etc.

So now I'm REALLY beginning to think the card is either bogus, a cheap clone or something else is going on.   Problem is, I don't really know if these two machines with these specific graphics cards actually should be that much different?   But it's bugging me because I thought with the never more powerful machine AND the 1060 GTX 6GB graphics card, I'd see a MAJOR difference....but I don't.   In fact, as I said....the Older machine with the Quadro 2000 may be out performing the never machine with the 1060 GTX.

Is there a way to actually test this card against benchmarks to see if it is authentic and working properly?   I would take a WAG (Wild word not allowed Guess) that this 1060 GTX card is working about about half capacity or something.

Please help if you can.  This is REALLY bugging me.

Thanks guys !!

 

[scianoir 610]

As I understand, FSX is more processor dependant rather than GPU dependant which may explain why you are not seeing much difference as the ‘hold up’, as it were, In processing data tends to be in the CPU rather than the GPU. After my GPU packed in recently, I replaced it with one which, on paper, looked about twice as powerful but I cannot say that I see any marked difference in performance with the new graphics card.

I’m sure someone who is more technically qualified than me can give you a more scientific explanation but that is how I understand it from a simplified viewpoint!

Bill

[pgde 571]

The freeware GPUZ utility will identify your card, chipset, etc.

[KFXE 13]

Thanks guys.

I "might" have found the problem.  Not totally sure , but I "think" this is it.....

I ran the program called UserBenchMark to check on my system.

The results showed the GPU working as expected and right in the middle of all the tests so I think I can now say the GPU card is ok.

But what I also found is that my MEMORY is WAY BELOW expectations.   I have two sticks of DDR4 2400 Corsair Vengence LPX installed.   But, when I went to the Corsair website it says that memory is compatible with Intel 100 Series, Intel 200 Series, Intel 300 Series, Intel X299.   I have an AMD AM4 System.  (FacePalm HERE)

Looks like all this time I've been running the wrong RAM Memory.  It works and I never got any kind of indication from the BIOS that it was wrong, but it's super slow.   What hurts even more is at the time I bought this memory, prices were spiked sky high so I think I paid $200 for this Ram that's now worth about $60.   UGH !

Even in FSX it ran good and was really smooth, but about like my MUCH older AM2 system.  So, I'm betting if I switch to AM4 approved RAM it will improve.   Not sure, but hoping.

The Corsair Vengence LPX appears to be ECC memory and my system requires NON ECC RAM.   I guess that also means it's "buffered" and my system needs UNbuffered RAM.

Edited December 26, 2019 by KFXE

[Bob Scott 7,715]

I think you're barking up the wrong tree...if the RAM works in the PC now, then there's nothing seriously wrong or incompatible.  Compatibility listings are non-exhaustive...they only list what they've tested, which does not mean that everything else is not compatible.  2400 MHz RAM is very slow, however.

The long pole in your performance problem is single-core processor speed...those machines run at 3.1 and 2.6 GHz respectively, which is relatively slow and not very far apart in terms of single-core performance.  Nothing in the sim can work any faster than the main thread running on the first core.  If the A10 can be overclocked to somewhere around 4GHz, that'd help a lot. 

 

[Luke 782]

1 hour ago, KFXE said:

Looks like all this time I've been running the wrong RAM Memory.

No, you've been running the wrong CPU. An A10 APU is what you buy grandma to surf Amazon and play Candy Crush.

Overclocking it is unlikely to make any significant difference. If you want AMD, get a Ryzen 2.

Cheers!

[KFXE 13]

Ok.  Good advice.

What does your CPU have to do with the appearance and realism of the Simulation?    I thought the GPU controlled how good (true to life) the simulation appears and the CPU controlled frame rate and how fast the simulation can run?

No?

Also, if my current RAM is slower than 98% of others with the same equipment, then how is addressing the CPU and NOT the RAM going to fix the issue?

Edited December 26, 2019 by KFXE

[Bert Pieke 7,584]

Most of what you see in FSX is generated by the CPU... The GPU just puts it on the screen.

P3D (newer) uses the GPU more, offloading the CPU in the process.

So, different GPUs are not going to change what you see in FSX.

I also agree that your RAM is probably just fine..

