Graphics Cards

[Fizwalker]

I am currently running with an ATI Radeon 7800 with 2048 MB VRAM. Would it be better to get a new card (said new card has to be around $100) or get another of the same card and set it up in `crossfire`?

[LaytheAerospace]

For KSP your video card doesn't matter much. It's a CPU problem, not GPU. That said, Crossfire carries a lot of issues with it. But a $100 card isn't going to be an upgrade over an HD 78xx, either.

I'd recommend you save your money. Dropping $100 on a video card when you already have an HD 78xx isn't going to make much of a difference, if any.

[Elthy]

Dont use crossfire on non highend cards! It wont double the performance! There are lots of issues with Crossfire/SLI which lower the performancegain by quite a bit.

Why do you have problems with your current card? You have a AMD Radeon HD 7850 or 7870, both are capable of running any game at medium-very high settings. You wont get a better GPU for 100$, afaik not even another 7850/7870.

[ZedNova]

Getting a new card won't matter if the bottleneck is your CPU. Your PC needs to be balanced to run well.

If you could list your specs in more detail, it would help people here find what you want easier.

[LaytheAerospace]

Dont use crossfire on non highend cards! It wont double the performance!

Agree. Also, I'd actually advise avoiding Crossfire entirely. I'm using a pair of 7970s right now. Most games work fine. Some go bat.... crazy, especially on a 120 Hz monitor. End up with crazy things like V-Sync thinking it needs to maintain 240 FPS instead of 120 (Skyrim, Dark Souls II), games that load only a black screen (Skyrim), games that crash over and over again (Final Fantasy XIII-2, Universe Sandbox), games that are unplayable due to visual artifacts (To the Moon) and countless other irritations.

If you want to use multiple GPUs, I strongly recommend going with Nvidia. Their drivers are far more stable, and SLI tends to outperform Crossfire, especially in the area of microstuttering (makes the game feel choppy regardless of framerate due to inconsistent frame delays).

[Slam_Jones]

If you really want to upgrade your GPU, I recommend going to the R9 series instead of crossfire. I upgraded from a 6850 to a R9 270 and the difference was impressive... even in KSP.

[Fizwalker]

For most other games, the card is fine. I was tired so I guess I wasn`t thinking quite straight.

My issue is --and it isn't game shattering-- is that with stations frame rate drops noticeably. I don`t really think it is a CPU issue--specs are below--but I could be wrong.

CPU: AMD FX9590

RAM: 16 GB

VidCard: Radeon HD 7850 (I think) with 2048 MB VRAM

Thank you for the replies!

[LaytheAerospace]

My issue is --and it isn't game shattering-- is that with stations frame rate drops noticeably. I don`t really think it is a CPU issue--specs are below--but I could be wrong.

You are wrong. It's a CPU issue. Physics is done on the CPU exclusively. Physics load is a nonlinear function (polynomial, my guess) of the number of parts loaded. Approaching stations causes more parts to load. Few computers can handle more than 1000 parts gracefully. I'm running a 3770k at 4.9 GHz (probably top 0.1% of desktop CPU power) and 1000 part ships are SLOW.

KSP is an extremely CPU heavy game. What's more, it's a class of problem that doesn't multithread. There's nothing that can be done to change this, it's a well known property of systems in which the future state depends on the present state (you must calculate all previous states before trying to calculate a future state). Which brings me to...

CPU: AMD FX9590

This is a problem. AMD chips are much slower than Intel chips at the same clock rate, and the architecture shares floating point units between pairs of cores, so if both cores need to do floating point math, they simply can't and have to stop dead while they wait for the other unrelated workflow to finish. Games, particularly games like KSP, use large amounts of floating point math.

AMD has run a very successful marketing campaign where they've convinced large numbers of people to buy their inferior products, because it has more cores. The problem is, most applications can't do anything with more cores. Applications which scale well with additional cores are fairly rare, and generally Intel chips outperform the AMD chips anyway because of the massive difference in efficiency and the fact that your chip effectively loses half its cores when presented with floating point workloads. It should really tell you something when AMD's top chip claims 8 cores at ~5 GHz and loses benchmark after benchmark to Intel's 4 core 4 GHz chip.

