AMD "Summit Ridge" Silicon Reserved for 8-core CPUs Initially

[doxology]

And then keep not sucking. That's going to be tough for a company in AMD's position.

That is kind of a sad graph because Nvidia has made some decent advances in different fields that last 5 years. Intel on the other hand, and while still the dominant CPU company, has really not been that impressive since AMD tanked with Bulldozer. Its as if they have just milked everything they can with very little upgrades each generation because they know they have no competition. There cheapest 4 core CPU the whole time has been around the $200 mark, and of course the fact that they haven't pushed the core count further during the whole time in the consumer market stuff. If Intel hadn't of been so dominant to begin with most people would of been complaining, but instead they look forward to the 5 or 6 percent increase each year. It has only just started getting to the point where an older Intel Core system should be upgraded, and that is mainly because of the things like PCI Express generational improvements, USB 3 and USB 3.1 or needing a USB C connector on your motherboard, or of course the switch to DDR4 memory. Those things will lead me to build my next PC rather than the incremental increases in performance on Intel's CPU's.

 

[Caring1]

But if AMD does that, everyone looses their minds!

To make loose, is to set free.
Constrain your brain folks, don't set it free.

 

[rruff]

Having more R&D doesn't always mean better products unluckily.

The thing is, AMD has become a very small company compared to their competitors, and that brings with it some serious handicaps.

 

[HD64G]

No. Want proof? Northwood vs Prescott, ivy vs Sandy or better yet another halfing of size 130nm clawhammer/Newcastle vs 65nm Brisbane/lima. There was no doubling of performance. Performance increase is based off of design changes not die size changes. Otherwise the current i7 would be what 2-3x as fast as the current models based off of die shrink alone.

The newer i-X generations benefit only the GPU part of the die and shrink the CPU part, so those aren't comparable examples to Zen vs BD. As for Northwood vs Prescott, the archi changes were awful and that's why the shrink didn't benefit them at all and forced them to go for another one in the next 2 years (Core 2). IPC is heavily benefited from shrinking manufacturing procedure if size per core remains the same even in same archtecture. Only case it doesn't is some bad-coded apps who adore only high clocks.

 

[Jism]

The thing is, AMD has become a very small company compared to their competitors, and that brings with it some serious handicaps.

Yes and no. It sold a couple of fabs, cause the costs to run / maintain fabs are enourmus. It just outsources the production of a chip and it can get good prices for that as well since AMD is doing multiple chips in PC & consolemarket. AMD can just focus on designing and selling chips, as it should.

Dont underestimate the underdog, as it did a couple of times before, beat both Intel and Nvidia with genius engineering. AMD needs good sales, and they are going on a different course wich hopefully does better then it did in the past.

The Bulldozer story was'nt a bad choice, but it was aimed at something that is'nt there yet. Going for more cores would have helped in for example video-encoding, but it lacked for programs / games that would require high single core performance.

They tackled that with putting in the turbo thing, you know, lower some core-speeds and up the first 2 or 4 cores with a few hundred MHz. In single-threaded that would work, but it's not the real answer to more IPC.

And by now, AMD released a CPU that should be on par with Intel now, and will only get better in the next refreshes, fixes and tricks engineers can do for that chip. AMD was always the better & more affordable vendor to go for in PC's. It still is. Yes their CPU consume some more power, but do you really drive your PC to run 24/7 at 100% usage that power would be a problem?

The money you save on AMD stuff can be invested in a better SSD, Ram or graphics card. Graphics cards as well; AMD just puts a proper product and if story's are true, we are going to have an exciting 2016 & 2017 coming up. I have my hopes and money set on AMD, and will go for another AMD AM4 upgrade.

My Thuban X6 at 4.2Ghz is still kicking ass, in many games, programs and is capable of doing the exact same thing Intel cpu's do.

 

[bug]

The newer i-X generations benefit only the GPU part of the die and shrink the CPU part, so those aren't comparable examples to Zen vs BD. As for Northwood vs Prescott, the archi changes were awful and that's why the shrink didn't benefit them at all and forced them to go for another one in the next 2 years (Core 2). IPC is heavily benefited from shrinking manufacturing procedure if size per core remains the same even in same archtecture. Only case it doesn't is some bad-coded apps who adore only high clocks.

