Discussion: AMD Ryzen

[8350rocks]

Completely irrelevant because AMD wasn't investing $100 mil, as the graph shows...

It looks like that line is at $100 Mil USD.

Can you please provide more concrete evidence of the spending numbers, because a vague graph with no specific points other than billion dollar increments is not specific enough to settle this.

The appearance of the graph is that AMD was spending around $100 mil USD in the beginning of the graph, which would be the approximate timeframe that K7/K8 was being developed.

You do realize that fiscal Q4 2016 is Jan-Mar 2017....yes?

Completely irrelevant because Zen will be not available to consumers until 2017.

Please read what I wrote...and I mean actually read it, your response to my post is inapprpriate...if you cannot see why, I am afraid we need not have future conversations.

Emphasis mine, in case you are not understanding, or since you are from Spain:

Sabes que el ultimo cuarto en el ano fiscal 2016 se ocura en los meses de enero a marzo en el ano 2017, verdad?

 

[con635]

Rumour from AT forum, new engineering sample, 8 core with ht still A0, 3.15ghz base, 3.3ghz all core turbo and 3.6ghz single core turbo and 95w, tiny grain of salt needed.

 

[cdrkf]

@con365 that sounds plausible, and wouldn't be a disaster. Hopefully they'll achieve higher clocks on the quad core part....

 

[-Fran-]

The "all core turbo" bit sounds weird to me, though. At least for me and current nomenclature, the "base clock" refers to the speed where all cores are being taxed, which would be in turn the "all cores turbo speed". So, that 3.15Ghz base and 3.3Ghz "all core turbo" sounds weird. I'm sure I am missing something here.

On the other hand, with a big "IF" implied, that 3.6Ghz Turbo sounds good. That means, if the scaling is not *that* terribad, hitting 4Ghz turbo with some tweaking under 125W sounds totally achievable.

Cheers!

 

[ComputerSecurityGuy]

That's within plausible predictions- I think most of us were expecting a 3.0-3.2gHz clock at 95W. That puts it right about on par with the i7-6900K. Only with a 95W TDP, although with AMD *actual* power consumption may differ. However that seems hopeful.


As for the quad core part I'm expecting 3.4-3.6gHz, which really would make it an attractive competitor to the i5 and a replacement for the now-unattractive Xeon E3 line.

So, anyway, it probably happened. Probably.

 

[gamerk316]

The "all core turbo" bit sounds weird to me, though. At least for me and current nomenclature, the "base clock" refers to the speed where all cores are being taxed, which would be in turn the "all cores turbo speed". So, that 3.15Ghz base and 3.3Ghz "all core turbo" sounds weird. I'm sure I am missing something here.

1/2 marketing, 1/2 power draw limitations. The fact the CPU only turbos 150 MHz on all cores kind of hints toward power draw limitations, which some of us (Juan) have been claiming for a while now. We'll see, but my first reaction? Bad news for AMD.

 

[ComputerSecurityGuy]

Not that bad. If we take the Blender demo as a fair demo of Zen, +/- maybe 5% depending on the benchmark, then it looks like Zen can hit the same clocks as Broadwell and hence pretty close to Broadwell performance, but possibly at a lower TDP (although being AMD who knows what actual power consumption is). For the consumer market it looks good, but the crucial actual power consumption figures are what will decide the server market.

 

[-Fran-]

The "all core turbo" bit sounds weird to me, though. At least for me and current nomenclature, the "base clock" refers to the speed where all cores are being taxed, which would be in turn the "all cores turbo speed". So, that 3.15Ghz base and 3.3Ghz "all core turbo" sounds weird. I'm sure I am missing something here.

1/2 marketing, 1/2 power draw limitations. The fact the CPU only turbos 150 MHz on all cores kind of hints toward power draw limitations, which some of us (Juan) have been claiming for a while now. We'll see, but my first reaction? Bad news for AMD.

But in technical terms, would that be like advertising a lower P state (power saving territory) as "base" and the "power state" as "all core turbo"? That would make absolutely no sense what-so-ever... But yeah... Marketing... So it is not out of possibility xD

Cheers!

 

[cdrkf]

The "all core turbo" bit sounds weird to me, though. At least for me and current nomenclature, the "base clock" refers to the speed where all cores are being taxed, which would be in turn the "all cores turbo speed". So, that 3.15Ghz base and 3.3Ghz "all core turbo" sounds weird. I'm sure I am missing something here.

1/2 marketing, 1/2 power draw limitations. The fact the CPU only turbos 150 MHz on all cores kind of hints toward power draw limitations, which some of us (Juan) have been claiming for a while now. We'll see, but my first reaction? Bad news for AMD.