So what can you do? Your best solution is probably to overclock. AMD chips overclock quite well, and gains are fairly linear. Increase your clock speed 25% while holding everything else constant, and you should see about 25% better performance.

Edited January 5, 2015 by LaytheAerospace

[Elthy]

So what can you do? Your best solution is probably to overclock. AMD chips overclock quite well, and gains are fairly linear. Increase your clock speed 25% while holding everything else constant, and you should see about 25% better performance.

An FX 9590 is allready overclocked (its just an FX 8350). With even higher clockspeed you will have even higher energy consumption but the energy needed by the 9590 is allready insane...

[LaytheAerospace]

Didn't know that, haven't been following developments in AMD chips since they got utterly demolished by Sandy Bridge, even though they came to market almost a year later with a product that claimed higher clock speeds and more cores.

For reference, here's a database of single threaded CPU benchmarks. Notice that there are a total of zero AMD chips in the list of previous world record holders. That's right, AMD has NEVER held the top spot in this CPU benchmark, ever.

Here's multithreaded. AMD has some spots here, but only their latest and greatest Opteron, and in every instance an Intel chip regained the throne mere days later. All told, AMD chips have held the crown a combined total of less than a year over the last ten. I'd also like to point out that it only took a pair of Xeons to beat four Opterons.

For comparison, this is the FX 9590 and here's the 8350. My favorite part is how the 8350 does significantly better than the 9590 on a bunch of benchmarks. AMD can't even compete with themselves, it seems.

AMD chips are cheaper for a reason. You get what you pay for.

[Edit] Forgot about this, the KSP CPU performance database.

Edited January 6, 2015 by LaytheAerospace

[cantab]

We can discuss CPUs plenty, but Fizwalker has the processor he has.

On any given system, KSP's performance primarily depends on what you, the player, do. The more parts you put on a rocket the more demanding the game is on your processor, and there comes a point where the processor can't keep up so the game slows down. On one computer that might be a hundred parts, on another it might be five hundred, but even the most powerful PC you can buy will lag if you build a complex enough ship. KSP lags for everyone.

(And some parts have more impact than others. Docking ports included, which space stations tend to be full of.)

So you have two choices. You either put up with the lag or you build to reduce part count. For a space station ways to reduce part count include using fewer, larger modules, placing modules with a tug so you don't need control stuff on every individual module, and eschewing unneeded structural parts.

Edited January 6, 2015 by cantab

[LaytheAerospace]

We can discuss CPUs plenty, but Fizwalker has the processor he has.

I tend to go a little overboard on the AMD bashing, to preemptively stop any pointless debates about whether or not AMD chips are really that much worse than Intel. It's a habit from the Tom's Hardware forums, where there's a nonstop stream of fanboys on either side of any debate.

Anyway, I did a little research on the OP's chip. It seems the 8320, 8350 and 9590 are all the same silicon (fun fact, AMD initially planned to sell the FX 9590 for $800, despite it being the exact same chip they'd been selling for $150 for a year). The only differences is the 8350 and 9590 have been "binned" (AMD pulled the best chips aside to guarantee performance) and run at higher stock settings. The differences in benchmarks are likely due to differences in popularity of the chips. More 8350s means the best 8350 benchmarks are better than the best benchmarks for the other two chips. The best 8320 is, theoretically, identical to the best 8350 and best 9590, due to random chance. However you'll see a higher average on the 9590 because of the binning.

So, you don't have a ton of overclocking headroom, because you're effectively already overclocked from 4.0 to 4.7 GHz, and AMD chips are already running ludicrously hot (220W, compared to 84W for a 4770k). However, you do have a binned chip, so you should be able to take it over 5.0 GHz without too much trouble. Average overclock on an FX 9590 with air cooling, according to HWBot, is 5.135 GHz, or about 10% over stock. It's not much, but if you can manage it without upgrading your cooling system, then it's literally free performance.

I think buying a new CPU cooler for this is probably a waste, though. You just don't have enough headroom to justify investing any money into overclocking. Your next opportunity for significantly better CPU performance is likely going to be the next CPU launch by AMD. Unfortunately, AMD recently announced that they aren't going to have anything for the enthusiast CPU space, and the FX series of processors is at an end. Their focus is increasingly on power efficiency for low cost APUs and mobile.

