AMD sketches out the next five years

[DanceswithUnix]

But now it's nearly build time for me, I find it hard to pick the Amd cpu and board (though I'll still be getting an Amd gpu), no matter how much I prefer to support them. Hate to say it, but that's the situation.

I know what you mean, if you are doing a high end build then AMD haven't really left you any choice.

I got my current motherboard & CPU 2 years ago and it was very competitive back then. Would I get them now? Dunno tbh. CPU performance has been going up so slowly that I don't think there is anything worth upgrading to yet but the price of getting an i5 has dropped a bit as have the Intel motherboards.

My setup uses:
£70 motherboard http://www.amazon.co.uk/M5A97-EVO-R2.../dp/B008RPZ5H8
£127 Fx8350 http://www.amazon.co.uk/AMD-FX8350-P.../dp/B009O7YUF6

I think 2 years on it is still the sweet spot for price/performance so if that doesn't tempt you then you probably do want to go the i7 route.

[ZaO]

I know what you mean, if you are doing a high end build then AMD haven't really left you any choice.

I got my current motherboard & CPU 2 years ago and it was very competitive back then. Would I get them now? Dunno tbh. CPU performance has been going up so slowly that I don't think there is anything worth upgrading to yet but the price of getting an i5 has dropped a bit as have the Intel motherboards.

My setup uses:
£70 motherboard http://www.amazon.co.uk/M5A97-EVO-R2.../dp/B008RPZ5H8
£127 Fx8350 http://www.amazon.co.uk/AMD-FX8350-P.../dp/B009O7YUF6

I think 2 years on it is still the sweet spot for price/performance so if that doesn't tempt you then you probably do want to go the i7 route.

Yeh it's looking like it's gonna be Intel again for me this time. My last cpu was the 2500K (I'd stick to that for now if I still had it), then a 955 before that. The 955 was really starting to struggle for my use around late 2011 (It was Skyrim and Bf3 that made me upgrade it), which is why I had to go for the 2500K. I'm not sure the FX chips were even out then.. Either that or I somehow didn't know about them, otherwise I probably would've got one to drop straight into the setup I had at the time.

I usually seem to upgrade cpu and gpu every couple of years (depending on progression of games and hardware). Maybe not at the exact same time as eachother, but usually within about 3-6 months. I often find it a good choice to wait for a new gen of something to come out, then get something high end from the last gen as the remaining stock is being sold off a bit cheaper. If the power requirements and price of the top of the line FX cpu were a bit lower, I'd consider getting an Am3 board and chucking one in there to serve me for a year or two. But it's not worth it imo. Having to buy a more powerful Psu also adds to the cost of the FX cpu's.

I do plan on getting a 290X though. Hopefully there'll still be some good ones left for sale when I'm finally able to buy. I'm also interested to see what the new amd gpu's will be like. I'm just trying to ignore rumours and wait for the real info Hopefully we'll get a proper release date for them soon..

[watercooled]

Yeah previous-gen flagships are often a good place to look for GPUs, just before the stock dries up.

WRT to power use though, you really shouldn't need to choose a different PSU because of an FX CPU; I've seen a few people claim that but it tends to be massively over-exaggerated, as do a lot of power claims when it comes to computing. Most single-GPU systems really, really don't need anything more than a ~500W PSU tops, and you're not going to save a lot (if anything) by getting something <500W.

[ZaO]

Yeah previous-gen flagships are often a good place to look for GPUs, just before the stock dries up.

WRT to power use though, you really shouldn't need to choose a different PSU because of an FX CPU; I've seen a few people claim that but it tends to be massively over-exaggerated, as do a lot of power claims when it comes to computing. Most single-GPU systems really, really don't need anything more than a ~500W PSU tops, and you're not going to save a lot (if anything) by getting something <500W.

The FX 9590 uses way more power than its rival, the 4690K (136W more I think). And the price of a psu could be quite significant if the FX cpu forces you to go from something like an 850W to a 1000W. Whatever extra it costs for a psu, that's an extra cost because of the cpu. Then there's the electricity costs.

[watercooled]

The 9000 series is an outlier - they're essentially just very high-leakage binned 8000 series aimed at overclocking. AMD marketing may occasionally push them as normal desktop parts but I really don't agree with that and wouldn't recommend them for stock clocks. Just stick to the 8000 series, the increase in stock clock speed doesn't nearly offset the jump in power consumption.

From what I've seen, the 4790k uses significantly more power than the 4770k for similar reasons. Higher leakage is often desirable for overclocking, but not for efficiency.

Electricity costs are another thing which tend to get exaggerated, assuming people have their CPU fully loaded 24/7 with Linpack which just isn't true for the vast majority of desktops.

[ZaO]

The 9000 series is an outlier - they're essentially just very high-leakage binned 8000 series aimed at overclocking. AMD marketing may occasionally push them as normal desktop parts but I really don't agree with that and wouldn't recommend them for stock clocks. Just stick to the 8000 series, the increase in stock clock speed doesn't nearly offset the jump in power consumption.