But in technical terms, would that be like advertising a lower P state (power saving territory) as "base" and the "power state" as "all core turbo"? That would make absolutely no sense what-so-ever... But yeah... Marketing... So it is not out of possibility xD

Cheers!

It's following Intel's example- they have 'all core' turbo on most of their cpu's as well.

I think the point to remember is that *all cores being used* isn't the same as *all cores being used 100%*. A lot of tasks don't necessarily tax the cores all that hard- so in those situations the CPU can wind up all the cores a bit.

I mean I have an i7 laptop (with the 6700hq) and whilst the base speed is technically only 2.6ghz, it usually runs with all 8 threads active around the 3.2 - 3.3ghz mark. It can ramp up as high as 3.5ghz in single thread.

@gamerK, for an 8 core, 16 thread part those speeds don't look like bad news at all to me, especially if that's at a rather conservative 95w tdp.

 

[-Fran-]

It's following Intel's example- they have 'all core' turbo on most of their cpu's as well.

I think the point to remember is that *all cores being used* isn't the same as *all cores being used 100%*. A lot of tasks don't necessarily tax the cores all that hard- so in those situations the CPU can wind up all the cores a bit.

I mean I have an i7 laptop (with the 6700hq) and whilst the base speed is technically only 2.6ghz, it usually runs with all 8 threads active around the 3.2 - 3.3ghz mark. It can ramp up as high as 3.5ghz in single thread.

@gamerK, for an 8 core, 16 thread part those speeds don't look like bad news at all to me, especially if that's at a rather conservative 95w tdp.

Half agree there (bolded): it's not on usage, but power/heat. Er... I know they're linked, but i'm just being a bit pedantic 😛

So, I agree on what you're implying and it makes sense, but I don't agree on that it directly translates to a marketable term. Hence why it wouldn't make sense to say: "yeah, we can have all 8 cores running at 3.3Ghz, PROMISE *crosses fingers* ". The speed in which the CPU can *throttle* back and forth depends on cooling first and manufacturing second. It would be great that AMD could promise such a thing, but I'm not keen on buying such a promise.

I think I explained a bit better what I meant there, right? 😛

Cheers!

EDIT: Typo.

 

[cdrkf]

It's following Intel's example- they have 'all core' turbo on most of their cpu's as well.

I think the point to remember is that *all cores being used* isn't the same as *all cores being used 100%*. A lot of tasks don't necessarily tax the cores all that hard- so in those situations the CPU can wind up all the cores a bit.

I mean I have an i7 laptop (with the 6700hq) and whilst the base speed is technically only 2.6ghz, it usually runs with all 8 threads active around the 3.2 - 3.3ghz mark. It can ramp up as high as 3.5ghz in single thread.

@gamerK, for an 8 core, 16 thread part those speeds don't look like bad news at all to me, especially if that's at a rather conservative 95w tdp.

Half agree there (bolded): it's not on usage, but power/heat. Er... I know they're linked, but i'm just being a bit pedantic 😛

So, I agree on what you're implying and it makes sense, but I don't agree on that it directly translates to a marketable term. Hence why it wouldn't make sense to say: "yeah, we can have all 8 cores running at 3.3Ghz, PROMISE *crosses fingers* ". The speed in which the CPU can *throttle* back and forth depends on cooling first and manufacturing second. It would be great that AMD could promise such a thing, but I'm not keen on buying such a promise.

I think I explained a bit better what I meant there, right? 😛

Cheers!

EDIT: Typo.

Hi, yeah I follow- although all I'd say at this point is *this info didn't come from AMD*. It's not like AMD are (currently) advertising the cpu's like this- I think this is more an observation by someone who's seen the sample (just like my laptop i7 typically sits with all cores above 3ghz despite being rated as a 2.6ghz base).

Now, if AMD release a slide boasting a '3.3ghz all core turbo speed' that might be a different matter- although advertising turbo at all is a bit of a sticky wicket imo as the *only speed* any modern cpu is *guaranteed of running at* is it's base speed. Turbo frequencies depend on the specific task and resulting thermal / power headroom.

 

[-Fran-]

That is true; thanks for reminding me it's based on a forum post. Accuracy might be lost in the translation somewhere along the way so I don't have to take it to heart or so seriously.

I'll just keep the positive aspect of it, putting the clock speeds north of the 3.2Ghz/3.5Ghz speeds I settled with and making 4Ghz a bit more feasible to pull (theoretically).

Cheers!

 

[juanrga]

Completely irrelevant because AMD wasn't investing $100 mil, as the graph shows...