The power desktop space, I'm afraid, has been completely abandoned by AMD. Really, though, they haven't had anything that could legitimately be called a "top of the line" CPU in a very long time. They're still trying to catch up to the 2500k, three generations later. Those benchmarks are at stock settings, by the way, with the brand new 4.7 GHz 9590 struggling to keep up with the four year old, 3.3 GHz 2500k.

[Camacha]

(fun fact, AMD initially planned to sell the FX 9590 for $800, despite it being the exact same chip they'd been selling for $150 for a year)

You do know Intel does the exact same thing, right? It is called binning, and common across the silicon industry. Sometimes good working chips are purposefully broken (by laser cutting connections on the chip, for instance) to be sold as simpler chips to keep product diversity up, but AMD tends to just leave them as is, so people might get lucky and score a better chip for little money.

Despite - or thanks to - its monster power consumption, the FX 9590 is a more than reasonably performing chip, as it should be on par or beyond a 2500K (which is part of the Sandy Bridge generation you mentioned). That was, and can still be, considered a quick chip. Modern ones are ~15-20% faster, but that is nothing Earth-shattering.

The power desktop space, I'm afraid, has been completely abandoned by AMD. Really, though, they haven't had anything that could legitimately be called a "top of the line" CPU in a very long time. They're still trying to catch up to the 2500k, three generations later. Those benchmarks are at stock settings, by the way, with the brand new 4.7 GHz 9590 struggling to keep up with the four year old, 3.3 GHz 2500k.

By saying that you are ignoring the sad fact that Intel has done little to improve performance those past few generations. They are much more economical when it comes to power, but are only slightly faster on the performance front. It does underline that AMD struggles in another way though, as in a healthy competitive market no technology company would have gotten away with that.

The FX is supposed to return in the somewhat near future, I have been told.

Edited January 7, 2015 by Camacha

[cantab]

It depends on the task of course. Passmark for example puts the FX 9590 and the i7 4790K quite close on overall performance*, but there's a much bigger gulf in single-threaded performance. Generally more tasks will benefit from faster cores (that you get with Intel) than from more "cores" (that you get with AMD); KSP is an extreme example of this.

*The forum throws a wobbly at a link with square brackets in so you'll have to look them up yourself.

[LaytheAerospace]

You do know Intel does the exact same thing, right? It is called binning, and common across the silicon industry.

Binning is one thing. Re-releasing a year old $150 chip as a your new top of the line chip (that you hope to charge unsuspecting customers $800 for) is something entirely different. Intel's top of the line chips are all very much different silicon, and have been for as long as I can remember. And AMD has always been the worst about this, selling low binned quad core chips as tri core with the offending core disabled, throughout the Phenom era.

So don't tell me everyone does this. AMD is alone in what they did with the FX 9590, and have a history of binning more aggressively than their competitors.

Sometimes good working chips are purposefully broken (by laser cutting connections on the chip, for instance) to be sold as simpler chips to keep product diversity up

Not at all applicable in this instance. The chips were already on the market, you can't go back in time and cut them down to differentiate them from your re-release. AMD gets no credit for this in the case of the 9590.

AMD tends to just leave them as is, so people might get lucky and score a better chip for little money.

And sometimes they try to re-release their $150 chip at $800 to trick people into thinking it was better than it really was so they can pay 530% more for the same performance.

I'm not trying to be snarky, I'm just trying to point out that AMD, like every other company on the planet, is motivated by profits. They aren't binning chips so you have a chance of getting a better chip at a discount, they're doing what they think will maximize their return on investment. Attributing it to kindness, or "so people might get lucky" is patently ridiculous. They want you to buy their more expensive, higher profit chips just as much as Intel does.

Despite - or thanks to - its monster power consumption, the FX 9590 is a more than reasonably performing chip, as it should be on par or beyond a 2500K (which is part of the Sandy Bridge generation you mentioned). That was, and can still be, considered a quick chip. Modern ones are ~15-20% faster, but that is nothing Earth-shattering.

