AMD CPUs, SoC Rumors and Speculations Temp. thread 2

[juanrga]

My original doubt about that 50% higher Cinebench score isn't solved. I am now reading two different (and contradictory) replies. One of them suggests that the sample was overclocked at 3.6GHz using extreme cooling. The other reply says that, on the contrary, the FX-8350 was underclocked to 2.4GHz to match the frequency of the sample.

Which is the valid reply?

I was not given specifics of the benchmark...or how the comparison was obtained.

However...both are possible options...they could have used any cooling solution to reach a specific clockspeed, they could have also underclocked as well...they could have even done both.

Both are valid options...and if you are comparing clock for clock, the manner in which they made the clocks equivalent is unimportant.

Therefore we can conclude that the benchmark is useless to us to extract real performance, but we can still do some remarks:

If the option valid is the second, then AMD comparing Zen to underclocked FX-8350 means that the IPC gain was of 50%, it wasn't "60%", which puts Zen is a poor shape.

If the first option is the valid, then AMD needing LN2 to hit 3.6GHz doesn't bother well for the kind of frequencies will be achieved by Zen on air or water, specially when considering that ES sometimes have much higher thermal headroom than final chips for the purpose of extreme stress testing.

 

[juanrga]

I heard something interesting today...apparently, on the 14LPP node, I am hearing they pick up approximately 40% more clock speed for the same power draw over the 28nm node on GPUs.

Obviously...that is not a direct comparison to the 32nm PD-SOI node...and clocks on GPUs are typically much lower than CPUs; however...that does bode well for the prospect of clock speed across the board.

That doesn't mean anything.

Foundries make claims as that for the optimal part of the process node. It is a kind of "up to" gain, not an average gain. Depending on the optimizations and the target, a process can provide gains in a region and lost in another. I will illustrate this with Intel change from 32nm to 22nm. The 22nm was more optimized for mobile, as a consequence we saw gains at lower frequencies, but extrapolating to higher frequencies the 32nm was better than 22nm. This is why SB i5/i7 were better overclockers than IB i5/i7. At arorund 1.3V IB hit a wall

A similar case for AMD? Kaveri vs Carrizo. Both done on 28nm, but one on 28SHP (optimized for higher frequencies) and the on 28GFA (optimized for power)

Again below certain cross point, Excavator achieves higher frequencies at same power, whereas the tendency is inverted at the higher frequencies. Steamroller is better at higher frequencies and that is the reason why Carrizo is mobile only. That is also why we don't see excavator-based 8C FX CPUs. the frequencies would be too low for compensating for the IPC gain over Piledriver and the overclocking headroom would be ridiculous.

Expect something similar for 14LPP (Low Power Plus) vs 28SHP (Super High Performance). 14LPP will be much better for frequencies around 1--2GHz then will be a crossing point and it will be poor than the former node. The million dollar question is where is the crossing point. Not even AMD engineers knows this, only Globalfoundries engineers have precise knowledge. I guess the crossing point is somewhat above 3GHz and this is why I don't expect 3.6GHz base clocks.

It is worth to mention what The Stilt said once AMD announced clocks for the RX480 at Computex. He was expecting much higher clocks:

If the final clocks for RX 480 are indeed as low as 1266MHz, it quite honestly doesn't look too good for 14nm LPP

If AMD isn't able to push the clocks significantly higher (1500MHz+) without blowing the power draw through the ceiling, that's extremely alarming IMO.

AMD partners have shipped Tahiti and Bonaire based cards clocked to 1150 - 1200MHz from the box, and these GPUs were built on TSMC 28nm HP process... ...Which is thought inferior to GlobalFoundries 28nm HPP for example. Not to mention that Bonaire for example is able to clock close to 1400MHz on air cooling.

So if AMD is really unable to ship a ~232mm² GPU clocked higher than 1266MHz, then my fears about the 14nm LPP have materialized even worse than I anticipated

Meanwhile nVidia was able to clock a 35% larger (232 vs. 314mm²) GPU 50% higher (1266MHz vs. 1898MHz), while only having 20% higher TDP

 

[cdrkf]

I heard something interesting today...apparently, on the 14LPP node, I am hearing they pick up approximately 40% more clock speed for the same power draw over the 28nm node on GPUs.

Obviously...that is not a direct comparison to the 32nm PD-SOI node...and clocks on GPUs are typically much lower than CPUs; however...that does bode well for the prospect of clock speed across the board.

