• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

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

More information about Zen from AMD at Hot Chips 28: http://hexus.net/tech/news/cpu/96112-amd-provides-information-zen-hot-chips-28/

At the Hot Chips symposium yesterday, AMD's Mike Clark talked attendees through the features of the Zen architecture using these same slides. Clark indicated that the top end of the Zen power envelope had been pushed down to under 100 watts. The power optimisation means that the Zen architecture can be used for products from <25 watts all the way up to about 100 watts. AMD's product range with thus be simpler.

The heralded "40 per cent generational improvement in instructions per clock" comes from a trio of contributory factors according to AMD. As you can see in the slide below, AMD has implemented a better core engine, a better cache system, and a lower power design. It is also not to be forgotten that AMD is moving from a 28nm to a 14nm LPP fabrication process at GlobalFoundries.

AMD's 'clean-sheet' Zen design borrows little from the firm's current range of processors. In the Zen CPU-Complex four cores share an 8MB L3 cache but are truly independent. Most of the CPU caches are said to be faster with the L3 bandwidth five times greater than in current AMD CPUs.

Zen uses SMT (simultaneous multi-threading) similar to current Intel processors. The slide below shows how the SMT system works in a block diagram.

The new Zen architecture looks competitive on paper and the single AMD comparative rendering benchmark win, against a current high-end Intel processor, is encouraging. However, it looks like it will be several months before we get the first AMD Zen chip(s) in the HEXUS labs.

So high end desktop Zen is under 100w, which is pretty insane for 8 cores/16 threads, assuming it's decently clocked as well.
 
More information about Zen from AMD at Hot Chips 28: http://hexus.net/tech/news/cpu/96112-amd-provides-information-zen-hot-chips-28/



So high end desktop Zen is under 100w, which is pretty insane for 8 cores/16 threads, assuming it's decently clocked as well.

Well, something's got to give right? So I don't see Zen clocking as high as Intel but as long as the power envelope is smaller, it'll be a hell of a chip for server farms. It'll also be an amazing choice in its APU flavor for a laptop chip...
 
Well, something's got to give right? So I don't see Zen clocking as high as Intel but as long as the power envelope is smaller, it'll be a hell of a chip for server farms. It'll also be an amazing choice in its APU flavor for a laptop chip...

Yeah, regardless of the debate over whether 14 LPP or 16FF+ is the better process, we can be certain Intel's 14nm is better than both of those.

So not to be too pessimistic, but if the 8c/16t is 95W then it won't be clocked high out of the box.

Although TDP is a bad indicator of overclockability. And also despite Intel saying Broadwell-E is 140W, I believe the i7 6850k consumes about the same amount of power as the i7 6700k.
 
Yeah, regardless of the debate over whether 14 LPP or 16FF+ is the better process, we can be certain Intel's 14nm is better than both of those.

So not to be too pessimistic, but if the 8c/16t is 95W then it won't be clocked high out of the box.

Although TDP is a bad indicator of overclockability. And also despite Intel saying Broadwell-E is 140W, I believe the i7 6850k consumes about the same amount of power as the i7 6700k.

I'm highly suspect of TDP, but for argument's sake, let's assume the Zen chip is at 100 and the 6900k (to which it was compared) is at 140.

Even with a 3GHz clock for Zen, the 6900k would be just 23% faster at 3.7GHz (that's based on a linear scale as 3.7/3.0=1.23333) given that their IPC is on-par.

This means that 10 Intel chips would perform similar to 12.3 Zen chips. But for 1400 watts I could run 14 Zen chips and thus get even more processing power for the same consumption. The TCO would work out in favor of Zen.

There's a lot of conjecture at this point, but it looks like Zen is indeed bringing back competition!
 
I'm highly suspect of TDP, but for argument's sake, let's assume the Zen chip is at 100 and the 6900k (to which it was compared) is at 140.

Even with a 3GHz clock for Zen, the 6900k would be just 23% faster at 3.7GHz (that's based on a linear scale as 3.7/3.0=1.23333) given that their IPC is on-par.

This means that 10 Intel chips would perform similar to 12.3 Zen chips. But for 1400 watts I could run 14 Zen chips and thus get even more processing power for the same consumption. The TCO would work out in favor of Zen.