From what I've seen, the 4790k uses significantly more power than the 4770k for similar reasons. Higher leakage is often desirable for overclocking, but not for efficiency.

Electricity costs are another thing which tend to get exaggerated, assuming people have their CPU fully loaded 24/7 with Linpack which just isn't true for the vast majority of desktops.

Yeh, but if you're looking for top performance as I am, that's the one you buy. And remember, the 4690K is the competitor for price and performance, not the 4790K. The power consumption difference is pretty high, especially as the FX 9590 also requires a water cooler on top of the already much higher power requirements without that. I can't be bothered to try and figure out exactly what the cost to electricity would be But, price, performance and power usage, plus the motherboard chipset used - Intel wins by enough to choose them, for me anyway.

Gaming will be pushing the the cpu hard. Not as much as Linpack, for sure, but still enough to be of concern.

As I say - I definitely prefer to support Amd over Intel and Nvidia. But it's not always a great choice when it comes to the cpu's. At least not since Sandy Bridge arrived. Let's hope Amd come up with something good soon

[watercooled]

I'm not disputing your choice, and of course it depends on what you're doing, however I'd find it hard to recommend a 9000 over an 8350. Look at benchmarks of the two; besides some slightly higher numbers here and there to brag about, the difference between the two is often quite negligible. And it would be even harder to tell the two apart in actual use. Don't forget they're exactly the same Vishera processor, just different bins and clock speeds.

It's not just AMD I'm picking on there though, for the same reason I'd generally not recommend a 4790k over a 4790 unless overclocking was a definite requirement; the extra cost and power consumption are unnecessary.

And same again for microscopic factory overclocks on GPUs; I'm sure that 1020MHz (vs 1000MHz stock) on my 280X makes a huge difference. Although in that case there's no obvious penalty; just I wouldn't really pay much attention to it when comparing cards.

[ZaO]

I'm not disputing your choice, and of course it depends on what you're doing, however I'd find it hard to recommend a 9000 over an 8350. Look at benchmarks of the two; besides some slightly higher numbers here and there to brag about, the difference between the two is often quite negligible. And it would be even harder to tell the two apart in actual use. Don't forget they're exactly the same Vishera processor, just different bins and clock speeds.

It's not just AMD I'm picking on there though, for the same reason I'd generally not recommend a 4790k over a 4790 unless overclocking was a definite requirement; the extra cost and power consumption are unnecessary.

And same again for microscopic factory overclocks on GPUs; I'm sure that 1020MHz (vs 1000MHz stock) on my 280X makes a huge difference. Although in that case there's no obvious penalty; just I wouldn't really pay much attention to it when comparing cards.

The FX 9590 is the competitor to the 4690/4690K for price and performance out of the box. That's why I'm comparing those two. I would buy the 9590 if I were getting an FX. I can't be bothered to save a very tiny amount (potentially nothing) on a 8350, when I would just have to buy a water cooler when I overclock it anyway (I also have the hassle of overclocking). I could just spend the same money and get a 9590, or get the 4690 with better performance in general, lower power requirements and a better chipset on a motherboard that has an upgrade path. And yes, the 5ghz compared to the 4.2ghz will make a big difference in games that are cpu heavy! It's not a small overclock at all..

[watercooled]

Not from what I've seen e.g. http://www.anandtech.com/show/8316/a...reme9-review/8
http://www.hardwarecanucks.com/forum...r-5ghz-13.html
http://www.xbitlabs.com/articles/cpu...0_5.html#sect0

Next to none of them are close to showing a perceptible difference.

And F1 is a particularly single-thread-heavy game. But like I said it depends on the exact games/applications you're running.

[ZaO]

Not from what I've seen e.g. http://www.anandtech.com/show/8316/a...reme9-review/8
http://www.hardwarecanucks.com/forum...r-5ghz-13.html
http://www.xbitlabs.com/articles/cpu...0_5.html#sect0

Next to none of them are close to showing a perceptible difference.

And F1 is a particularly single-thread-heavy game. But like I said it depends on the exact games/applications you're running.

Ok, so just comparing the 8350 and 9590 here.. The benchmarks you've linked do show some good differences. 5-10 fps is a good improvement in my book. People including myself will definitely overclock a cpu for that difference! If it's the difference between 50 and 60fps, or holding 60fps solid rather than too inconsistent, I'm in. Besides, those games are not depending on the cpu as much as others. Even with the games in those benches, change the settings or hardware setup, and it could look quite different. Without even searching for benchmarks, I think Arma 3 should show some good gains from the extra speed. But I agree - the difference you see depends on what programs you're running, of course...