That doesn't mean anything.

Foundries make claims as that for the optimal part of the process node. It is a kind of "up to" gain, not an average gain. Depending on the optimizations and the target, a process can provide gains in a region and lost in another. I will illustrate this with Intel change from 32nm to 22nm. The 22nm was more optimized for mobile, as a consequence we saw gains at lower frequencies, but extrapolating to higher frequencies the 32nm was better than 22nm. This is why SB i5/i7 were better overclockers than IB i5/i7. At arorund 1.3V IB hit a wall

A similar case for AMD? Kaveri vs Carrizo. Both done on 28nm, but one on 28SHP (optimized for higher frequencies) and the on 28GFA (optimized for power)

Again below certain cross point, Excavator achieves higher frequencies at same power, whereas the tendency is inverted at the higher frequencies. Steamroller is better at higher frequencies and that is the reason why Carrizo is mobile only. That is also why we don't see excavator-based 8C FX CPUs. the frequencies would be too low for compensating for the IPC gain over Piledriver and the overclocking headroom would be ridiculous.

Expect something similar for 14LPP (Low Power Plus) vs 28SHP (Super High Performance). 14LPP will be much better for frequencies around 1--2GHz then will be a crossing point and it will be poor than the former node. The million dollar question is where is the crossing point. Not even AMD engineers knows this, only Globalfoundries engineers have precise knowledge. I guess the crossing point is somewhat above 3GHz and this is why I don't expect 3.6GHz base clocks.

It is worth to mention what The Stilt said once AMD announced clocks for the RX480 at Computex. He was expecting much higher clocks:

If the final clocks for RX 480 are indeed as low as 1266MHz, it quite honestly doesn't look too good for 14nm LPP

If AMD isn't able to push the clocks significantly higher (1500MHz+) without blowing the power draw through the ceiling, that's extremely alarming IMO.

AMD partners have shipped Tahiti and Bonaire based cards clocked to 1150 - 1200MHz from the box, and these GPUs were built on TSMC 28nm HP process... ...Which is thought inferior to GlobalFoundries 28nm HPP for example. Not to mention that Bonaire for example is able to clock close to 1400MHz on air cooling.

So if AMD is really unable to ship a ~232mm² GPU clocked higher than 1266MHz, then my fears about the 14nm LPP have materialized even worse than I anticipated

Meanwhile nVidia was able to clock a 35% larger (232 vs. 314mm²) GPU 50% higher (1266MHz vs. 1898MHz), while only having 20% higher TDP

I think that last quote may well be jumping the gun a bit. There are multiple reasons why the RX480 is clocked at 1266 mhz, and only one of them is 'it can't go any faster'

I'd say before we can come to that conclusion, we also need to consider:
- The RX480 is likely not the full Polaris 10 gpu, as it has a rather strange core count if it is.

- GCN has always been a relatively low clock design, especially if we're talking efficiency which is evidently a focus for AMD this time around. For example, many GCN gpu's clocked at 800 mhz offered comparable efficiency in terms of perf/w to nVidia. However when AMD ramped these up to above the 1ghz point on 28nm to match NV for outright performance, the efficiency went through the floor despite being on the same process. That suggests the GCN architecture isn't designed for high clocks and instead favours higher shader counts. NV on the other hand have been keeping the shader counts modest and the performance gains in Pascal come from much higher clocks instead.

- AMD may still have plenty of clock speed head room in Polaris 10, and are choosing not to use it, due to the target market segment for this gpu. This is made all the more likely if they intend to offer a higher core count, higher clocked variant of Polaris 10 in the near future.

 

[jimmysmitty]

If Zen is clocked at 2.4GHz it is not the end of the world unless they are on par with Intel per clock with too old of a generation of CPU, especially since Intels 14nm is hitting base clocks of 4GHz.

I think it is better for this discussion if we use the clocks of the 8-core Intel cpu against 8-core Zen, which is 3.2GHz base and 3.7GHz turbo (i7-6900k). We can't mix 8-cores from AMD against 4-cores from Intel.

And why not? We currently do, FX 8350 vs an i7 or even an i5.

I agree though as that will be the market they will try to compete in but even so Intel has a clear clock speed advantage if the 2.4GHz is a top end clock speed. We wont know until closer to release.

