AMD - Piledriver chitchat

[AETAaAS]

You mean like this one? http://www.scan.co.uk/products/gigab...-dvi-hdmi-mitx

But yeah, this late in the game I would wait for Kaveri

Sigh. I guess it's a just a little more waiting for the right CPU... Maybe more motherboards will come out in the meanwhile. For a fair period, it was only AsRock with news of an FM2+ miniITX part...

I having a feeling its more the case,here is a £280 to £300 worth of competitors products,which are slower in the game we run.

My view was that AMD primarily wanted to compare with the 630. The 4770k was there to say; "if you think we are bottlenecking the graphic card, you are insane".

[Biscuit]

My view was that AMD primarily wanted to compare with the 630. The 4770k was there to say; "if you think we are bottlenecking the graphic card, you are insane".

That makes more sense. Still doesn't prove a lot though :-\
No one buys a 630 with serious hope of gaming

[kalniel]

That makes more sense. Still doesn't prove a lot though :-\
No one buys a 630 with serious hope of gaming

Nor, as yet, an APU I presume.

[Biscuit]

Nor, as yet, an APU I presume.

Well there is a difference between seriously considering gaming and gaming seriously. Neither apply to the 630 however an APU can be gamed on at a low to medium level

[kompukare]

Probably truer to say it's an embedded product, rather than mobile. Intel list it as desktop and it's got a 65W TDP - not very mobile friendly

Sure. Or that it is even truest that those two parts (i5-4570R, i5-4670R) are iNtel Apple parts. But then I guess Iris Pro (and to a lesser extend HD3000, HD4000 and HD4600) are mostly something Intel is doing because of Apple.

My view was that AMD primarily wanted to compare with the 630. The 4770k was there to say; "if you think we are [CPU] bottlenecking the graphic card, you are insane".

I think you are spot on there. But of course with multiplayer games like this and SC2 etc., the actual map makes a huge difference. Benchmark built into games rarely show actual multiplayer gaming. That's why I don't have a problem with those 640x480 CPU scaling benches - for most things they are silly (FPSs into the 100s etc.) but for benching CPUs they are handy.

[Terbinator]

That's why I don't have a problem with those 640x480 CPU scaling benches - for most things they are silly (FPSs into the 100s etc.) but for benching CPUs they are handy.

But they are still silly as it's a ridiculous setting no one uses.

[kalniel]

But they are still silly as it's a ridiculous setting no one uses.

Well only silly if you are claiming it's a representative benchmark of someone's likely experience. But they're usually not, they're usually benchmarks of CPU performance and to assess that you have to remove other bottlenecks.

[CAT-THE-FIFTH]

The issue is that running a benchmark at low resolution and low settings,is problematic in a few areas IMHO:
1.)The CPU loading profile itself will probably be different,and you might find yourself hitting an artificial bottleneck. This would be a combination of the game engine and the drivers. Even in a game some settings are CPU heavier,so running at lower settings will reduce this load.
2.)It does not indicate what the actual bottleneck in the game is,ie,GPU or CPU or both at normal settings. How long will it take for modern graphics card at 1920X1080 to show the CPU bottleneck?? If it is GPU limited at 1920X1080,how much more GPU power would you need to see a bottleneck at that resolution and how many years??
3.)By the time the CPU might be a bottleneck in a game,further patches or expansions might load the CPU differently.

The problem is that it leads to people spending way too much on CPUs,when they really would be better served by getting a cheaper CPU and spending the rest on a better graphics card. It even leads to situations where people make upgrades from good enough CPUs.

I would rather look at benchmarks at the normal resolutions of 1680X1050,1920X1080/1200 and 2560X1440/1600 to see what is more important with a game.

[kalniel]

But CPU benchmarks aren't about how you experience the game, they're about testing different CPUs.

You wouldn't normally test a Ferrari, McLaren and Porsche in a town with a 30mph speed limit, despite the fact they're going to be used in towns a lot of time. You want to know what the differences in performance are.

[CAT-THE-FIFTH]

But CPU benchmarks aren't about how you experience the game, they're about testing different CPUs.

You wouldn't normally test a Ferrari, McLaren and Porsche in a town with a 30mph speed limit, despite the fact they're going to be used in towns a lot of time. You want to know what the differences in performance are.

People are running their games at 1680X1050,1920X1080/1200 and 2560X1440/1600 not 640X480. If the realworld loading is different than artificially low settings,it tells you NO useful information.

People are reading those benchmarks at 640X480 to infer 1680X1050,1920X1080/1200 and 2560X1440/1600 performance not 640X480 performance.

The CPU loading profile is going to differ between higher and lower resolutions and different game settings,as the drivers and game engine will be not doing exactly the same thing. The bottlenecks are different.

How many reviews bother to study the CPU loading at different settings and resolutions,down to how many threads are used,etc??