What you fail to understand, is that the 40% (?) expected IPC improvement already takes into account the die shrink and the increased number of transistors.

 

[nunyabuisness]

AMD said Q2 they would send out samples. So not long before we see what it can really do.

 

[bug]

AMD has good and bad just like Intel. Remember Intel stuck with netburst just as long as amd stuck with fx

If only it was that simple. Netburst was bad, but people still bought it (in part because of intel's shenanigans, in part because they didn't know any better). And intel has alternative revenue sources. AMD's only alternative is GPUs. And people have no trouble switching to intel. And AMD is much smaller than intel (thus it feels the pain much faster).

 

[cdawall]

The newer i-X generations benefit only the GPU part of the die and shrink the CPU part, so those aren't comparable examples to Zen vs BD. As for Northwood vs Prescott, the archi changes were awful and that's why the shrink didn't benefit them at all and forced them to go for another one in the next 2 years (Core 2). IPC is heavily benefited from shrinking manufacturing procedure if size per core remains the same even in same archtecture. Only case it doesn't is some bad-coded apps who adore only high clocks.

First gen core 2 was 65nm same as Prescott netburst again that was a gain from architecture not die shrink. Die shrinks don't add performance they shrink the size. Why you can't get that through your head is beyond me.

 

[medi01]

Problem is they need to go from sucking to exceptional*

*Great porn title.

ROFLMAO

 

[HD64G]

What you fail to understand, is that the 40% (?) expected IPC improvement already takes into account the die shrink and the increased number of transistors.

I fail to understand, or most people assume it is like that? You might be correct but it isn't absolutely clear as Zen is a totally new arch targeting a big jump in IPC. And if they are to have their 8 core Zen CPUs at 100W TDP, just by the shrink from their 125W FX-8370 32nm to 14nm the performance should skyrocket either by letting them fill each core my many more transistors (most possible in order to gain more by SMT that needs bigger cores), either by upping clocks much more (less possible in 1st gen Zen), not only go up by 40% (+15% from BD vs Excavator) me thinks.

First gen core 2 was 65nm same as Prescott netburst again that was a gain from architecture not die shrink. Die shrinks don't add performance they shrink the size. Why you can't get that through your head is beyond me.

1) I talked about gains in size about the balance in APUs between the iGPU and the CPU sections. CPU is shrinking analogically to the gains in nm but iGPU is filling the die space gained by this, so it is logical for the CPU section not to be faster. If they needed to make it faster they could easily do so.
2) Even in same arch (45nm) AMD managed to make a great jump from Phenom II BE 955 X4 3.2GHz to Thuban X6 1100 3.3GHz in the same 125W. We are talking about a 50% more transistors in same procedure. So, why not to have more than 40% when we go to a much bigger core by the transistor count it will have?

And most important: All the above are pure estimations judjing from the past, so nothing sure, nothing to have tensions on it eh? Have fun speculating. October-November is Zen's launch. By then, all will be clear.

 

[cdawall]

I fail to understand, or most people assume it is like that? You might be correct but it isn't absolutely clear as Zen is a totally new arch targeting a big jump in IPC. And if they are to have their 8 core Zen CPUs at 100W TDP, just by the shrink from their 125W FX-8370 32nm to 14nm the performance should skyrocket either by letting them fill each core my many more transistors (most possible in order to gain more by SMT that needs bigger cores), either by upping clocks much more (less possible in 1st gen Zen), not only go up by 40% (+15% from BD vs Excavator) me thinks.



1) I talked about gains in size about the balance in APUs between the iGPU and the CPU sections. CPU is shrinking analogically to the gains in nm but iGPU is filling the die space gained by this, so it is logical for the CPU section not to be faster. If they needed to make it faster they could easily do so.
2) Even in same arch (45nm) AMD managed to make a great jump from Phenom II BE 955 X4 3.2GHz to Thuban X6 1100 3.3GHz in the same 125W. We are talking about a 50% more transistors in same procedure. So, why not to have more than 40% when we go to a much bigger core by the transistor count it will have?