Of course the real fun is in the fact that this has done nothing but fuel more rumors and it will eventually fuel more until release causing more crap than needs be.

If I can make 1 prediction- Zen 'top' clock speed isn't going to be 2.4ghz. Looking at this logically, there'd be 0 point in AMD developing and launching an 'almost as high IPC' as Intel with a serious clock speed deficit. That wouldn't gain them anything from where they are now.

Now, I can accept that base clocks could possibly be sub 3ghz (at least on the 8 core parts), however in that case I expect aggressive boost frequencies. I'd also think we'd see higher tdp parts than 95w if the clocks are that low (I mean again AMD have proven they'll push power in order to be performance competitive repeatedly in the past, typically going up to 140W if we ignore the 220w 9590).

AMD are expecting to be 'competitive' with equivalent Intel parts- that to me suggests we should see something close to Intels 8 core. That would suggest base clock in the low 3ghz ballpark. I mean the *current* engineering samples are at 2.4ghz already, and engineering samples are always pretty low clocked compared to a consumer product.

Edit: To put this another way doing a little bit of *very rough math*

The FX 8350 scores 1.1 in the Cinebench 11.5 single thread test. Assuming (generously) that Zen IPC is 1.6x that of Vishera (as the 40% is above Excavator), then at the same speed (4ghz) Zen would score a respectable 1.76 points.

Now if we state that the maximum frequency for Zen is just 2.4 ghz, then the Zen ST score would be just 1.06 points, i.e. slower in real terms than the current FX cpu. That wouldn't be a worthwhile product for them to launch.

I guess it remains to be seen then. All we can do is hope for the best and prepare for the possibility of the worst.

 

[cdrkf]

If Zen is clocked at 2.4GHz it is not the end of the world unless they are on par with Intel per clock with too old of a generation of CPU, especially since Intels 14nm is hitting base clocks of 4GHz.

I think it is better for this discussion if we use the clocks of the 8-core Intel cpu against 8-core Zen, which is 3.2GHz base and 3.7GHz turbo (i7-6900k). We can't mix 8-cores from AMD against 4-cores from Intel.

And why not? We currently do, FX 8350 vs an i7 or even an i5.

I agree though as that will be the market they will try to compete in but even so Intel has a clear clock speed advantage if the 2.4GHz is a top end clock speed. We wont know until closer to release.

Of course the real fun is in the fact that this has done nothing but fuel more rumors and it will eventually fuel more until release causing more crap than needs be.

If I can make 1 prediction- Zen 'top' clock speed isn't going to be 2.4ghz. Looking at this logically, there'd be 0 point in AMD developing and launching an 'almost as high IPC' as Intel with a serious clock speed deficit. That wouldn't gain them anything from where they are now.

Now, I can accept that base clocks could possibly be sub 3ghz (at least on the 8 core parts), however in that case I expect aggressive boost frequencies. I'd also think we'd see higher tdp parts than 95w if the clocks are that low (I mean again AMD have proven they'll push power in order to be performance competitive repeatedly in the past, typically going up to 140W if we ignore the 220w 9590).

AMD are expecting to be 'competitive' with equivalent Intel parts- that to me suggests we should see something close to Intels 8 core. That would suggest base clock in the low 3ghz ballpark. I mean the *current* engineering samples are at 2.4ghz already, and engineering samples are always pretty low clocked compared to a consumer product.

Edit: To put this another way doing a little bit of *very rough math*

The FX 8350 scores 1.1 in the Cinebench 11.5 single thread test. Assuming (generously) that Zen IPC is 1.6x that of Vishera (as the 40% is above Excavator), then at the same speed (4ghz) Zen would score a respectable 1.76 points.

Now if we state that the maximum frequency for Zen is just 2.4 ghz, then the Zen ST score would be just 1.06 points, i.e. slower in real terms than the current FX cpu. That wouldn't be a worthwhile product for them to launch.

I guess it remains to be seen then. All we can do is hope for the best and prepare for the possibility of the worst.

Yeah, I guess we'll find out soon enough. I just think 36 CU is a strange number for the full GPU. Once we get a proper die shot I'm sure people will work out what full Polaris looks like. I mean people correctly identified that Tonga in fact had 6 memory interfaces for a 384 bit bus, although AMD never utilized all 6 (I guess the added memory bandwidth didn't equate to enough extra performance to be worth the cost?).

