R9 270 slower than 9800GT - new card in legacy system

[zelcs]

Disclaimer: I do know what CPU bottle-necking is, and this is not supposed to be, primarily, a gaming setup.

So I replaced 9800GT with R9 270 (Gigabyte) in my old AM2 board (Gigabyte M57SLI-S4, nVidia nForce 570 chipset) with a Brisbane Athlon X2 5000+, and I am getting low GPU usage and framerate in games.

For example, in 3DMark 06 I got about 20% GPU usage with about 85-95% CPU usage.

3DMark 2006 results were worse overall:

9800GT / R9 270
SM2.0: 4635 / 3937
SM3.0: 5067 / 5621
Overall: 10336 / 9620

In 3DMark 2013, Ice Storm result was worse as well:

9800GT / R9 270
Ice Storm graphics: 63K / 61K
Cloud Gate graphics: 7.5K / 29.5K
Fire Strike graphics: n.a. / 5.4K

These are all pure graphics tests, so I cannot comprehend how a slow CPU/RAM/system could possibly prevent the card from loading the level in it's 2GB memory and rendering it as fast as possible.

Or are there geometry calculations still depending on CPU even for these "pure graphics" tests? Otherwise, how can one explain why an R9 270 can hit lows of 18 - 20 fps in 3DMark 2006 "Return to Proxyconn" and not go past ~25% GPU usage.

Heat is no issue and power, I am pretty sure, isn't either, as I have been running a pair of these 270s in this system (Corsair VS650watt) at maximum utilization and nonstop day and night. (Don't ask me why, you probably know already).

It just feels like there is some software limitation there. Something in the drivers that perhaps relies on some newer CPU instructions or higher system/PCI-E BUS speed.

It feels like these modern day drivers are not optimized for something like Athlon 64, which probably contributes to it more than the fact that Athlon X2 system does have a pretty obvious difficulty with feeding the card with data fast enough.

What is your experience with old system + new card situations? Perhaps I should try more ancient drivers? How far in the past can I go as far as drivers?

 

[SPBHM]

3dmark vantage is less CPU limited and can run on both cards,

 

[SiliconWars]

Or are there geometry calculations still depending on CPU even for these "pure graphics" tests? Otherwise, how can one explain why an R9 270 can hit lows of 18 - 20 fps in 3DMark 2006 "Return to Proxyconn" and not go past ~25% GPU usage.

Return to Proxyconn was actually really heavy on the CPU (the 18-20 fps part would be the bit where the camera sweeps around the legs on a machine?). Firefly forest and the flying ship one were much better tests of a GPU.

 

[Termie]

A quick look at the only test that can run on both cards that actually matters, Cloud Gate, demonstrates that the 270 is 4.5x faster.

Other than that, you are completely and totally CPU-limited, and yes, it affects all the tests you've mentioned. The tests aren't even remotely "pure graphics" tests, not when you're running them on an ancient CPU and a brand-new GPU. And the CPU burden of AMD's drivers is greater, which is why your performance is lower in some tests. Your CPU dates back to 2005, so none of this is all that surprising.

 

[zelcs]

Sub 20s in Proxy was when that steam tank fell and the door opened and a bunch of new troopers started firing at the camera.

Firefly ran at pathetic ~30 - 35 fps average, breaking into high 20s at times. Both Firefly and Proxyconn are supposed to be "Shader Model 2.0" tests and that's where 270 lost to 9800 and made the overall score lower as well.

SM3.0 test was no way near the potential of a 2.8 billion transistor card either... Flying ship test was best at 70+ avg fps, but Deep Freeze was rather pathetic at 45 - 49 fps average.

And how can 3DMark 2013 Ice Storm SM2.0 test be so bad is beyond me.

I have no 9800GT datapoint for Vantage as I no longer own that card. I will run Vantage on 270 and report back.

 

[beginner99]

Other than that, you are completely and totally CPU-limited, and yes, it affects all the tests you've mentioned. The tests aren't even remotely "pure graphics" tests, not when you're running them on an ancient CPU and a brand-new GPU. And the CPU burden of AMD's drivers is greater, which is why your performance is lower in some tests. You CPU dates back to 2005, so none of this is all that surprising.

this

 

[DigDog]

I guess you probably dont want to hear this, but ..
http://pcpartpicker.com/p/2GQ49

 

[zelcs]

3DMark Vantage results:

Overall: 8691
Graphics: 16488
CPU: 3593

 

[SiliconWars]

3DMark Vantage results:

Overall: 8691
Graphics: 16488
CPU: 3593

http://uk.hardware.info/reviews/4976/5/amd-radeon-r9-270-review-3dmark-vantage

29,268. That's how much the CPU is bottlenecking the GPU, even in Vantage.