There's a lot of conjecture at this point, but it looks like Zen is indeed bringing back competition!

Which, following your "for the sake of argument" assumptions, would bring us onto this - how many chips can AMD produce and how fast? If they could pull off your hypothetical (and maybe they can), then production becomes the big question.
 
TDP is not wattage used. My FX9590 varies considerably from 45+ up to 170+ depending on cores used, voltages and loading.

If, as I hope that the Zen will have all overclocking routes left open like the FX, the user will be able to run at higher clocks by adjusting clocks, voltages and cooling. That is why we are here after all.
 
A 4770k is not a 'midrange' chip. :/ At the time, the 4770k was the best CPU you could buy and was only overtaken by its refresh, the 4790k due to better thermal paste(seriously, that's the only difference).

Not quite, there was a couple of features that got disabled for Haswell K series processors, TSX & VT-d, They we're re-enabled for the 4790k so basically it was no different except better thermals and clocks for the gamer, But, if you do tasks that use the other techs the 4790k was the better option.
 
not condemning them i have had amd cpus systems for years still have one.its just in this day and age and tech you don't jump in front without serious changes or big money.

we all know where AMD sit in the same position of 10-15 years where they have sat. what was the last competitive cpu realistically ?

there is your answer. for budget people it will be ideal but for top end intel hold it.

would i buy a amd cpu ? yes in a heart beat if it was in the right budget for what i require.

i have no loyalty to brands only what is best for my money.cpus from amd i owned.

amd 2000xp , 3000xp , 939 3500 , 940 3800x2 , 5200 x2 , 940 x4 , 955x4,

so i dont mind amd just we wont get a revelation when its out.


You're either trolling or delusional. Intel have done literally nothing since Sandy Bridge. They haven't had to, but AMD have more than caught up by the look of it, and ought to take huge bites out of Intel's market share as a result.

They didn't even need to take such a big jump to win back a lot of business. Most don't realise, but Excavator beat Haswell in raw integer performance, and wasn't too far behind in FP, it just had major latency issues as with all Bulldozer family products.

Which ever way you spin it, beating Broadwell-E core-for-core, thread-for-thread and clock-for-clock in Blender means Zen will be very quick. Indeed, given that Summit Ridge is only 2 channel DDR-4 and Blender ought to benefit from Broadwell-E's 4 channel fairly significantly, I'd say it's a stupendous result ....

Yeah I get the requirement to have comparable clock-for-clock for comparison reasons.

But to only get 3ghz out of a engineering sample? Seems a bit low to me.

Stop the FUD. That's very high for an 8 core, 16 thread engineering sample.

Many of the Broadwell-E engineering samples were in the low 2.xGhz, some below 2.
 
Which ever way you spin it, beating Broadwell-E core-for-core, thread-for-thread and clock-for-clock in Blender means Zen will be very quick. Indeed, given that Summit Ridge is only 2 channel DDR-4 and Blender ought to benefit from Broadwell-E's 4 channel fairly significantly, I'd say it's a stupendous result ....
I highly doubt AMD put 4 RAM sticks in the Broadwell-E machine when theirs only needed 2.
 
I highly doubt AMD put 4 RAM sticks in the Broadwell-E machine when theirs only needed 2.

I highly doubt they didn't, but the bandwidth in broadwell is certainly over exaggerated in terms of it's usefulness and likely wouldn't have made any difference in the result.

http://www.pcworld.com/article/2982...ing-truth-about-their-performance.html?page=2

Maybe not the best source but the only applications that benefit are those shifting through a vast amount of data like a rar file and doing extremely basic operations. When you try and do complex operations on smaller files(which is most work that doesn't come down to basic file transfers) bandwidth stop making that much of a difference.

http://www.anandtech.com/show/8959/...-3200-with-gskill-corsair-adata-and-crucial/4

http://www.anandtech.com/show/8959/...-3200-with-gskill-corsair-adata-and-crucial/5

This isn't comparing dual/quad channel, but still effectively a what almost 50% bandwidth gain from slowest to fastest. Again we see that Cinebench gains almost nothing, winrar has modest gains, anything that is almost solely focused on throughput is faster, almost anything that is relatively computationally heavy is not remotely bandwidth limited. This doesn't prove that removing 50% of the bandwidth wouldn't produce a bandwidth limit but combined with the other link and you can see in most cases it really doesn't.