As for Zen, I choose to remain cautiously optimistic that it'l be a decent product. I'm not expecting miracles, just something that will actually represent an option to buyers for a range of scenarios. For gaming, it possibly won't be the best value option if it's lowest configuration is a hex core, but then again lower core count options are due next year so that doesn't strike me as a problem at this point assuming the per core performance is there.

 

[sleeper357]

I do not see the issue with lower clocked zen for mainstream consumers coming in at 45watt to 65 watt and enthusiast @ 3.5ghz with higher tdp.

without knowing test specs it is speculation but the fact there is a zen part out there capable of 3.5Ghz is good news.

 

[jdwii]

I do not see the issue with lower clocked zen for mainstream consumers coming in at 45watt to 65 watt and enthusiast @ 3.5ghz with higher tdp.

without knowing test specs it is speculation but the fact there is a zen part out there capable of 3.5Ghz is good news.

The thing is a 3.5Ghz part has not been seen and its rumored we will only see that with turbo mode.

 

[juanrga]

It is worth to mention what The Stilt said once AMD announced clocks for the RX480 at Computex. He was expecting much higher clocks:

If the final clocks for RX 480 are indeed as low as 1266MHz, it quite honestly doesn't look too good for 14nm LPP

If AMD isn't able to push the clocks significantly higher (1500MHz+) without blowing the power draw through the ceiling, that's extremely alarming IMO.

AMD partners have shipped Tahiti and Bonaire based cards clocked to 1150 - 1200MHz from the box, and these GPUs were built on TSMC 28nm HP process... ...Which is thought inferior to GlobalFoundries 28nm HPP for example. Not to mention that Bonaire for example is able to clock close to 1400MHz on air cooling.

So if AMD is really unable to ship a ~232mm² GPU clocked higher than 1266MHz, then my fears about the 14nm LPP have materialized even worse than I anticipated

Meanwhile nVidia was able to clock a 35% larger (232 vs. 314mm²) GPU 50% higher (1266MHz vs. 1898MHz), while only having 20% higher TDP

I think that last quote may well be jumping the gun a bit. There are multiple reasons why the RX480 is clocked at 1266 mhz, and only one of them is 'it can't go any faster'

Spoiler

I'd say before we can come to that conclusion, we also need to consider:
- The RX480 is likely not the full Polaris 10 gpu, as it has a rather strange core count if it is.

- GCN has always been a relatively low clock design, especially if we're talking efficiency which is evidently a focus for AMD this time around. For example, many GCN gpu's clocked at 800 mhz offered comparable efficiency in terms of perf/w to nVidia. However when AMD ramped these up to above the 1ghz point on 28nm to match NV for outright performance, the efficiency went through the floor despite being on the same process. That suggests the GCN architecture isn't designed for high clocks and instead favours higher shader counts. NV on the other hand have been keeping the shader counts modest and the performance gains in Pascal come from much higher clocks instead.

- AMD may still have plenty of clock speed head room in Polaris 10, and are choosing not to use it, due to the target market segment for this gpu. This is made all the more likely if they intend to offer a higher core count, higher clocked variant of Polaris 10 in the near future.

I fully agree the Stilt is not offering a definitive proof. He is only speculating. The force in his argument appears when what he says about clocks is complemented by the rest of the information about Zen:

• 
  ■ AMD is reducing by >20% its targets for clocks. Original target was 4.0--4.1 GHz, now are considering 3.3--3.4GHz.
  ■ People with access to engineering samples report unusually low clocks.
  ■ No one except AMD is choosing 14LPP for high performance chips. Any other company avoids 14LPP.
  ■ AMD is only announcing 8C chips; i.e., the chips that would be less disadvantaged by low clocks. 6C isn't even mentioned although I would expect them from yields arguments.

All the information, leaks, and speculations point towards lower clocks. It is possible that all that information is wrong and the coincidence just a casualty? Yes, but unlikely.

 

[juanrga]

I do not see the issue with lower clocked zen for mainstream consumers coming in at 45watt to 65 watt and enthusiast @ 3.5ghz with higher tdp.

without knowing test specs it is speculation but the fact there is a zen part out there capable of 3.5Ghz is good news.

To add to jdwii excellent answer, only Zen @2.4GHz is known at this time.