I don't think either of us need to prove the other wrong or anything! I'm just saying - if I were to buy the 8350, I'd overclock it anyway, so I might as well just buy the 9590 and have the job done for me with a water cooler already included. The price would be very similar! But as it happens, I'm getting an Intel cpu as I think it's currently the best choice. Here's to hoping Amd comes out with something good in the near future

[watercooled]

That's the thing though - 10fps at 120fps means a whole lot less than it would at 30fps. With very few to no exceptions, the differences in the reviews I linked are not that significant. F1 is a fairly CPU-dependent game and not very well threaded it seems, hence i3>Vishera.

E.g. I'd like to see anyone honestly tell the difference between:
93.9>99.4
136.2>141.4
97>98.5
etc

Even a few of the bigger (in percentage terms) differences like
44.2>47 are fairly small (2.8fps), and interestingly both are higher than the 4770k (CoH2 Anandtech) anyway.

Like you say if it was 10fps difference at below 60fps then it could be worthwhile, but we're not seeing anything like that.

Likewise I'm not trying to disprove anything, just showing why I'd nearly always go with an 8350 over a 9590, with one of the only exceptions being hobby overclocking.

Depending on how exactly they're binned, it's possible that a 8350 OC'd to 9590 clocks could use less power than the actual 9590, but I'm just extrapolating and can't seem to find any results after a quick search. It would be interesting to see though so I'll post back if I find something.

Edit: I agree with this, The Stilt being a well-known overclocker/modder. http://www.xtremesystems.org/forums/...=1#post5235195

[ZaO]

That's the thing though - 10fps at 120fps means a whole lot less than it would at 30fps. With very few to no exceptions, the differences in the reviews I linked are not that significant. F1 is a fairly CPU-dependent game and not very well threaded it seems, hence i3>Vishera.

E.g. I'd like to see anyone honestly tell the difference between:
93.9>99.4
136.2>141.4
97>98.5
etc

Even a few of the bigger (in percentage terms) differences like
44.2>47 are fairly small (2.8fps), and interestingly both are higher than the 4770k (CoH2 Anandtech) anyway.

Like you say if it was 10fps difference at below 60fps then it could be worthwhile, but we're not seeing anything like that.

Likewise I'm not trying to disprove anything, just showing why I'd nearly always go with an 8350 over a 9590, with one of the only exceptions being hobby overclocking.

Depending on how exactly they're binned, it's possible that a 8350 OC'd to 9590 clocks could use less power than the actual 9590, but I'm just extrapolating and can't seem to find any results after a quick search. It would be interesting to see though so I'll post back if I find something.

Edit: I agree with this, The Stilt being a well-known overclocker/modder. http://www.xtremesystems.org/forums/...=1#post5235195

Well framerate is a lot to do with consistency. A solid 30fps may feel better than a framerate that is jumping up and down from 40-50 for example. So in those higher numbers with 10fps difference, it could be noticeable depending on the refresh rate with vsync or frame lock. Though tbh, I've always aimed for 60fps with my games. All my monitors have had a 60hz refresh rate max. So I don't really have much experience of playing above 60fps to know how noticeable those jumps would be up there, other than a few times where I turned vsync off just to see how my framerate would be without it. The screen tearing made it too hard to tell anyway lol Also, playing at 60fps is expensive enough as it is!

I get what you're saying. The 8350 might be better when overclocked than the 9590. So yeh, power draw could very possibly/probably be lower. How much lower? I don't know.. Unless it was by quite a large amount, I'd still get the 9590 and save myself the hassle of running many lengthy prime95 tests. God knows how much I hate doing that haha.. But it's got to be done if you want to be stable eh!

[DanceswithUnix]

Well it sounds like GF are ramping up their 14nm process which should help one aspect:

http://www.fudzilla.com/news/process...m-announcement.

[watercooled]

GloFo might be producing some of the 14nm Exynos SoCs. To quote myself:

Chipworks speculate that some of the 14nm Exynos wafers might be from GloFo: http://www.chipworks.com/en/technica...ung-galaxy-s6/

Interesting if true, as it would mean they're fairly close in their 14nm ramp.

[DanceswithUnix]

GloFo might be producing some of the 14nm Exynos SoCs. To quote myself:

Well aren't Apple tying up a lot of Samsung's fab capacity? That would make a lot of sense.

[watercooled]

It's just interesting if true as it presumably means there's some decent compatibility between the processes at Samsung and GloFo to make porting less difficult.

Plus shipping processors on GloFo 14nm would be something more solid than company announcements too. I'm quite surprised by how quickly 14nm has appeared after 20nm.

You often get the 'but Intel is more than a full node ahead' line repeated, however GloFo/Samsung 20nm is already far denser than Intel's 22nm (Intel 22nm is actually closer to GF/S 28nm). The 14nm is less dense than Intel's as we already know, but it's nowhere near a full node behind as some would have you believe. https://www.semiwiki.com/forum/conte...lead-10nm.html
And of course metal layer density is only one part of the story. At the end of the day these ..nm numbers are pretty much just company-specific names for the nodes.