That low resolution and low settings benchmark might show one CPU faster than the other simply because one setting at low might cause less cores to be used for example.

If a game with a high end graphics card is not CPU limited at 1920X1080,then the 640X480 benchmarks are pointless,since the game is GPU limited not CPU limited. You cannot say when the graphics cards will be fast enough to hit the CPU limitation at 1920X1080 as it could be years.

All these low resolution benchmarks have done is make people spend too much money on CPUs,when they are better served buying a faster graphics card.

It also explains why most gamers who don't have high K series Core i5 are still running plenty of games fine,ie,their GPU is the bottleneck not a CPU.

[kalniel]

People are running their games at 1680X1050,1920X1080/1200 and 2560X1440/1600 not 640X480. If the realworld loading is different than artificially low settings,it tells you NO useful information.

It tells you how the CPU performs at that resolution. Nothing more, nothing less. The real world loading difference between reducing GPU demand at low resolution and throwing a bigger GPU at the same resolution is not hugely different - where not GPU limited the CPU chucks out loads of frames and becomes limiting.

People are reading those benchmarks at 640X480 to infer 1680X1050,1920X1080/1200 and 2560X1440/1600 performance not 640X480 performance.

Well that's their own stupid fault then, not the reviewers

[watercooled]

That sort of review is misleading though TBH - not everyone will scrutinise the settings used for testing, or indeed understand the impact they may have on the results.

These reviews are plain assuming CPU performance scales perfectly from these low settings up to realistic settings, and make no effort to prove or demonstrate this. Real-world tests run using real settings, in real games (online or otherwise) have time and time again shown that performance does not scale this way.

They are simply useless and biased unless you happen to be running the latest games at 640x480 on your £1000 rig and want 300fps.

Edit: Oh and an example I've just thought of, since the subject of using these 'benchmarks' to assess how well the CPU will cope under stress in the game. Say something like an i3 scores higher than something like a 6300 at low-res, sitting staring at a wall. Great, with one render thread being about the only thing stressing the CPU, and things like game simulation, physics, net code, and so on, using other threads but sitting nearly idle, the i3 can crank out an extra 10fps. Now add some more players into the game, throw a few grenades, demolish some buildings; those previously near-idle threads suddenly load up and the dual core tanks in fps.

[kompukare]

Well the most detailed of these benches are probably the pcgameshardware.de ones; the GameGPU.ru ones aren't as useful since that site has a fair few CPUs missing.

But when the do, for example, BF4 CPU scaling article they run at 1280x720 (might be the lowest BF4 runs at?) and do various tests like frequency and core count scaling.

However, these tests are not meant to be taken in isolation. To get a good idea of what to buy you really have to look at their GPU scaling plus the CPU scaling tests. Unfortunately, there is no site which allows people to input their monitor resolution, acceptable AA, GPU budget and CPU budget and gives them a custom bench. Which is fair enough: there's simply too much data (10 x CPU + 10 x GPU = 100 tests plus resolutions and settings).

Cat is right of course: if people only casually look at the benches or ask in one of those 'Intel are gods' forums they will end up blowing most of their budget on a CPU. PC gaming enthusiast do love a bit of e-pen and elitism but of course fail to realise if PC gaming looks so expensive to enter into, then there will be less PC games making PC gaming even more expensive etc.

[watercooled]

Oh I'm not saying *all* CPU benchmarks are pointless, just the ones that run at silly settings and assume perfect scaling. GameGPU.ru seem to have most of the market segments covered with CPUs IIRC? They might not have literally every CPU going, but you can often get a fairly accurate estimation of say an i5 2500 of there is a 2500k present.

[kompukare]

GameGPU.ru seem to have most of the market segments covered with CPUs IIRC? They might not have literally every CPU going, but you can often get a fairly accurate estimation of say an i5 2500 of there is a 2500k present.

Last ones I saw had no IB. Something about not buying CPUs but only using what the companies have sent them (guess they never give reviewers samples back). Guess the i5 performance whether SB, IB or Haswell is well enough known though, but i3's in IB/Haswell would be nice to see there too - just in case found a fuse they could leave unblown on those to catch up with the FX6300 for instance; FX6100 to FX6300 is over 30% faster in CoD Ghost after equalising for frequency.

Also, no cheap Intel (Pentiums / Celerons) or FM2 CPUs. Guess Intel and AMD PR don't send those out much either.

[Biscuit]

It tells you how the CPU performs at that resolution.

Which is utterly useless information. Nothing more.
Adding this information into a review is downright irresponsible. It suggests that they are considering as part of their conclusion to the review and that other people should factor it in their buying decision when it offers no real life value.
Gaming benchmarks should be tested in realistic gaming environments, if you want to test a CPU in an environment which is unlimited by the GPU then use a benchmark which will do that without prejudice.