 

[cdrkf]

I do not see the issue with lower clocked zen for mainstream consumers coming in at 45watt to 65 watt and enthusiast @ 3.5ghz with higher tdp.

without knowing test specs it is speculation but the fact there is a zen part out there capable of 3.5Ghz is good news.

To add to jdwii excellent answer, only Zen @2.4GHz is known at this time.

Which would be a disappointment as it would be about the same (or slightly slower) than an FX 8350 in single thread. It would still offer good performance uplift in multi threaded applications thanks to offering 8 full cores and double the threads (hard to estimate how it's multi core scaling will compare). I doubt that will be the case though.

 

[juanrga]

Current Zen silicon only hits 2.4GHz, but I expect them to get clocks higher on final silicon. Said that, I doubt that they will break the 3GHz barrier.

 

[salgado18]

Current Zen silicon only hits 2.4GHz, but I expect them to get clocks higher on final silicon. Said that, I doubt that they will break the 3GHz barrier.

I'm no expert, but I believe they will get at least 3GHz base on the octocore, even if it means using aggressive power management and marketing twists.

 

[8350rocks]

If Zen is clocked at 2.4GHz it is not the end of the world unless they are on par with Intel per clock with too old of a generation of CPU, especially since Intels 14nm is hitting base clocks of 4GHz.

I think it is better for this discussion if we use the clocks of the 8-core Intel cpu against 8-core Zen, which is 3.2GHz base and 3.7GHz turbo (i7-6900k). We can't mix 8-cores from AMD against 4-cores from Intel.

And why not? We currently do, FX 8350 vs an i7 or even an i5.

I agree though as that will be the market they will try to compete in but even so Intel has a clear clock speed advantage if the 2.4GHz is a top end clock speed. We wont know until closer to release.

Of course the real fun is in the fact that this has done nothing but fuel more rumors and it will eventually fuel more until release causing more crap than needs be.

If I can make 1 prediction- Zen 'top' clock speed isn't going to be 2.4ghz. Looking at this logically, there'd be 0 point in AMD developing and launching an 'almost as high IPC' as Intel with a serious clock speed deficit. That wouldn't gain them anything from where they are now.

Now, I can accept that base clocks could possibly be sub 3ghz (at least on the 8 core parts), however in that case I expect aggressive boost frequencies. I'd also think we'd see higher tdp parts than 95w if the clocks are that low (I mean again AMD have proven they'll push power in order to be performance competitive repeatedly in the past, typically going up to 140W if we ignore the 220w 9590).

AMD are expecting to be 'competitive' with equivalent Intel parts- that to me suggests we should see something close to Intels 8 core. That would suggest base clock in the low 3ghz ballpark. I mean the *current* engineering samples are at 2.4ghz already, and engineering samples are always pretty low clocked compared to a consumer product.

Edit: To put this another way doing a little bit of *very rough math*

The FX 8350 scores 1.1 in the Cinebench 11.5 single thread test. Assuming (generously) that Zen IPC is 1.6x that of Vishera (as the 40% is above Excavator), then at the same speed (4ghz) Zen would score a respectable 1.76 points.

Now if we state that the maximum frequency for Zen is just 2.4 ghz, then the Zen ST score would be just 1.06 points, i.e. slower in real terms than the current FX cpu. That wouldn't be a worthwhile product for them to launch.

Zen ES has 2.4 clockspeed at the moment. At this same point in the lifecycle of BD/PD, ES clocks were the same.

Food for thought...clocks will improve.

 

[8350rocks]

I heard something interesting today...apparently, on the 14LPP node, I am hearing they pick up approximately 40% more clock speed for the same power draw over the 28nm node on GPUs.

Obviously...that is not a direct comparison to the 32nm PD-SOI node...and clocks on GPUs are typically much lower than CPUs; however...that does bode well for the prospect of clock speed across the board.

That doesn't mean anything.

Foundries make claims as that for the optimal part of the process node. It is a kind of "up to" gain, not an average gain. Depending on the optimizations and the target, a process can provide gains in a region and lost in another. I will illustrate this with Intel change from 32nm to 22nm. The 22nm was more optimized for mobile, as a consequence we saw gains at lower frequencies, but extrapolating to higher frequencies the 32nm was better than 22nm. This is why SB i5/i7 were better overclockers than IB i5/i7. At arorund 1.3V IB hit a wall

A similar case for AMD? Kaveri vs Carrizo. Both done on 28nm, but one on 28SHP (optimized for higher frequencies) and the on 28GFA (optimized for power)

Again below certain cross point, Excavator achieves higher frequencies at same power, whereas the tendency is inverted at the higher frequencies. Steamroller is better at higher frequencies and that is the reason why Carrizo is mobile only. That is also why we don't see excavator-based 8C FX CPUs. the frequencies would be too low for compensating for the IPC gain over Piledriver and the overclocking headroom would be ridiculous.

