*** AMD "Zen" thread (inc AM4/APU discussion) ***

[Martini1991]

Hence people getting so excited about rumours of AMD producing an APU with HBM.

And when it's out in the flesh and exists then great.

 

[eddyr]

Even DDR4 would be a significant help to AMDs iGPUs.

3Ghz DDR4 would be getting near the performance of the 128Bit GDDR5 on the RX460, certainly 2.5x as fast as 2Ghz DDR3.

Only with quad channel DDR4, which I can't see happening on desktop Zen APU. In dual channel it will realistically be ~50GB/s if the platform spec is 3GHz DDR4.

 

[humbug]

Only with quad channel DDR4, which I can't see happening on desktop Zen APU. In dual channel it will realistically be ~50GB/s if the platform spec is 3GHz DDR4.

I didn't even know they had dual channel DDR4, seems completely pointless, just DDR4 badged DDR3

Does sky lake not have quad channel?

 

[mmj_uk]

lol just admit you were wrong.

 

[humbug]

lol just admit you were wrong.

What?

 

[AllBodies]

We've been seeing over exaggerations to the IGP performance for years now.
The bottleneck has always been the bandwidth.

Yeah, I'd expect it to be more like 750 Ti performance.

I don't think we'll start seeing APUs have properly 'disruptive' performance until they have HBM integrated on the die. And even then, they'll probably cap out at sub-£200 GPU performance.

 

[nashathedog]

I just saw this but I guess it's what you're all talking about, It's a damn shame, I'm currently on a 4790k and it seems that Zen's a bust so I'll wait for the replacement after skylakes tick/tock and go with that

http://www.techspot.com/news/65945-leaked-benchmarks-amd-zen-falls-short-intel-haswell.html

I was really hoping for a decent 8 core gaming cpu in Zen.

 

[SiDeards73]

I just saw this but I guess it's what you're all talking about, It's a damn shame, I'm currently on a 4790k and it seems that Zen's a bust so I'll wait for the replacement after skylakes tick/tock and go with that

http://www.techspot.com/news/65945-leaked-benchmarks-amd-zen-falls-short-intel-haswell.html

I was really hoping for a decent 8 core gaming cpu in Zen.

Nash, thats a rehash of the leaked benchies that have been around for a few days, its taken from an Engineering sample thats been people think has been downclocked for stability while testing etc..

people are expecting the released product to be clocked significantly better, most theories put it at minimum on par with Haswell i believe.

Ontop of that AOTS is apparently not using all cores on a processor of this type, i think i read it wont use more than 6 cores.

Also that whole article is based off a WCCFtech clickbait article, id take it all with a huge pinch of salt, im waiting to see some actual official reviews of the product before i condemn it, im hoping to trade my 4770k in for it, if it matches 4770k perf but i get double the cores its a win in my book, as i also get DDR4 etc.

 

[nashathedog]

Nash, thats a rehash of the leaked benchies that have been around for a few days, its taken from an Engineering sample thats been people think has been downclocked for stability while testing etc..

people are expecting the released product to be clocked significantly better, most theories put it at minimum on par with Haswell i believe.

Ontop of that AOTS is apparently not using all cores on a processor of this type, i think i read it wont use more than 6 cores.

Also that whole article is based off a WCCFtech clickbait article, id take it all with a huge pinch of salt, im waiting to see some actual official reviews of the product before i condemn it, im hoping to trade my 4770k in for it, if it matches 4770k perf but i get double the cores its a win in my book, as i also get DDR4 etc.

I'll make do too if it's clocked so that single thread performance matches my 4790k while having 8 cores and 16 threads plus DDR4 and a few other new platform additions, I just looked in my account and i bought my 4790k in June 2014 so I've been on it for 2 years.
The biggest surprise is that it cost me £269.00 2 years ago as an early adopter (first batch here at OCUK) but today it'd cost me £320

I hope AMD are able to reign in the pricing trend, They'll more than likely go with the flow though

 

[Lokken86]

I'll make do too if it's clocked so that single thread performance matches my 4790k while having 8 cores and 16 threads plus DDR4 and a few other new platform additions, I just looked in my account and i bought my 4790k in June 2014 so I've been on it for 2 years.
The biggest surprise is that it cost me £269.00 2 years ago as an early adopter (first batch here at OCUK) but today it'd cost me £320

I hope AMD are able to reign in the pricing trend, They'll more than likely go with the flow though

Intel domination getting boring isn't it. I probably won't even upgrade even if Zen was better than my 3570k by a long way. The 3570 is just too dam good. 4 years old now I think and still spitting out everything I throw at it. I'll more than likely wait til Zens "tock" and see if that is worthy.
But I definitely would like to try an AMD cpu for once.

