New Zen microarchitecture details

[crashtech]

Seeing as AMD once had CPUs that bettered Intel's, the statement is rather puzzling. Maybe most of the intended audience is unaware of that history.

 

[CentroX]

Seeing as AMD once had CPUs that bettered Intel's, the statement is rather puzzling. Maybe most of the intended audience is unaware of that history.

bettered Intel in performance, not marketshare.

 

[crashtech]

bettered Intel in performance, not marketshare.

Which is all that should matter in a CPU forum!

 

[Insomniator]

So 6 more months for a CPU 'competitive' with 18 month old SkyLake? Plus APU performance at the level of 4 year old consoles, which has also been done by ~$90 750ti's for 2-3 years by then. Sweet!

I just don't get their target segment. Its like some niche ultra budget area where saving 30 bucks makes all the difference in the world.

 

[Dresdenboy]

bettered Intel in performance, not marketshare.

Depends on market:

http://www.theregister.co.uk/2004/09/09/amd_vs_intel_retail/

 

[deasd]

Damned if you do, damned if you dont. There is no winning for AMD.

I finally have to agree with you when I read some new posts above...... this thread should be closed now.

 

[MiddleOfTheRoad]

bettered Intel in performance, not marketshare.

Nope.

The Athlon's performance advantage gave AMD over half the retail market at one point:
http://www.geek.com/chips/amd-outsells-intel-in-us-retail-market-560369/

Those days are behind them -- but the revisionist history on this forum is just stupid.

 

[monstercameron]

I finally have to agree with you when I read some new posts above...... this thread should be closed now.

Yeah, the anti hype killed it. Im done.

 

[tential]

How isn't Zen relevant?

I don't care if it's faster than Skylake or whatever on a per core basis. It's already faster than what's out now in the FX lineup which is fast enough for gaming.

It's not like gaming CPU requirements are going up. So then that leaves core count....

So Zen is already 6-8 cores.
Intel would need to boost its next line to at LEAST 6 cores to be competitive. 8 cores 16 threads is quite interesting...

There's no way I get a 4 core/8 thread intel CPU over an 8 core 16 thread AMD CPU if AMD has over sandy/ivybridge performance or even on par with that with updated instructions. At that point, I want the extra cores. I can literally have 2 PCs through VM with the AMD solution which is very attractive to me.

 

[Killrose]

Hopefully we get a ZEN "nip slip" at Computex. If we don't I fear Zen will just end up being better than the current AMD line-up but still far, far short of Intel's current, let alone incomming CPU's.

I like my FX8320e@4.5ghz turbo. It does everything I need it to do. Does good for price performance, but it would be hard to recommend for a new build.

 

[crashtech]

Yeah, the anti hype killed it. Im done.

You have to admit that AMD has a formidable challenge to basically leapfrog the performance of several generations of its competition's CPUs. Even many who wish AMD to succeed are justifiably skeptical. Personally I hope they knock it out of the park, it would be a real gas to have an AMD system again like I did all through the early to mid 2000's.

 

[scannall]

You have to admit that AMD has a formidable challenge to basically leapfrog the performance of several generations of its competition's CPUs. Even many who wish AMD to succeed are justifiably skeptical. Personally I hope they knock it out of the park, it would be a real gas to have an AMD system again like I did all through the early to mid 2000's.

Just being several nodes ahead of when BD was introduced is a running start. Not to mention, knowing what they screwed up with that design.

I just want to see what it can do, because I'm a tech junkie. If forced to make a prediction, it'd be roughly Ivy Bridge IPC. But hey, it's about as accurate a guess as anyone else. And if I'm wrong either way I won't feel bad.

It might be fun to see AMD do something unexpected in 2017. Introduce a Zen APU with 16Gb of HBM2 ram for example. That one might actually change the game a bit.

 

[frozentundra123456]

How isn't Zen relevant?

I don't care if it's faster than Skylake or whatever on a per core basis. It's already faster than what's out now in the FX lineup which is fast enough for gaming.

It's not like gaming CPU requirements are going up. So then that leaves core count....

So Zen is already 6-8 cores.
Intel would need to boost its next line to at LEAST 6 cores to be competitive. 8 cores 16 threads is quite interesting...

There's no way I get a 4 core/8 thread intel CPU over an 8 core 16 thread AMD CPU if AMD has over sandy/ivybridge performance or even on par with that with updated instructions. At that point, I want the extra cores. I can literally have 2 PCs through VM with the AMD solution which is very attractive to me.