Expect something similar for 14LPP (Low Power Plus) vs 28SHP (Super High Performance). 14LPP will be much better for frequencies around 1--2GHz then will be a crossing point and it will be poor than the former node. The million dollar question is where is the crossing point. Not even AMD engineers knows this, only Globalfoundries engineers have precise knowledge. I guess the crossing point is somewhat above 3GHz and this is why I don't expect 3.6GHz base clocks.

It is worth to mention what The Stilt said once AMD announced clocks for the RX480 at Computex. He was expecting much higher clocks:

If the final clocks for RX 480 are indeed as low as 1266MHz, it quite honestly doesn't look too good for 14nm LPP

If AMD isn't able to push the clocks significantly higher (1500MHz+) without blowing the power draw through the ceiling, that's extremely alarming IMO.

AMD partners have shipped Tahiti and Bonaire based cards clocked to 1150 - 1200MHz from the box, and these GPUs were built on TSMC 28nm HP process... ...Which is thought inferior to GlobalFoundries 28nm HPP for example. Not to mention that Bonaire for example is able to clock close to 1400MHz on air cooling.

So if AMD is really unable to ship a ~232mm² GPU clocked higher than 1266MHz, then my fears about the 14nm LPP have materialized even worse than I anticipated

Meanwhile nVidia was able to clock a 35% larger (232 vs. 314mm²) GPU 50% higher (1266MHz vs. 1898MHz), while only having 20% higher TDP

Just to add a comment here.

The 480 was designed specifically with the goal of PPW and maximizing that PPW. There will be cards more power hungry than that, and the card will still be able to overclock past 1500-1600 from what I am hearing.

 

[sleeper357]

Which would be a disappointment as it would be about the same (or slightly slower) than an FX 8350 in single thread. It would still offer good performance uplift in multi threaded applications thanks to offering 8 full cores and double the threads (hard to estimate how it's multi core scaling will compare). I doubt that will be the case though.

But that would not jive with the 40% Dr Su spoke of would it? it would make since for a midrange apu in my opinion.

 

[cdrkf]

Which would be a disappointment as it would be about the same (or slightly slower) than an FX 8350 in single thread. It would still offer good performance uplift in multi threaded applications thanks to offering 8 full cores and double the threads (hard to estimate how it's multi core scaling will compare). I doubt that will be the case though.

But that would not jive with the 40% Dr Su spoke of would it? it would make since for a midrange apu in my opinion.

I was posing a hypothetical situation- to be clear I expect Zen to be clearly superior to any Dozer' derived core in single thread. A word of caution though, Lisa Su stated Zen has 'more than 40% higher IPC' (which stands for Instructions Per Clock). That is categorically not the same as saying 'Zen is 40% faster'.

The major issue with the bulldozer line of cores is they sacrificed single thread IPC in favour of better multi thread scaling for a given die area, and tried to compensate with very high clock speed. AMD and Intel's most successful architectures have however all focused on lower speeds and doing more per clock cycle, i.e. Higher IPC. Many thought the design of Bulldozer was odd, given AMD beat Intel in the reverse situation back with the Pentium 4. At the time Intel pushed their 'netburst' design for very high clocks (there were P4 parts clocked as high as modern cpu's, up into the 4ghz range), but lower IPC. Around the same time, AMD released the athlon 64 with much higher IPC and significantly lower clocks- the Athlon 64 won almost every benchmark going and used less power at the same time.

That brings us back to Zen- AMD have stated they are ditching the low IPC, high frequency design the construction cores are using, in favour of a more traditional high IPC design. So when running at the same speed, assuming AMD's assertion holds up, then a Zen CPU will do 40% more work than a Dozer' cpu. That said, that is assuming the two are at the same speed. The thing is though, high IPC designs are usually harder to hit high clocks, and there are questions over the higher speed capability of the manufacturing process they are using. The long and short of it is that Zen isn't going to be clocked at frequencies as high as AMD's current FX offerings (which go as high as 4.7ghz base, 5ghz turbo for the top end 9590). So chances are the performance boost going for FX -> Zen in single thread is going to be less than 40% as Zen will be at a clock disadvantage.

