AMD Prank King 2016, FAIL!!

[SinOfLiberty]

[Tim]

All I'll say is, AMD better find a way to crank up the clock speeds or get boost working correctly in the final silicon. Not impressed with the games, in the encoding department it seems to be doing as advertised, however, not so good in the power draw.

[AliG]

Can you provide the source on this?

Also if I'm reading the graph correctly (which is a big if, it's not very clear), it appears the listed frequencies are 3.15/3.4 GHz. I'm under the impression that AMD is bumping up the base clock to 3.4 GHz, and boost higher from there. Even if I'm entirely wrong on the frequencies, the fact it slots between the 6800k and 6900k means it'll still take a lot of server market share if the pricing rumors are true. If I'm right, then it should be directly inline with the 6900k.

Not discounting their results, these could be very real ES benchmarks. Just trying to get a better picture of where production silicon should fit.

[AliG]

Found the source:
http://www.overclock.net/t/1619110/c...zen-benchmarks

For the record, clocks are 3.1/3.3 with this sample - which actually paints a very different performance picture. Most people on the other forum seem pretty optimistic (with the expectation the clocks will be 3.4 GHz+).

[drmrlordx]

Canard PC/Doc Teraboule.

[EternityZX9]

Exciting times if true...but I'm not going to believe anything until legit reviews come out.

[Tim]

Exciting times if true...but I'm not going to believe anything until legit reviews come out.

I suppose it depends on what people use their PCs for, but when it comes to the games (what I mainly use my PC for) I can't be more disappointed. That power draw is a bit nasty too.

[AliG]

I suppose it depends on what people use their PCs for, but when it comes to the games (what I mainly use my PC for) I can't be more disappointed. That power draw is a bit nasty too.

Why?

You can't possibly expect an underclocked octo-core to compete against the 6700k? Not even the 6900k does as well in games.

Wait for the quadcore cutdown versions for gaming comparisons.

[zalbard]

Why?

You can't possibly expect an underclocked octo-core to compete against the 6700k? Not even the 6900k does as well in games.

6900K can be easily overclocked to similar levels, though.

IMO, Intel artificially limits the clock rates of these chips.

[vario]

Sooo.... Whats with the title ?
Given that this ES Sample has lower clocks than the "advertised" in Zen promo results are inline with what was shown ...
MT Performance inline with 1000$ chip .
Gaming Performance around Haswell with same frequency ...
So whats exactly bad ?
Am i missing something ?
Did people expect MT performance of Broadwell-E AND AT THE SAME TIME ST performance of a full blown skylake ?

[vario]

6900K can be easily overclocked to similar levels, though.

IMO, Intel artificially limits the clock rates of these chips.

I was under the impression that Broadwell-E doesnt overclock particurarly well, worse than haswell-E even.
But yes, if AMD wants to compete on not default settings, they must be able to do at least 4.0ghz.

[AliG]

6900K can be easily overclocked to similar levels, though.

IMO, Intel artificially limits the clock rates of these chips.

I don't think it's fair to compare overclocked results vs a stock 6700k. 6700k can easily be overclocked to 4.5 GHz, with good chips hitting 5 GHz. With that logic, there's no point to the 6900k because it can't hit 5 GHz 24/7.

2 very different chips with different target audiences. I think it's perfectly fine to expect quadcores to compete in gaming, and octocores to be kings of server computation.

[BeepBeep2]

I don't see anything fail about these benchmarks.
Rumors are 3.15 base, 3.3 Turbo / 3.5 single core turbo, same OPN as the Cinebench leak that was cropped and called out as fake (I had a hunch that it might not be)

Launch speeds are going to be 8% higher, and the results here are normalized to the 6600K.
At 3.4 GHz I suspect that productivity "percentage" number to be in the 180's, gaming at least @ 6800K level, and power consumption roughly equal to the 6900K.

Looks fine to me.

[kromosto]

It all depends on pricing...

[CrimInalA]

With AMD it always seems to depend on pricing , never on performance

[kromosto]

Actually amd havent offered anything for me even they were not price dependent.

[Chickenfeed]

I was under the impression that Broadwell-E doesnt overclock particurarly well, worse than haswell-E even.
But yes, if AMD wants to compete on not default settings, they must be able to do at least 4.0ghz.

Based on what I've gathered the 6/8 core Broadwell generally do at least 4.2 and good samples top out at 4.4. 4.5+ generally takes excessive voltage and heat dump goes through the roof. Haven't had my CPU up and running long enough to fully collaborate this but seems about right.

I'm with you on Zen needing to have head room at 4+. I'll be bummed if the 16 thread part can't crack 4 but without a doubt the 8s should. Turbo on the mid range parts will likely flirt with 4 on their own.

If the mid range parts can sneak up to haswell level overclocks and the price is right, they'll give x99 and upcoming X299 a good kick in the pants. Should have most of the benefits of intels HEDT and a lower platform cost. I wrestled with wether it was worth it to me to go broadwell e over skylake but in the end the added pcie lanes and 4 threads sold me. Zen should hopefully make that an even more interesting situation in the near future.

[vario]

Based on what I've gathered the 6/8 core Broadwell generally do at least 4.2 and good samples top out at 4.4. 4.5+ generally takes excessive voltage and heat dump goes through the roof. Haven't had my CPU up and running long enough to fully collaborate this but seems about right.

I'm with you on Zen needing to have head room at 4+. I'll be bummed if the 16 thread part can't crack 4 but without a doubt the 8s should. Turbo on the mid range parts will likely flirt with 4 on their own.