 

[zelcs]

I guess you probably dont want to hear this, but ..
http://pcpartpicker.com/p/2GQ49

Not as cheap in Latvia. The cheapest counterparts I could find in cheapest shops:

34.90 € - INTEL Celeron G1820 2,70GHz LGA1150 BOX
48.73 € - MSI B85M-P32 m-ATX LGA1150
28.11€ - Crucial 4GB DIMM DDR3 1600MHz

Total: 111.73 EUR / 153.23 USD

Besides, I woud want at least two PCI-E 16x slots and Crossfire.

And my plan was to wait until Broadwell... I may not make it.....

 

[BrightCandle]

We don't actually know anything about broadwell performance because by current accounts it seems to be about a year away. But if its anything like the improvements we have been seeing with SB -> IB then its at most going to bring about 10% performance, and in most games that really isn't going to matter very much especially since your going to be GPU limited with an R270. But that X2 is pretty old and slow now.

 

[Insert_Nickname]

Not as cheap in Latvia. The cheapest counterparts I could find in cheapest shops:

34.90 € - INTEL Celeron G1820 2,70GHz LGA1150 BOX
48.73 € - MSI B85M-P32 m-ATX LGA1150
28.11€ - Crucial 4GB DIMM DDR3 1600MHz

Total: 111.73 EUR / 153.23 USD

That alone represents a huge upgrade from what you have already. I'd actually go so far as to say that's a bargain...

Besides, I woud want at least two PCI-E 16x slots and Crossfire.

And my plan was to wait until Broadwell... I may not make it.....

Don't bother crossfire-ing lower-end cards. You be much better of moving to the next tier (280-280X) graphics card or even the tier after (290-290X) before you should be considering crossfire. Especially with such a CPU. You be massively CPU bound even with the single 270 using a Celeron, and that's not even mentioning the potential driver issues with crossfire. Everything is easier with a single card setup...

If you're waiting for Broadwell, you might as well wait a little longer for Skylake and get the completely new platform as a bonus. We don't even know if Broadwell will be released on the desktop currently.

 

[SPBHM]

3DMark Vantage results:

Overall: 8691
Graphics: 16488
CPU: 3593

graphics for a 9800GT is around 6000 I think, so yes... already seeing a big difference, but 3600 for the CPU is extremely low, you would be better with a 7770 and a fast CPU.

 

[Spjut]

You could try finding a cheap Phenom II X4 CPU. Your motherboard can take some of them if you update the BIOS.
Just make sure you get the correct revision
http://www.gigabyte.com/products/product-page.aspx?pid=2287#dl

Is it Windows Vista?
I do think it's odd that the performance actually is worse for you. Having a CPU bottleneck usually means that you can increase the resolution, MSAA etc without the performance dropping by much.

 

[el etro]

Your motherboard is not PCi-Express v2? My P5vd2-vm with a C2D at 1.8Ghz hanged my GTS 450 in the past. Until i trade the system for a FX-4100, everything become ok.

Would be good to try a Athlon X2 7750 overclocked in this rig to test if the bottleneck is so great like this.

1) These are all pure graphics tests, so I cannot comprehend how a slow CPU/RAM/system could possibly prevent the card from loading the level in it's 2GB memory and rendering it as fast as possible.


2)It just feels like there is some software limitation there. Something in the drivers that perhaps relies on some newer CPU instructions or higher system/PCI-E BUS speed.


3)It feels like these modern day drivers are not optimized for something like Athlon 64, which probably contributes to it more than the fact that Athlon X2 system does have a pretty obvious difficulty with feeding the card with data fast enough.


4)Perhaps I should try more ancient drivers? How far in the past can I go as far as drivers?

1) Because it still relies on CPU. Is like try to test a GTX 780Ti Lightning at 4k with a Pentium G620...

2) This is the question. but newer gaming coding may require CPU instructions that your 5200+ don't have.

3) Lets state one thing: Athlon x2 5000+ performance is out of any standard(even NUC processors performance standard) nowdays.

4)IMO will not solve anything.

 

[KingFatty]

...with about 85-95% CPU usage.

I think that's your problem right there. If you looked deeper, it's possible one of the CPU cores is fully 100% too.

As someone else mentioned, it's likely that the difference in scores between the two video cards can be explained by various overhead of the GPU drivers that is assigned to the CPU. Normally the CPU can easily handle it, but you are so CPU limited/saturated/bottlenecked that the computer can't get it done, so the weaker GPU with weaker need for CPU help comes out ahead.