The other thing that is a difficulty for them is Intel not only have as high (or slightly higher IPC) than where Zen is predicted to fall, they also have the best manufacturing tech in the industry so have managed to push their clocks as high as AMD's FX parts (well apart from the water cooled enthusiast parts). This means Zen is going to be slower in single thread than the latest Intel kit. That needn't be an issue though, as it's still going to be a huge jump, they do need to get reasonable clocks out of it though. In multi thread it's likely to be pretty strong, as Zen is offering 8 full cores with SMT (basically same idea as hyperthreading) so 16 threads. It might not be as fast as Intel's best 8 core, but I'd expect it'll be close behind and certianly better than the Intel 6 core parts. That puts AMD right back at the high end of the market, however being able to justify those kind of prices does have the prerequisite that Zen doesn't have any major compromises that go with it- that is why getting close to Intel on single thread performance is important.

The current FX parts weren't bad in multi-thread when they came out, but it was always a case of 'similar multi thread to Intel with terrible single thread'. They didn't really offer an advantage. Hopefully this time around it will be 'better multi thread (for the price at any rate) with good (albeit not quite AS good) single thread'. That then becomes something worth considering, and was a similar position to where AMD were with Phenom II X6 vs the original Core i7.

 

[grsychckn]

Current Zen silicon only hits 2.4GHz, but I expect them to get clocks higher on final silicon. Said that, I doubt that they will break the 3GHz barrier.

Is this article FUD? Zen ES @ 3GHz. Seems like it claimed 3GHz two months ago and you have a reply in the comments about 3.5GHz turbo.

 

[juanrga]

Spoiler

I heard something interesting today...apparently, on the 14LPP node, I am hearing they pick up approximately 40% more clock speed for the same power draw over the 28nm node on GPUs.

Obviously...that is not a direct comparison to the 32nm PD-SOI node...and clocks on GPUs are typically much lower than CPUs; however...that does bode well for the prospect of clock speed across the board.

That doesn't mean anything.

Foundries make claims as that for the optimal part of the process node. It is a kind of "up to" gain, not an average gain. Depending on the optimizations and the target, a process can provide gains in a region and lost in another. I will illustrate this with Intel change from 32nm to 22nm. The 22nm was more optimized for mobile, as a consequence we saw gains at lower frequencies, but extrapolating to higher frequencies the 32nm was better than 22nm. This is why SB i5/i7 were better overclockers than IB i5/i7. At arorund 1.3V IB hit a wall

A similar case for AMD? Kaveri vs Carrizo. Both done on 28nm, but one on 28SHP (optimized for higher frequencies) and the on 28GFA (optimized for power)

Again below certain cross point, Excavator achieves higher frequencies at same power, whereas the tendency is inverted at the higher frequencies. Steamroller is better at higher frequencies and that is the reason why Carrizo is mobile only. That is also why we don't see excavator-based 8C FX CPUs. the frequencies would be too low for compensating for the IPC gain over Piledriver and the overclocking headroom would be ridiculous.

Expect something similar for 14LPP (Low Power Plus) vs 28SHP (Super High Performance). 14LPP will be much better for frequencies around 1--2GHz then will be a crossing point and it will be poor than the former node. The million dollar question is where is the crossing point. Not even AMD engineers knows this, only Globalfoundries engineers have precise knowledge. I guess the crossing point is somewhat above 3GHz and this is why I don't expect 3.6GHz base clocks.

It is worth to mention what The Stilt said once AMD announced clocks for the RX480 at Computex. He was expecting much higher clocks:

If the final clocks for RX 480 are indeed as low as 1266MHz, it quite honestly doesn't look too good for 14nm LPP

If AMD isn't able to push the clocks significantly higher (1500MHz+) without blowing the power draw through the ceiling, that's extremely alarming IMO.

AMD partners have shipped Tahiti and Bonaire based cards clocked to 1150 - 1200MHz from the box, and these GPUs were built on TSMC 28nm HP process... ...Which is thought inferior to GlobalFoundries 28nm HPP for example. Not to mention that Bonaire for example is able to clock close to 1400MHz on air cooling.