If the mid range parts can sneak up to haswell level overclocks and the price is right, they'll give x99 and upcoming X299 a good kick in the pants. Should have most of the benefits of intels HEDT and a lower platform cost. I wrestled with wether it was worth it to me to go broadwell e over skylake but in the end the added pcie lanes and 4 threads sold me. Zen should hopefully make that an even more interesting situation in the near future.

Thing is, AMD is going to use defective (or non defective if they will have more orders for them) 8 core chips to make 4 and 6 core chips.And hopefully they will have some sort of automation to select stronger cores , but it could mean they wont clock much higher, as FX4xxx series was only marginally better clocker than the 8xxx .I hope im wrong, but it seems the process is still being tuned in.All in all it could mean 4 core Zens can be similar clockers to the 8 core ones, with something along the 100-200Mhz uplift.Still, it would grant them decent standings in the not overclocked chips, but it wont be enough to close the gap to the K chips.
Historically speaking, AMDs SECOND attempt at a chip was much more successful , tbred B, X2 F3s , phenom C3 and some others comes to mind ;-) .
However they can again do what i call AMD promotions, unlocking lost cores ...

[Particle]

I'm not yet convinced that AMD is going to offer 6-core SKUs. They have historically offered products in core counts matching a multiple of the base module size. With Stars that was one core so we saw 2, 3, 4, and 6 core offerings over time. With BD/PD/SR/EX that was two cores so we saw 4, 6, and 8 core offerings. With Zen that will be four cores, so we may only ever see 4 and 8 core products depending on if they stay with that trend.

[FlanK3r]

Its solid CPU, this was ES with lower clocks, not finals silicon And of course, the clocks are lower, because CPU is not high clocking Kaby. But the oponent is example i7-6850K.

[vario]

I'm not yet convinced that AMD is going to offer 6-core SKUs. They have historically offered products in core counts matching a multiple of the base module size. With Stars that was one core so we saw 2, 3, 4, and 6 core offerings over time. With BD/PD/SR/EX that was two cores so we saw 4, 6, and 8 core offerings. With Zen that will be four cores, so we may only ever see 4 and 8 core products depending on if they stay with that trend.

Well AMD has 3 tiers SR3 SR5 and SR7 .Only one die fabricated initially .It would be dumb not to follow Intels way of having 4-6-8 cores (yes intel has also 2 and 10 and many more but they are not competing in these segments for now) .Im pretty sure AMD can disable whichever cores they like on a chip , also doesnt that mean that 4 core versions will have larger L3 cache per core ? Could be interesting.
Of course Sr5 could be just HT disabled part :-/ with slashed L3.Your way of thinking can be backed up little by this french article, they had 4 samples, 2 4 core and 2 8 core ones...
http://www.guru3d.com/news-story/french-magazing-posts-engineering-sample-amd-ryzen-processor-benchmarks.html

Anyhow, i hope for 6 core parts, OR attractive 8 core price because its gonna be tough to compete without 6 core one.
Maybe 8 core NO HT part is the supposed 300-400$ one.

[BeepBeep2]

Are we sure there is only one die? There are quad core ES out there AFAIK. If AMD thinks they can competitively sell the quad core parts I don't think they want to cannibalize margins by selling gimped 8 core dies.

[qcmadness]

Are we sure there is only one die? There are quad core ES out there AFAIK. If AMD thinks they can competitively sell the quad core parts I don't think they want to cannibalize margins by selling gimped 8 core dies.

Taping out a new 14-nm design is very costly.

[BeepBeep2]

Taping out a new 14-nm design is very costly.

Sure it is, but we've already seen quad core ES (allegedly existing), and AMD has positioned itself to sell a lot of these if the performance is good.
I don't think AMD would want to sell 8c/16t ~200-250mm dies when they could sell 4c/8t dies at half the area and greatly increase margins on the lower end of the addressable market. What would it cost AMD to create the 4 core die, another $100m? It seems to me like that would be the smartest choice if they are expecting several billion USD in revenue.

Expensive wafers on an expensive process, fitting twice as many dies on a wafer reaps rewards with high volume. Not everyone is going to want a $300-600 8c/16t, I think a lot of people will also be interested in a $150-250 4c/8t.

[AliG]

Taping out a new 14-nm design is very costly.

True, but finalizing transistors leakage and density is the hard part, not necessarily scaling chip size. I would assume that if an octo-core can be made with reasonable yield (i.e. they can pack enough efficiently running transistors into the desired wafer space), cutting the design down to a quadcore with less cache should be rather doable.

Sure it is, but we've already seen quad core ES (allegedly existing), and AMD has positioned itself to sell a lot of these if the performance is good.
I don't think AMD would want to sell 8c/16t ~200-250mm dies when they could sell 4c/8t dies at half the area and greatly increase margins on the lower end of the addressable market. What would it cost AMD to create the 4 core die, another $100m? It seems to me like that would be the smartest choice if they are expecting several billion USD in revenue.

Expensive wafers on an expensive process, fitting twice as many dies on a wafer reaps rewards with high volume. Not everyone is going to want a $300-600 8c/16t, I think a lot of people will also be interested in a $150-250 4c/8t.

I agree with this. You may see some 8 cores that don't meet voltage spec be cut down to 6 cores to improve manufacturing yield, but that's assumes AMD has a horribly inefficient manufacturing process to not have a single native quadcore chip.