It looks like that line is at $100 Mil USD.

Can you please provide more concrete evidence of the spending numbers, because a vague graph with no specific points other than billion dollar increments is not specific enough to settle this.

The appearance of the graph is that AMD was spending around $100 mil USD in the beginning of the graph, which would be the approximate timeframe that K7/K8 was being developed.

The graph clearly shows AMD wasn't spending $100 mil, but more.

You do realize that fiscal Q4 2016 is Jan-Mar 2017....yes?

Completely irrelevant because Zen will be not available to consumers until 2017.

Please read what I wrote...and I mean actually read it, your response to my post is inapprpriate...if you cannot see why, I am afraid we need not have future conversations.

Emphasis mine, in case you are not understanding, or since you are from Spain:

Sabes que el ultimo cuarto en el ano fiscal 2016 se ocura en los meses de enero a marzo en el ano 2017, verdad?

Again, completely irrelevant to the point made about Zen launch.

P.S.: That is not Spanish.

 

[juanrga]

That's within plausible predictions- I think most of us were expecting a 3.0-3.2gHz clock at 95W. That puts it right about on par with the i7-6900K. Only with a 95W TDP, although with AMD *actual* power consumption may differ. However that seems hopeful.


As for the quad core part I'm expecting 3.4-3.6gHz, which really would make it an attractive competitor to the i5 and a replacement for the now-unattractive Xeon E3 line.

So, anyway, it probably happened. Probably.

Indeed, most people was expecting 3.0--3.2GHz for the octo-core and a bit more for the quad-core. For the quad-core I was expecting 3.3GHz base and 3.7GHz turbo.

http://www.tomshardware.co.uk/forum/id-2986517/discussion-amd-zen/page-11.html#18572956

But the more relevant part of the leak (if it is legit) is this:

The second SKU is a 4-core one with AMD's HT. It's got a 65W TDP and the base clock is still 2900 MHz. All-core turbo is 3100 MHz, max turbo is 3400 MHz. I don't know if it's only an SKU for testing mainboards or something is not okay with the clock-wattage correlation. I mean on higher clocks the 4-core SKU steps into the 95W TDP envelope, AMD can't keep the wattage low. Maybe GloFo's 14nm process needs some maturing... Frankly I don't have a clue what's in the background.

Both chips are made on the same process node, but then why has the octo-core clock increased from 2.8GHz to 3.15GHz, but the quad-core has increased only from 2.8GHz to 2.9Ghz? Process maturity would affect both chips in a similar way.

The leaker suggests some defect with the SKU, but another possibility is possible, than the octo-core chip is a golden-chip.

 

[con635]

As for the quad core part I'm expecting 3.4-3.6gHz, which really would make it an attractive competitor to the i5 and a replacement for the now-unattractive Xeon E3 line.

So, anyway, it probably happened. Probably.

Imagine it was branded athlon 1060k and was £60, we can dream cant we

 

[juanrga]

Some remarks:

1. I think the Blender demo was a best case scenario (even for non-AVX software). Using the average 40% officially claimed by AMD we find Zen IPC is at the Sandy Bridge level more or less.

2. HEDT is a niche market. The only reason why AMD is releasing octo-core CPUs is because (i) those are based in server dies that didn't pass quality test and (ii) the higher core count will reduce the clock problem that AMD has by choosing 14LPP process.

3. The problems for Zen will be efficiency on servers and clocks for mainstream desktop users. A quad-core 3.3GHz Zen will not compete well against a 4.2GHz quad-core KBL.

4. TDP is the biggest unknown. AMD officially rates Zen up to 95W, but during the Blender demo they claimed that Zen targets TDP similar to BDW-E and this one is rated at 140W. I have a feeling that those official 95W are in reality more close to 125W.

 

[-Fran-]

The second SKU is a 4-core one with AMD's HT. It's got a 65W TDP and the base clock is still 2900 MHz. All-core turbo is 3100 MHz, max turbo is 3400 MHz. I don't know if it's only an SKU for testing mainboards or something is not okay with the clock-wattage correlation. I mean on higher clocks the 4-core SKU steps into the 95W TDP envelope, AMD can't keep the wattage low. Maybe GloFo's 14nm process needs some maturing... Frankly I don't have a clue what's in the background.

Both chips are made on the same process node, but then why has the octo-core clock increased from 2.8GHz to 3.15GHz, but the quad-core has increased only from 2.8GHz to 2.9Ghz? Process maturity would affect both chips in a similar way.

The leaker suggests some defect with the SKU, but another possibility is possible, than the octo-core chip is a golden-chip.