Well, that depends on price. Lets look at some numbers (speculative of course). Say Zen has 8 cores at a 25% single thread deficit (considering both ipc and overclocking) to BW-E/Skylake-E. That is equivalent to 6 core Intel in MT. Now if AMD provides intel hex core performance at 300.00, then obviously, it is an attractive buy, if one can use all the cores. However, I totally disagree with the AMD fans that think they will price Zen this way. If zen is competitive with hex core intel, which is what I expect actually, slightly better in multi and lower in ST, I expect the price to be close to the 5820k or whatever the equivalent intel chip is at the time. Actually, for most users, I still think 8 cores is somewhat overkill. For a heavy user, I think hex core with high per core performance is still the sweet spot.

 

[ElFenix]

http://www.microcenter.com/product/437203/Core_i7-5820k_33_GHz_LGA_2011-V3_Tray_Processor

 

[Abwx]

I'm not suggesting anything. I'm just interpreting the data AMD themselves has provided.

AMD themselves say 40% more IPC over Excavator core. They also say that Excavator based Carrizo for example has four cores. So based on this, what exactly indicates that Zen core would have higher performance than a Excavator compute unit (two cores, according to AMD).

http://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9Mjg0ODkyfENoaWxkSUQ9LTF8VHlwZT0z&t=1

http://products.amd.com/en-us/search/APU/AMD-FX-Series-Processors/AMD-FX-Series-Processors-for-Laptops/FX-8800P-with-Radeon™-R7-Graphics/123

If you dont mind let s use CB 11.5 rather than R15, this latter is a suspicious update that saw AMD chips losing 10% in respect of 11.5 when compared to Intel CPUs, it s enough that CB is Intel optimised so we wont add another "optimisation" over one that already brought 15-18% advantage to Intel in respect of CB R10 scores...

What matter is the FP throughput of a module since a single EXV core can extract only 60% of the FPU capabilities, as you noticed a single thread is served by two ALUs, a Zen core is nothing else than a fused EXV module where the four ALUs are used in a single core, hence this core should have the same (an even better looking at Mathias analysis..) throughput when loaded with two threads, this imply that ST perf in CB will be improved by much more than 40%..

I've blathered on about this particular subject too much already, but seriously, if AMD takes a core that is as wide as - if not wider than (fp) - an XV module, improves the cache subsystem, rebuilds/improves the front end, AND adds SMT on top so that the core can either split resources between one or two threads like an Intel CPU and yet somehow manages to produce less throughput per thread with the new core compared to an XV module, then AMD is beyond pathetic. Consider that a hypothetical 8m XV running @ 3.4 GHz should score anywhere from 1225-1376 assuming it uses cores identical to those in Carrizo, which it probably wouldn't (it would have L3, probably bigger L2, and maybe some other goodies, or maybe not).

 

[guskline]

http://www.microcenter.com/product/437203/Core_i7-5820k_33_GHz_LGA_2011-V3_Tray_Processor

WOW! Broadwell E is really close to release!

 

[superstition]

Threadcrapping and trolling are not allowed
Markfw900

this thread should be closed now.

Yeah, the anti hype killed it. Im done.

I, for one, am still interested in the subject of Zen's architecture. I would appreciate it if people would heed the moderator's notes.

Hype is mostly relevant for marketing concerns, not architectural ones.

 

[Soulkeeper]

Anyone know the dates/events we are waiting on to get more information ?

 

[rtsurfer]

WOW! Broadwell E is really close to release!

NDA up May 27th.

Spoiler alert, clocks like crap.

 

[Dresdenboy]

Anyone know the dates/events we are waiting on to get more information ?

AMD might talk about Zen at Computex on June 1st. They announced the launch of 7th gen APU (BR), info about Polaris and more.
http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=2170028

 

[Arachnotronic]

AMD might talk about Zen at Computex on June 1st. They announced the launch of 7th gen APU (BR), info about Polaris and more.
http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=2170028

Why do they say "Polaris updates"? Why not "Polaris Launch" like they do with 7th gen APUs?

 

[Justinbaileyman]

When is the Bristol Ridge launch happening? Next month maybe or am I way off??

 

[LTC8K6]

When is the Bristol Ridge launch happening? Next month maybe or am I way off??

Computex.

 

[Phynaz]

Why do they say "Polaris updates"? Why not "Polaris Launch" like they do with 7th gen APUs?

Because they aren't launching?

 

[Dresdenboy]

Because they aren't launching?

Exactly that.

This will be a problem if GTX 1070 becomes available months earlier. So far AMD has mss gains in all segments not addressed by GTX 1080.

Some site even wrongly assumed that we will hear about Zen APUs at Computex...