Thing is most heavy throughput stuff is either moving files around or unraring, or server level duty of dealing with way more things at a time, very little applies to home use. The reason x99 is popular is really the extra cores, not the bandwidth. For Blender there would likely be almost no performance difference as with cinebench... but ultimately Zen doesn't have quad channel, produces that performance level and isn't bandwidth limited so why would Haswell be unless it was far less efficient with memory/cache which isn't likely. Also key is, AMD is just showing it matching the x99 platform chips... the reality is it will be competing with quad core i7's in the lower price range higher mainstream area.
 
The higher core versions might be quite competitive with Broadwell-E but the lower core count versions will be fighting a modified Skylake,aka,as Kaby Lake and the top Skylake SKUs are running at around 4GHZ.

The 4C/8T chips will only have one CCX active,but it will be interesting to see if AMD can get close enough to the clockspeeds Kaby Lake is targetting though.
 
Seeing as bulldozer was brought up, what exactly happened with that cpu? I remember after it was launched talk going round of what they produced wasn't what it was supposed to be and a lot of changes had to happen top even get it out at all? Almost as if they got so far into making it, realised certain elements weren't feasible and then had to go with what they could at the time. :confused:
 
Seeing as bulldozer was brought up, what exactly happened with that cpu? I remember after it was launched talk going round of what they produced wasn't what it was supposed to be and a lot of changes had to happen top even get it out at all? Almost as if they got so far into making it, realised certain elements weren't feasible and then had to go with what they could at the time. :confused:

I dont think it was that bad. I do feel that most applications do not get tweaked enough to cater to the strengths from what I remember seeing. Basically intel instruction set takes preference so hard to compare its intended performance when code is optimized in such a way.

It certainly felt that way in the piledriver series.
 
Last edited:
Not all software benefit from parallelism and especially games, just look at the consoles for evidence of that.. they have 8 cores but really slow ones and a lot of games are barely managing a steady 30fps even with heavy console specific optimization and low level API's.

Processor manufacturers need to find the right balance between number of cores and overall core speed, Intel have pretty much nailed it with their quad core line in recent years whereas AMD went too far over to number of cores.
 
Last edited:
One of the best things about the Zen platform will be the high number of PCIe lanes. I think it has enough for 4x PCIe 3.0 x16 which may be important with PCIe 4 arriving next year, somewhat indicating that PCIe 3.0 is likely to become restrictive sometime soon. Running crossfire on a skylake system at x8 a card isn't appealing on a base system I want to keep for at least 5 years. The premium Intel charge for PCI lanes on the x99 platform is also an unattractive prospect. It's a shame 4.0 won't be ready for Zen but I hear there may be some pre specked instances beforehand that a bios upgrade may effectivly make into 4.0. Meh, Full 3.0 x16 will be fine, I do wonder about crossfire running over the PCIe lanes also being an issue down the line.

Is there a confirmed spec sheet at this point? It's a while back that I saw slides and such.
 
not condemning them i have had amd cpus systems for years still have one.its just in this day and age and tech you don't jump in front without serious changes or big money.

we all know where AMD sit in the same position of 10-15 years where they have sat. what was the last competitive cpu realistically ?

there is your answer. for budget people it will be ideal but for top end intel hold it.

would i buy a amd cpu ? yes in a heart beat if it was in the right budget for what i require.

i have no loyalty to brands only what is best for my money.cpus from amd i owned.

amd 2000xp , 3000xp , 939 3500 , 940 3800x2 , 5200 x2 , 940 x4 , 955x4,

so i dont mind amd just we wont get a revelation when its out.

It is gutting the last great AMD cpu they had was single core. Barton 3200+, I recall paying £320 in 2003?

I only had Thunderbird 1.3 on Gigabyte VIA then 1.4 on nForce, Athlon XP 2200+, two Barton 3200+ on two NF7-S v2, Athlon 939 4600+x2 and last was Phenom II 1090t x6.


My memories of the best of AMD will be the Quake 3/mods years vs Pentium 3.
 
Back
Top Bottom