 

Edited December 26, 2019 by Bert Pieke

[Bob Scott 7,715]

2 hours ago, KFXE said:

What does your CPU have to do with the appearance and realism of the Simulation?    I thought the GPU controlled how good (true to life) the simulation appears and the CPU controlled frame rate and how fast the simulation can run?

No?

Also, if my current RAM is slower than 98% of others with the same equipment, then how is addressing the CPU and NOT the RAM going to fix the issue?

In default (DX9) FSX, most of the work to compose a frame is done in the CPU.  The GPU does some finishing work, such as anti-aliasing, but the CPU has to first layer and place all of the draw calls for each object in the frame in the proper order.

If you use the incomplete DX10 "preview" mode in FSX along with the appropriate utilities like Steve Parson's DX10 Fixer, then more of the frame composition work can be offloaded to the GPU.  For example duplicated objects like trees and clumps of grass need only be passed to the GPU once with multiple draw calls referencing the same object.  In this case, a more powerful GPU does help some.  P3D makes much better use of the GPU using DX11 and 64-bit driver code.

CPU throughput depends on clock speed, a metric we call IPC (instructions-per-clock) which is the efficiency with which the CPU can perform instructions, and RAM latency (the speed at which RAM can answer a read/write request--a function of memory clock speed and memory latency).  So both CPU speed and RAM latency matter, and throughput is a balance of the two.  If the RAM is very slow, then the CPU compensates for that by inserting wait cycles during which a particular instruction in the pipeline does nothing as it sits idle waiting for the amount of time it takes for the memory subsystem to respond to a read request.  It's actually somewhat more complex still, because the CPU pipelines multiple instructions at once, so for each clock step there can be a bunch of instructions being processed at once, reducing the effect of waiting on RAM, and the CPU also uses a very fast block of internal cache memory, so that whenever it issues a read request it actually gets a small block of RAM and puts it into its cache, which speeds things up as long as the program doesn't branch elsewhere and have to discard/overwrite the pre-cached memory that it read.

So the bottom line is that CPU clock speed and IPC generally make more of a difference than RAM latency, but only to a point.  Both matter.

Regards

 

[kevinfirth 1,392]

7 hours ago, w6kd said:

In default (DX9) FSX, most of the work to compose a frame is done in the CPU.  The GPU does some finishing work, such as anti-aliasing, but the CPU has to first layer and place all of the draw calls for each object in the frame in the proper order.

If you use the incomplete DX10 "preview" mode in FSX along with the appropriate utilities like Steve Parson's DX10 Fixer, then more of the frame composition work can be offloaded to the GPU.  For example duplicated objects like trees and clumps of grass need only be passed to the GPU once with multiple draw calls referencing the same object.  In this case, a more powerful GPU does help some.  P3D makes much better use of the GPU using DX11 and 64-bit driver code.

CPU throughput depends on clock speed, a metric we call IPC (instructions-per-clock) which is the efficiency with which the CPU can perform instructions, and RAM latency (the speed at which RAM can answer a read/write request--a function of memory clock speed and memory latency).  So both CPU speed and RAM latency matter, and throughput is a balance of the two.  If the RAM is very slow, then the CPU compensates for that by inserting wait cycles during which a particular instruction in the pipeline does nothing as it sits idle waiting for the amount of time it takes for the memory subsystem to respond to a read request.  It's actually somewhat more complex still, because the CPU pipelines multiple instructions at once, so for each clock step there can be a bunch of instructions being processed at once, reducing the effect of waiting on RAM, and the CPU also uses a very fast block of internal cache memory, so that whenever it issues a read request it actually gets a small block of RAM and puts it into its cache, which speeds things up as long as the program doesn't branch elsewhere and have to discard/overwrite the pre-cached memory that it read.

So the bottom line is that CPU clock speed and IPC generally make more of a difference than RAM latency, but only to a point.  Both matter.

Regards

 

This is one of the most concise and understandable explanations of the relationship between CPU and memory I've read.  Beautiful.

Took me a long time to grasp the technical basics of getting a good balanced system...

1. Buy a 6/8 core CPU, unlocked...

2. overclock it as well as possible...