 

[zelcs]

In BF3 my CPU useage is pegged 100% both cores and constantly.

In newer games like BF3 or Planetside2 I do actually feel increase in FPS, especially when i crank up AA, texture size etc.

But the increaase is still marginal, considering the huge difference in pure compute power between the two cards.

What surprised me was the actual DOWNGRADE in DX9 tests like 3DMark 13 Ice Storm and 3DMark 06 SM2.0. I expected the new GPU to "take off" in those kind of tests and not be CPU bound at all...

About Phenom upgrade:

I do have the latest BIOS on my board and I have checked Gigabytes site and it does list certain Phenom II quads on supported CPU list, although it warns that "If you install AMD AM3/AM2+ CPU on AM2 motherbord, the system bus speed will downgrade from HT3.0(5200MHz) to HT1.0(2000 MT/s".

No idea how that HT speed would affect gaming. I also don't know if such upgrade would give the necessary X2 X3 X4 fps increase in most games.

Locally there are pretty much no decent Phenoms for sale for decnet price. Ebay prices are better, but still quite high and would require 15 euro shipping, which is so hard to justify for what is still a piece of old electronics.

Couple of shops locally offer "Phenom II X2 521 Dual Core, 3,5GHz, Socket AM3" for ~40 euro.

Would that be 40 euro well spent or would I just make a fool of myself?

Edit: Turns out those X2 Phenoms are X4s with one half disabled... That's pretty stupid, Athlon II makes more sense.

 

[SPBHM]

I think going for anything lower than a Phenom II X4 945/955 (95w) which your board supports with the latest bios, is not worth it, try finding used on local forums, ebay or something.

otherwise you could buy something newer, like an A55 board + X4 750K, or Core i3 + h61/h81 and so on...

 

[2is]

R9 270 on an Athlon X2? I have a couple X2's laying around and they can't even play the last several revisions of COD with acceptable performance. I'll bet about everything you do will be CPU bound with that processor.

 

[Slomo4shO]

Since no one else bothered to ask, run BF 4 on both cards. Use mantle and document your results

 

[ZGR]

That is one slow CPU. My old build has an Athlon X2 and it at least 1 core bottlenecks in a ton of basic tasks.

Time for a new motherboard and CPU!

 

[zelcs]

Used Phenom IIs are ridiculously pricey.

So, in an act of desperation, I swapped my Athlon X2 with the Phenom II 910E I had in my webserver, and, let me tell you, it made all the difference in the world!

I am running the 910E at 3.2 Ghz and here are results before and after:

3DMark 2006
Athlon X2 2.6 / Phenom II 3.2 Ghz
SM2.0: 4635 / 6825
SM3.0: 5067 / 9594
Overall: 9620 / 18307

3DMark 2013
Ice Storm graphics: 61K / 142,5K
Cloud Gate graphics: 29.5K / 37.6K
Fire Strike graphics: 5.4K / 5.6K

Battlefield 3 and Planetside 2 goes from competitively-unplayable to no lag at all and good fps throughout at Ultra settings.

So what have I (or we) learned today?
1.) AMD Phenom II is faster than Athlon X2 5000+.
2.) Driving graphics card APIs still requires substantial work from CPU
3.) Modern games are properly multithreaded
4.) Motherboard (as CPU, RAM and GPU interconnect) is not THAT important. And so isn't the speed of the RAM (DDR2 vs DDR3).
5.) Modern graphics card drivers may relay on some newer CPU instruction set additions or architectural characteristics

P.S. The Athlon X2 is just about enough to fufill my server duties until I buy a brand new Intel Broadwell or Skylake system for my desktop.

 

[SPBHM]

funny outcome, so you had a Phenom II X4 doing file server work and was trying to play BF3 with a dual core k8?

but the difference is what you could expect,

 

[zelcs]

The 910E was doing much more than just fileserving. It was running one VM with websites (with solid amount of traffic) and another for a few gameservers.

So it wasn't a foregone conclusion that the Athlon K8 X2 would actually be good enough.

I haven't checked the gameservers but website seems to respond OK. But load averages are just about on the limit

Having bought a couple of R9 270, I realized that the place I needed processing power most atm is home (with all the newer games).

910E will return to server work when I have a new desktop build.

I know I am going offtopic, but I have to say that the 910E has been a splendid server CPU. Even with chepo consumer mobo without ECC ram or anything, the reliability has been excellent during the past 1,5 years I have been webserving from this box.

And it's just 65 watts at default 2.6 Ghz!

 

[tweakboy]

A 9800GT owned and still owns ? Don't be ridiculous. Games will ruln in slow motion. gl