That reason has more weight by far.

The top "golden" SKU will most likely be meant for Pro stuff (server) first and consumer last (if at all), then everything else goes consumer way. Plus, if that is a harvested die from an early engy sample, then makes sense for it to be clocked that low at first glance. Or they were trying to simulate a 4 core with the 8 one disabling stuff via microcode in the test MoBos, which would also explain the clocks to keep the TDPs low.

I do believe the clocks for the 4C/8T part won't be that bad, but they won't be Haswell level either. Sandy was 3.4Ghz/3.8Ghz?

Cheers!

 

[esrever]

Those clocks look great for ES.

 

[juanrga]

Those clocks look great for ES.

Sometimes final stepping ES achieve higher clocks than production chips. Late bulldozer ES could hit 5GHz.

 

[juanrga]

The second SKU is a 4-core one with AMD's HT. It's got a 65W TDP and the base clock is still 2900 MHz. All-core turbo is 3100 MHz, max turbo is 3400 MHz. I don't know if it's only an SKU for testing mainboards or something is not okay with the clock-wattage correlation. I mean on higher clocks the 4-core SKU steps into the 95W TDP envelope, AMD can't keep the wattage low. Maybe GloFo's 14nm process needs some maturing... Frankly I don't have a clue what's in the background.

Both chips are made on the same process node, but then why has the octo-core clock increased from 2.8GHz to 3.15GHz, but the quad-core has increased only from 2.8GHz to 2.9Ghz? Process maturity would affect both chips in a similar way.

The leaker suggests some defect with the SKU, but another possibility is possible, than the octo-core chip is a golden-chip.

That reason has more weight by far.

The top "golden" SKU will most likely be meant for Pro stuff (server) first and consumer last (if at all), then everything else goes consumer way.

There are rumors about a consumer black series with higher clocks, something similar to the 9000 series.

 

[jdwii]

Guys this is a 8 core part Intel themselves only have a 3.2Ghz base and 3.7Ghz turbo(TDP 140). Think some people think this is something like the fx series which aren't true 8 core parts or at least not as full of a "processor" as a Intel 8 core.

I guess we are talking about 4ALU+2AGU so maybe power consumption is down over that or maybe its down a little over it having only a 256bit FPU? We also have to think about GF 14nm too.

 

[jimmysmitty]

Guys this is a 8 core part Intel themselves only have a 3.2Ghz base and 3.7Ghz turbo(TDP 140). Think some people think this is something like the fx series which aren't true 8 core parts or at least not as full of a "processor" as a Intel 8 core.

I guess we are talking about 4ALU+2AGU so maybe power consumption is down over that or maybe its down a little over it having only a 256bit FPU? We also have to think about GF 14nm too.

To be fair Intels 10 core is also a 140w TDP part and is base of 3GHz with boost to 3.5GHz. As well we still need to know if the 95w TDP is a TDP or ADP rating as both Intel and AMD seem to change depending on a lot of factors, although Intel seems to have stuck with TDP since Core 2.

 

[simon12]

Anyone know why AMD limited it to 95W when 125W could allow higher clocks?

 

[jimmysmitty]

Anyone know why AMD limited it to 95W when 125W could allow higher clocks?

My only thoughts would be a limitation of the 14nm LPP rather than anything else.

 

[cdrkf]

Anyone know why AMD limited it to 95W when 125W could allow higher clocks?

My only thoughts would be a limitation of the 14nm LPP rather than anything else.

Yeah with every process and design there is a 'point of diminishing returns'- case in point is Carrizo- very efficient (encroaching on Intel level perf/W) at low clocks (so very competitive vs Intel low power laptop / tablet parts), but doesn't scale up past 65w.

In the case of Zen- you'll find that the jump from 95w - 125w nets virtually no performance but jumps up the power consumption considerably which will ruin the perf/w figures. That might have been ok in the past but these days people don't tolerate speed at the expense of everything else- power consumption is something that gets scrutinized closely. I think AMD are better off being 10% behind Intel in raw performance but with equivalent perf/w, rather than matching them but being 20% behind in efficiency (numbers are just as an example, not a prediction 😛).

It'll be interesting to see how it clocks and the power curve, it's probably likely the 14nm LLP process doesn't scale up very well, based on the GPU releases (looking at the specs for the GTX 1050ti, that interestingly hits about the same frequencies that AMD have with Polaris at 1200mhz range and *NOT* the 1500 - 1600 range the other gtx 10xx boards go up to which I think gives a comparison of how 14nm llp compares against TSMC's 16nm ff+ process i.e. it's not an especially high clocking process).