3. Buy the lowest latency mem possible (it really does make economic sense when you understand the above post)

4. if you want to go hardcore, you can try to overclock the mem

5. overclock the CPU cache, something I think a lot of people overlook

6. apply for a new mortgage and upgrade your GPU to the best possible 😉

7. Make best use of the CPU cores by running an affinity mask (search @SteveWfor guru level explanations about why this important point makes such a difference)

[Christopher Low 5,332]

LOL @ No.6

[KFXE 13]

15 hours ago, w6kd said:

In default (DX9) FSX, most of the work to compose a frame is done in the CPU.  The GPU does some finishing work, such as anti-aliasing, but the CPU has to first layer and place all of the draw calls for each object in the frame in the proper order.

If you use the incomplete DX10 "preview" mode in FSX along with the appropriate utilities like Steve Parson's DX10 Fixer, then more of the frame composition work can be offloaded to the GPU.  For example duplicated objects like trees and clumps of grass need only be passed to the GPU once with multiple draw calls referencing the same object.  In this case, a more powerful GPU does help some.  P3D makes much better use of the GPU using DX11 and 64-bit driver code.

CPU throughput depends on clock speed, a metric we call IPC (instructions-per-clock) which is the efficiency with which the CPU can perform instructions, and RAM latency (the speed at which RAM can answer a read/write request--a function of memory clock speed and memory latency).  So both CPU speed and RAM latency matter, and throughput is a balance of the two.  If the RAM is very slow, then the CPU compensates for that by inserting wait cycles during which a particular instruction in the pipeline does nothing as it sits idle waiting for the amount of time it takes for the memory subsystem to respond to a read request.  It's actually somewhat more complex still, because the CPU pipelines multiple instructions at once, so for each clock step there can be a bunch of instructions being processed at once, reducing the effect of waiting on RAM, and the CPU also uses a very fast block of internal cache memory, so that whenever it issues a read request it actually gets a small block of RAM and puts it into its cache, which speeds things up as long as the program doesn't branch elsewhere and have to discard/overwrite the pre-cached memory that it read.

So the bottom line is that CPU clock speed and IPC generally make more of a difference than RAM latency, but only to a point.  Both matter.

Regards

 

Awesome explanation. Thanks for taking the time to write that.

Your honor, the defense has no further questions regarding the CPU and GPU. 😎

[KFXE 13]

17 hours ago, Bert Pieke said:

I also agree that your RAM is probably just fine..

 

Except that the latency is horrible and I can't fix it with BIOS tweaking.  On Machine ONE, the Corsair LPX ECC RAM appears to be the bottleneck according to UserBenchMark.

Just for fun, I ordered 2 x 8GB sticks of NON ECC, UNbuffered RAM for Machine ONE.  I'll let you know how it BenchMarks after I receive it and install it.  I can't think of anything more to try and UserBenchMark tests say the Corsair LPX ECC RAM I have now is horribly latent.  My Motherboard specifically calls for NON ECC RAM.  My mistake when I bought it.  Although I've read that in many (but not all) cases the board will simply ignore any extra bits.

Thanks and Happy New Year!

 

Edited December 27, 2019 by KFXE

[KFXE 13]

On 12/26/2019 at 11:49 AM, Luke said:

No, you've been running the wrong CPU. An A10 APU is what you buy grandma to surf Amazon and play Candy Crush.

Overclocking it is unlikely to make any significant difference. If you want AMD, get a Ryzen 2.

Cheers!

 

Would I need to go any further than a 2nd Gen Ryzen 7  2700X - 4.3GHz  CPU ?

[KFXE 13]

I'm wondering if the benefits outweigh the cost difference?   They seem awfully close in performance.

My current processor (This Processor, Navy Blue)compared to two different Ryzen CPU's.  Of course, either Ryzen will blow away my current CPU no doubt.

The prices are commonly found prices on ebay for these previously owned processors .

Even the Ryzen 7 3700X is only a VERY small bit better than the 2700X according to CPU-Z  Only the Ryzen Thread Ripper is significantly better.

What I'm saying is why go for the more expensive $300+ 3700X if it's only slightly better than the $120 1700X ?

Edited December 27, 2019 by KFXE