And most important: All the above are pure estimations judjing from the past, so nothing sure, nothing to have tensions on it eh? Have fun speculating. October-November is Zen's launch. By then, all will be clear.

SIZE DOESN'T CHANGE PERFORMANCE. Adding transistors can THAT'S NOT A DIE SHRINK THAT's A CHANGE IN DESIGN. Do you need me to try this in a different language?

 

[rruff]

And by now, AMD released a CPU that should be on par with Intel now, and will only get better in the next refreshes, fixes and tricks engineers can do for that chip. AMD was always the better & more affordable vendor to go for in PC's. It still is. Yes their CPU consume some more power, but do you really drive your PC to run 24/7 at 100% usage that power would be a problem?

By the time we see it, I doubt it will match Intel. And where have those "genius engineers" been the last few years? Are they bringing Keller back?

I don't understand your "better and more affordable" comment. Last time I upgraded I didn't even look at Intel initially. I wasn't up on computer components but I was an AMD fan. They sure looked better on paper. Faster clocks and I could get a 6-core for the same price as a 2-core/2-HT Intel chip. Seemed like a no brainer. But when checked reviews and benchmarks I discovered the truth. The Intel chip would trounce AMD in any task running 4 threads or less, and AMD would barely come out on top at 6. If the power consumption under load was the only difference I wouldn't care, but it also sucked a lot more at idle. The electric difference worked out to ~$20/yr which is significant on a $100 processor. If I'd OC'd it to close the performance gap, it would have been even worse. Same thing happened with GPUs when Nvidia came out with Maxwell.

I think the brightest spot currently for AMD are the latest APUs. CPU performance isn't that good, but idle consumption is down and load power isn't that bad. For very casual gamers they handily beat Intel's iGPU, so they have a niche where they are the clear winner. Unfortunately it looks like the APUs are going to stay on 28nm for a good while yet.

 

[HD64G]

SIZE DOESN'T CHANGE PERFORMANCE. Adding transistors can THAT'S NOT A DIE SHRINK THAT's A CHANGE IN DESIGN. Do you need me to try this in a different language?

OK, calm down, no need to get hot headed. I understand what you say as it is too simple not to, you just do not get what I have already explained in detail by now. Think of previous gens between Athlon CPUs when being shrunk in process might help a bit. A few months away from the final result we are now. Peace. Over.

 

[medi01]

Speaking of AMD PR and why they are silent.

I know the yadayada 980Ti/TX OMG Fury didn't trounce them, but if you recall, that:

a) they never chased TX
b) 980Ti is a downscaled TX
c) at 4k Fury is competitive nevertheless

Actual product is competitive, at least, beats competition in certain (not so unlikely) scenarios.

Now, what the heck can your PR drum beat about, if all what you will be able to do is compete with mid range Intel CPU which is a couple of generations old? (if even that)

Only Apple could successfully (to an extent) twist reality in such situations. (not that they had to after switching from IBM)

 

[cdawall]

OK, calm down, no need to get hot headed. I understand what you say as it is too simple not to, you just do not get what I have already explained in detail by now. Think of previous gens between Athlon CPUs when being shrunk in process might help a bit. A few months away from the final result we are now. Peace. Over.

What you are saying has nothing to do with a die shrink.

 

[Aquinus]

OK, calm down, no need to get hot headed. I understand what you say as it is too simple not to, you just do not get what I have already explained in detail by now. Think of previous gens between Athlon CPUs when being shrunk in process might help a bit. A few months away from the final result we are now. Peace. Over.

Die shrinks primarily improve leakage. Performance only improves if the shrink allows for higher clock speeds. IPC is solely dependent on architecture and latency of the various sub-systems with-in a CPU. There is some improvement in performance but only if the circuit is tuned to take advantage of the reduced latency of a smaller circuit which entirely depends on the architecture of the CPU in the first place. Either way, a die shrink only improves power consumption given the same exact circuit at a reduced operating voltage.