Saying AMD's best ever desktop CPU is only on par with a chip Intel released 4 years ago using less than half the power at 2/3 the clock speed with half the cores is faint praise indeed. And the differences may have been a mere 15-20% against a 2500k, but a modern Intel chip is much faster.

http://www.anandtech.com/bench/product/836?vs=1289

http://www.anandtech.com/bench/product/836?vs=288 (in case you meant to claim that the 4770k was only 15-20% faster than a 2500k, which is also not true)

Single threaded CPU benchmarks have the 4770k as much as 100% faster than the 9590, and we're comparing them at stock settings. When the 4770k's much greater overclocking headroom comes into play, the difference is bigger still. And it's not like AMD is winning the multithreaded benchmarks, despite it's 33% clock speed advantage and 100% core count advantage. It should be winning by a minimum of 100% in multithreaded integer workloads, but it's not because its performance is garbage.

By saying that you are ignoring the sad fact that Intel has done little to improve performance those past few generations. They are much more economical when it comes to power, but are only slightly faster on the performance front.

So you're going to tell me that AMD is doing better improving performance by re-releasing their $150 chip as a new high end part a year later? Or that Intel has its enormous performance advantage over AMD because they've advanced performance so much less than AMD did over the same period of time? If not, then this criticism of Intel in defense of AMD doesn't make a lot of sense. You should be criticizing AMD for failing to improve performance at all, not Intel for steady 20% gains one generation after the next.

I'll leave you with a reminder that we've already benchmarked KSP. There is no debate to be had over which chip performs better in KSP. Intel wins by a large margin, even at lower clock speeds.

Edited January 7, 2015 by LaytheAerospace

[ZedNova]

I thought this thread's original subject was graphics cards, not Intel vs AMD CPUs.

[LaytheAerospace]

I thought this thread's original subject was graphics cards, not Intel vs AMD CPUs.

The original subject was asking about improving performance in KSP. The OP made the common mistake of assuming he needed a new GPU to do that (makes sense, it's how you improve performance in 99.9% of games). The GPU issue having been put to rest, the discussion moved on to the component that actually matters for KSP, the CPU.

[GigaG]

I have a Core i5-650 3.2GHz and a Radeon HD 5450 1GB. I have such a crappy GPU. So would KSP be helped by a better GPU if you have a horrid one?

[ZedNova]

I have a Core i5-650 3.2GHz and a Radeon HD 5450 1GB. I have such a crappy GPU. So would KSP be helped by a better GPU if you have a horrid one?

Yes, certainly. Although it won't take a 500 dollar monster to run KSP on full graphic settings.

[GigaG]

^Yeah I have another more graphically demanding program that I'm thinking about getting a new GPU for. What is the best GPU that would balance well with the Core i5-650? Looking for nVidia.

[ZedNova]

^Yeah I have another more graphically demanding program that I'm thinking about getting a new GPU for. What is the best GPU that would balance well with the Core i5-650? Looking for nVidia.

I'd recommend a GTX 750 if your budget is under 200 USD. If you can spend around 220 USD i'd get a GTX 760.

Edited January 8, 2015 by ZedNova

[Camacha]

^Yeah I have another more graphically demanding program that I'm thinking about getting a new GPU for. What is the best GPU that would balance well with the Core i5-650? Looking for nVidia.

Balance is a non-issue. Buy the fastest card your money will get, your CPU will outpace it in most cases anyway. When it is limited to a single thread the CPU might come into play, but that is only the case with a select number of games.

[...]

Look like the habits of the Tom's Hardware boys rubbed off on you

[LaytheAerospace]

Look like the habits of the Tom's Hardware boys rubbed off on you

Meh, it's only fanboyism when you refuse to change your opinions based upon the evidence. I'm just an ....... that's obsessed with CPU power .

For the record I was a huge AMD fan right up to the Phenom generation. I even cheered when they beat Intel to 1.0 GHz, and was horribly disappointed when Intel's first 1.0 chip trashed theirs in benchmarks.

As far as a GPU upgrade, I recommend anyone who's lost start here:

http://www.tomshardware.com/reviews/gaming-graphics-card-review,3107.html

[Fizwalker]

Laythe, while I understand your position on CPUs it's not really relevant to my original post. No one is advocating for AMD or Intel here. I was looking to improve the system I ALREADY have, not buy a new one. No where in your posts do you do that. That said, please keep it relevant to the question I am asking about.

To the other replies that have been on topic, I appreciate them, and thank you! Fewer parts in my stations is is!

[cantab]

750 Ti is a good choice in its price bracket, latest-gen card and very power efficient.