So if AMD is really unable to ship a ~232mm² GPU clocked higher than 1266MHz, then my fears about the 14nm LPP have materialized even worse than I anticipated

Meanwhile nVidia was able to clock a 35% larger (232 vs. 314mm²) GPU 50% higher (1266MHz vs. 1898MHz), while only having 20% higher TDP

Just to add a comment here.

The 480 was designed specifically with the goal of PPW and maximizing that PPW. There will be cards more power hungry than that, and the card will still be able to overclock past 1500-1600 from what I am hearing.

What he is stating is that he was expecting AMD to get higher clocks in the same power consumption slot. Aka he was expecting a higher PPW (Performance Per Watt). Then he goes further and claim that the problem must be in the 14LPP node because the competence (on 16FF node) could hit 50% higher clocks on a 35% larger die with only a 20% TDP penalty. I.e. competence improved PPW by about 69%.

 

[juanrga]

Current Zen silicon only hits 2.4GHz, but I expect them to get clocks higher on final silicon. Said that, I doubt that they will break the 3GHz barrier.

Is this article FUD? Zen ES @ 3GHz. Seems like it claimed 3GHz two months ago and you have a reply in the comments about 3.5GHz turbo.

My 3.5GHz claim therein was about turbo and I was replying to a poster that expected a "14nm 5ghz monster".

My claim here about breaking the 3GHz barrier is about base clocks. The 2.4GHz is also a base clock.

The article you mention is based in another article from an Italian site. The Italian author has been publishing leaks in its own site and in forums. He leaked inexistent up to 20-core Zen FX and Opteron CPUs, HBM included on Carrizo, Bristol Ridge coming with 16CUs, and other wrong stuff. He still claims >4GHz clocks for 8C Zen.

 

[sleeper357]

cdrkf is was a rhetorical? And would the 40% ipc increase be based on amds latest excavator core ? athlon 845 not the 9590

 

[cdrkf]

cdrkf is was a rhetorical? And would the 40% ipc increase be based on amds latest excavator core ? athlon 845 not the 9590

Ah sorry 😛 Yeah they've said 40% IPC over Excavator core, so more than 40% over Piledriver in the desktop FX cpu's (although it will depend on application).

 

[juanrga]

Zen ES has 2.4 clockspeed at the moment. At this same point in the lifecycle of BD/PD, ES clocks were the same.

Food for thought...clocks will improve.

BD ES clocks weren't the same. First BD silicon was clocked at 2.8GHz.

Clocks will improve, sure. The question is how much? If we take the original BD clock (2.8GHz) and final silicon clock (3.6GHz) and assume that improvement will be proportional then we obtain 2.4 / 2.8 x 3.6 = 3.1GHz for Zen.

 

[ComputerSecurityGuy]

That's not bad compared to Broadwell-E, if we assume that. Which is what a six or eight core Zen would be competing against.

 

[8350rocks]

Current Zen silicon only hits 2.4GHz, but I expect them to get clocks higher on final silicon. Said that, I doubt that they will break the 3GHz barrier.

Is this article FUD? Zen ES @ 3GHz. Seems like it claimed 3GHz two months ago and you have a reply in the comments about 3.5GHz turbo.

Yeah, basically...

I know the last ES I heard about was 2.4. If they have one at 3 GHz, that would certainly be news to me....and I would want to know.

If they really had a 3 GHz ES, that would legitimately place claims for top end clocks around 4 GHz into the realm of possibility. I do not expect to see that happen. Once the process is more mature, possibly close to 4 GHz if it turns out quite well.

Zen ES has 2.4 clockspeed at the moment. At this same point in the lifecycle of BD/PD, ES clocks were the same.

Food for thought...clocks will improve.

BD ES clocks weren't the same. First BD silicon was clocked at 2.8GHz.

Clocks will improve, sure. The question is how much? If we take the original BD clock (2.8GHz) and final silicon clock (3.6GHz) and assume that improvement will be proportional then we obtain 2.4 / 2.8 x 3.6 = 3.1GHz for Zen.

At this point it could possibly end up there...I am told they are still shooting for 3.5 base clock. Though, what they finally arrive at remains to be seen. I safely expect north of 3 GHz, by how much is the only true question.

 

[juanrga]

What frequency supports the memory controller on Bristol Ridge? 2400MHz? 3600MHz?