 

[humbug]

I just saw this but I guess it's what you're all talking about, It's a damn shame, I'm currently on a 4790k and it seems that Zen's a bust so I'll wait for the replacement after skylakes tick/tock and go with that

http://www.techspot.com/news/65945-leaked-benchmarks-amd-zen-falls-short-intel-haswell.html

I was really hoping for a decent 8 core gaming cpu in Zen.

Wccf and techspot are the dimmest bulb in the draw, always have been, they aren't seeing anything beyond how long the bars on the chart are.

This is an unreleased sample of Zen clocked 25% lower than the Intel in a game I think only scales to 6 or 8 threads. Normalised it's IPC is around that of the Intel.

These people are actually so idiotic as to blatantly contradict themselves, they say it's not good and then go on to say it's much better that their previous generation, if they understood it at a higher level than the length of the bars they would realise that with twice as many threads it's only scoring about 30 higher.
You can't have it both ways, you can't lower IPC and higher performance, it's so obvious it's only using 8 threads, if that.

 

[DragonQ]

Wccf and techspot are the dimmest bulb in the draw, always have been, they aren't seeing anything beyond how long the bars on the chart are.

This is an unreleased sample of Zen clocked 25% lower than the Intel in a game I think only scales to 6 or 8 threads. Normalised it's IPC is around that of the Intel.

These people are actually so idiotic as to blatantly contradict themselves, they say it's not good and then go on to say it's much better that their previous generation, if they understood it at a higher level than the length of the bars they would realise that with twice as many threads it's only scoring about 30 higher.
You can't have it both ways, you can't lower IPC and higher performance, it's so obvious it's only using 8 threads, if that.

Agreed, I couldn't believe how stupid that article was when I read it. How can they conclude it's a "bad start" for Zen and complain that it can't beat Intel's 3 year old Haswell CPU when it's clocked nearly 1 GHz lower?? Surely they must know that 3.2 GHz is not going to be the speed of the fastest Zen chip...

 

[Lokken86]

Agreed, I couldn't believe how stupid that article was when I read it. How can they conclude it's a "bad start" for Zen and complain that it can't beat Intel's 3 year old Haswell CPU when it's clocked nearly 1 GHz lower?? Surely they must know that 3.2 GHz is not going to be the speed of the fastest Zen chip...

It might be 3.2Ghz who knows.

 

[humbug]

Agreed, I couldn't believe how stupid that article was when I read it. How can they conclude it's a "bad start" for Zen and complain that it can't beat Intel's 3 year old Haswell CPU when it's clocked nearly 1 GHz lower?? Surely they must know that 3.2 GHz is not going to be the speed of the fastest Zen chip...

I'm going to look at it in more detail tomorrow, but yes essentially I don't believe 2,8 to 3.2Ghz will be the release speed of Zen, tho it might be for the 16 threader, or it might only get another 200Mhz.
It's the big fat chip of the range, the one sucking up the most power, it's likely to be a bit reigned in to keep it within the 95 watt envelope it look like they set for themselves.
Every chip has a zone beyond which it requires very diminished power to performance returns, 10% more MHz could cost 30‰ more power, on a fat 8 core 16 thread CPU 30% more power could result in 95 watts becoming 120 watts, be that as it may it's better to keep the fat one on a diet.

For the smaller chips like the 6 core you could add another 10% Mhz and keep within the 95 watts, the 4 core perhaps 15% for 85 watts.

So I'm really not worried about the MHz on this chart at all, it's not far off where I would have expected the big flagship chip to be, and it is just a test sample.

16 thread: 3Ghz - 3.2Ghz turbo
12 thread: 3.3 GHz - 3.5Ghz turbo
8 thread: 3.6Ghz - 4Ghz turbo

I don't expect them to clock as high as SkyLake or even quite as high as DevilsCanyon. Intel's process is more mature.

 

[drunkenmaster]

It might be 3.2Ghz who knows.

I can't remember a single ES that was at shipping clock speeds from AMD and most for Intel as well, though sometimes if they did a new stepping very late before intending to launch they might get closer but in general Intel ES are well below shipping retail chip speeds also.