No offense but, @cdawall is right, you don't know what you're talking about and you're simply talking over each other. Even within major microarchitectures, there are significant differences between different processes. In fact, one could argue that a die shrink has considerable lower level changes that impact a transistors switching capability and breakdown voltage given the change in circuit size and channel width between both ends of any given transistor in the CPU. A great example of this is Sandy Bridge to Ivy Bridge, it's the same micro-architecture but Ivy Bridge introduced tri-gate transistors which is a CPU-wide change aside from the μ-Arch. It was a required change the influenced the entire CPU merely to move to a smaller node. Don't confuse the two because they're very independent from each other. One determines how the circuit operates (how switches turn on and off and its characteristics,) and other determines how it functions (μArch; what logic does it actually execute.)

 

[RealNeil]

You guys know a lot more about this shit than I do,.....
I hope that it works, and it kicks ass, takes out the trash, and polishes the car!

Hopefully, it's not too long of a wait to see how it does.
Even if it's merely ~good~ I'll probably get it in some form. (we can't let Intel make all of the money)

 

[doxology]

The one thing we know is that this architecture should in theory be completely new in design and different in most ways from what we saw with Bulldozer.

We can't make any sort of comparisons about it with anything from AMD previously as those arguments would be pointless, again because this is an entirely new design that was designed from the ground up. It really sounds like AMD threw out all their previous designs and is starting over.

I honestly don't really expect it to be a giant Intel killer, but I do expect it to get close enough that we can see Intel finally try some new stuff and hopefully we can stop being stuck at an 4 core consumer market CPU design.

Now if it could be clock for clock even or better than Intel that would be awesome. AMD usually offers a cheaper product than Intel and I just want to see some legit competition that would finally bring prices down while also forcing Intel to push the industry forward. If Intel does that, then AMD has to keep up at that point.

Time will tell I guess.

 

[FordGT90Concept]

Yeah, Zen is more like Nehalem and sons than Bulldozer and sons. Architecturally, it is brand new.

 

[RealNeil]

The one thing we know is that this architecture should in theory be completely new in design and different in most ways from what we saw with Bulldozer.

We can't make any sort of comparisons about it with anything from AMD previously as those arguments would be pointless, again because this is an entirely new design that was designed from the ground up. It really sounds like AMD threw out all their previous designs and is starting over.

I honestly don't really expect it to be a giant Intel killer, but I do expect it to get close enough that we can see Intel finally try some new stuff and hopefully we can stop being stuck at an 4 core consumer market CPU design.

Now if it could be clock for clock even or better than Intel that would be awesome. AMD usually offers a cheaper product than Intel and I just want to see some legit competition that would finally bring prices down while also forcing Intel to push the industry forward. If Intel does that, then AMD has to keep up at that point.

Time will tell I guess.

Any real competition is more than welcome as I see it.

 

[cdawall]

No offense but, @cdawall is right, you don't know what you're talking about and you're simply talking over each other.

I meant offense. People being completely wrong and not comprehending why annoys me. Quite honestly the conversation was annoying me. Kinda like trying to convince a wall it was a window.

 

[HD64G]

I meant offense. People being completely wrong and not comprehending why annoys me. Quite honestly the conversation was annoying me. Kinda like trying to convince a wall it was a window.

I have wrote in previous posts that I might be wrong. Your big ego is YOUR problem. Find a solution yourself with that, no one can help you...

 

[Aquinus]

I meant offense. People being completely wrong and not comprehending why annoys me. Quite honestly the conversation was annoying me. Kinda like trying to convince a wall it was a window.

I have wrote in previous posts that I might be wrong. Your big ego is YOUR problem. Find a solution yourself with that, no one can help you...

Now now, ladies. Play nice.

 

[cdawall]

I have wrote in previous posts that I might be wrong. Your big ego is YOUR problem. Find a solution yourself with that, no one can help you...

Yep you said you might be wrong and then said the same thing for 8 posts. All of which was wrong. There was no ego in those posts btw. I started out nicely saying that you were incorrect you posted the same thing multiple more times so I just posted louder.

But here just in case you missed it die shrinks don't changer performance.