 

[DragonQ]

I'm going to look at it in more detail tomorrow, but yes essentially I don't believe 2,8 to 3.2Ghz will be the release speed of Zen, tho it might be for the 16 threader, or it might only get another 200Mhz.
It's the big fat chip of the range, the one sucking up the most power, it's likely to be a bit reigned in to keep it within the 95 watt envelope it look like they set for themselves.
Every chip has a zone beyond which it requires very diminished power to performance returns, 10% more MHz could cost 30‰ more power, on a fat 8 core 16 thread CPU 30% more power could result in 95 watts becoming 120 watts, be that as it may it's better to keep the fat one on a diet.

For the smaller chips like the 6 core you could add another 10% Mhz and keep within the 95 watts, the 4 core perhaps 15% for 85 watts.

So I'm really not worried about the MHz on this chart at all, it's not far off where I would have expected the big flagship chip to be, and it is just a test sample.

16 thread: 3Ghz - 3.2Ghz turbo
12 thread: 3.3 GHz - 3.5Ghz turbo
8 thread: 3.6Ghz - 4Ghz turbo

I don't expect them to clock as high as SkyLake or even quite as high as DevilsCanyon. Intel's process is more mature.

The 16-thread i7-6900K turbos to 4 GHz. I wouldn't expect Zen to be that far off. You're right though, if the 16-thread version might be slightly slower than the 12 and 8 thread versions, which means gamers will have to choose between cores and single thread performance as always.

 

[nashathedog]

The 16-thread i7-6900K turbos to 4 GHz. I wouldn't expect Zen to be that far off. You're right though, if the 16-thread version might be slightly slower than the 12 and 8 thread versions, which means gamers will have to choose between cores and single thread performance as always.

That's where the decisions become harder, I want good gaming performance but I was hoping to get a decent 8 core 16 thread cpu. I definitely want more cores than my 4790k has and certainly don't see any point in i7 owners like myself moving to Zen if they have to drop cores to keep up with what they've had for the last 2, 3 or 4 years. If I do end up staying with a 4 core 8 thread cpu it needs to be significantly faster than Devils canyon. So if Zen can't deliver I'll probably wait on Intels 10nm release.

 

[h4rm0ny]

Agreed, I couldn't believe how stupid that article was when I read it. How can they conclude it's a "bad start" for Zen and complain that it can't beat Intel's 3 year old Haswell CPU when it's clocked nearly 1 GHz lower?? Surely they must know that 3.2 GHz is not going to be the speed of the fastest Zen chip...

Thirded (or fourthed, or whatever we're up to). Techspot articles are an ugly mix of ignorant and lazy. Even when they do correct themself in the article and note that the final clockspeeds could be higher (which of course they will because final clocks are ALWAYS higher than the engineering sample), they over-correct and say "...if it's 4GHz". I mean in theory it could be but on 14nm that would be very high. My guesstimate is probably around 3.6GHz which I freely admit is a guess, but at least it's an educated one that fits in with what Intel have achieved on that process. If I were writing an article such as that I would at the least do some BASIC ARITHMETIC and state what the calculated IPC increase is based on those figures.

Really we can't read too much into an AOTS benchmark. It's fascinating to watch these things slowly start to leak, but we're very far from a realistic picture here.

 

[DragonQ]

...they over-correct and say "...if it's 4GHz". I mean in theory it could be but on 14nm that would be very high.

i7-6700K is 4 GHz and 14 nm but I agree that AMD's version will likely be not as clockable, at least at first.

 

[humbug]

Thirded (or fourthed, or whatever we're up to). Techspot articles are an ugly mix of ignorant and lazy. Even when they do correct themself in the article and note that the final clockspeeds could be higher (which of course they will because final clocks are ALWAYS higher than the engineering sample), they over-correct and say "...if it's 4GHz". I mean in theory it could be but on 14nm that would be very high. My guesstimate is probably around 3.6GHz which I freely admit is a guess, but at least it's an educated one that fits in with what Intel have achieved on that process. If I were writing an article such as that I would at the least do some BASIC ARITHMETIC and state what the calculated IPC increase is based on those figures.

Really we can't read too much into an AOTS benchmark. It's fascinating to watch these things slowly start to leak, but we're very far from a realistic picture here.

Arithmetic is something thay would need to lockup in a dictionary.