Discussion: AMD Ryzen

Page 9 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.


I post in other forums and I can sure you that many posters (including certain Italiian tech 'writer') claimed that the power issue of the RX480 was an invention from Toms. After other sites reproduced the issue, they continued claiming that the power issue was an invention. After AMD acknowledged the issue and announced a fix, one half of the posters continue claiming that the issue doesn't exist and that AMD is releasing the fix only because users believed the lies of the media. Another half now admits that the issue is real but claims that the media is biased because Nvidia cards also violate PCIe specs, but this is hidden as part of a worldwide conspiracy. They have been mentioning a number of cards that supposedly violated the spec and that Toms didn't notice. The funny part is that they use power figures made _public_ by Toms to pretend that the Nvidia issue is being maintained in secret. Of course, all the Nvidia cards they mention run within the spec. Pathetic!

Same will happen with Zen. Any site that makes a honest review and notices the flaws/issues will be attacked by that people.
 
After reading the discussions and the sources, some things left me feel wanting for more questions. The starting prices of the new processors, and the motherboards supporting the socket once they enter the consumer market.

The newest Intel motherboards support the PCIe gen3, which I would need the bandwidth since I don't have a solid read of what stresses a GPU (gaming or workstation) will subject to when I start the 3D modeling and texturing as part of the game development process. The major problem is that these motherboards belong to processor sockets whose asking price is already my entire projected budget.

If I am going after a processor that costs $1000-1500+ to meet the purposes, I'm doing something wrong and financially irresponsible. If I am settling for the AMD FM2+ processors or PCIe gen 2 graphics cards, as far as my research and inquiries can conclude, I would asking for trouble to arrive late of the development process and result in risky scrapping and revisions. The stuff I want to do in my spare time once I learn the related skills in my own time is game development, but making 3D models and composing music and audio on my own and at optimally $0 software budget are major hurdles in my PC build planning (taking advantage of one-time free OS opportunities, so I won't bleed more funds getting Windows, as Linux distribution diversity is not helping with my thoughts about game testing and game engine decision-making).

Not going to hold my breath of what the Zen architecture can do, since my greater concerns in life is being a CIS college student keeping financial status in the mental background. Though starting prices of Intel meant that I can only have Intel builds that meet the needs when I have much larger resource pool (produces more wasted time). I am patient, but timing is a priority when I want to take up development as a interest-based hobby and use any revenue from that to become self-sustaining and no longer need investment at the expense of my future job earnings that are better allocated towards financial aid repayments.
 
All we have to go on is speculation based on AMD's past history at this point. The only information we really have at the moment is that it should have IPC around that of Haswell, there will be a 8 core 16 thread, (lets ignore the fact that AMD usually has insane overclocking ability), so that leaves us with a chip matching the performance of a i7-5960X (theoretically).

Now let's look at Intel. Haswell was a few generations back but in reality every new generation of chips has only had an increase of 5-10% and with no competition the prices have gone up as well. Do I believe that Intel can create a more powerful chip? Yes I do, but they have left themselves open. Even if Zen has the performance of a few generations back, it is still in performance range of the newer chips thanks to Intel.

The real question here is price. They already said the first chips out will be for enthusiasts. Dose that mean the 8 core Zen will match the price of the i7-5960X at $1000?! That's how much Intel thinks enthusiasts should pay! Is AMD going to under-cut them and go for $800? Still a hard pill to swallow but still a good deal. Or do you think AMD will go right for the throat and price 8 core 16 thread Zen against the 4 core 8 thread i7? It lacks the GPU but are you really going to use an i7 GPU anyway?

I think AMD really shines in the APU department. Every where I turn people are talking about Zen APU with the power of a RX-460. Sounds good, but I really hope for an APU with the power of an i5 and the graphics of an RX-480 even if it is a $300-$400 chip. It probably wouldn't make much sense business wise but it would allow the small-form-factor builds to really take off and this isn't a place Intel can really compete given their iGPU sucks and a gaming PC can only be so small when you are forced to use a dGPU.
 


According to data given by AMD Zen has IPC around Sandy Bridge. That "insane overclocking ability" was the result of Bulldozer family being a speed-demon microarchitecture (i.e. designed for higher clocks) plus using a special SOI process node derived from IBM foundry tech. This time, Zen is a brainiac microarchitecture (i.e. designed for higher IPC) and it will be made on a mobile oriented process optimized for low clocks: 14LPP. Overclocking abilities of Zen will be very bad.



Every new generation from Intel only has 5--10% gain when using legacy x86 software. When using the new AVX256 instructions Haswell is up to 70% faster clock-for-clock than Ivy Bridge. The x86 ISA is not scalable and Intel did hit a performance wall, which had been predicted in academia decades ago.

Prices are not going up because of lack of competition, but because the PC market is in free fall and because classical rules of silicon scaling stopped to work. Newest nodes are very expensive to develop and get working because we are hitting the physical limits of silicon. This is the reason why both AMD and Nvidia did skip the 20nm node.

Pricing is a complete mystery. I guess that Zen 8-core will be priced similar to Broadwell 6-core, because Zen is a simpler smaller product made on a cheaper node.



Intel has the fastest APUs now and I don't think this is going to change soon. The RX460 has 14CU and GDDR5. Zen APUs are rumored to have DDR4 and up to 11CU.
 


Juan, I'd agree with most of that although I think you are underestimating the IPC somewhat. According to AMD Zen is +40% IPC over Excavator (which itself is quite significantly faster per clock than Bulldozer). That should put it near to Haswell as others have said- I don't recall any recent information debunking that prediction. The main (potential) issue is likely to be clock speeds.
 


I am also using the 40% over Excavator claimed by AMD. On page 1 of this thread I wrote:

 


You are forgetting L3 cache to start with...
 


There is no L3 on Excavator.
 


I think that is his point- although I guess 40% IPC would be over actual silicone rather than a hypothetical faster excavator with l3...

What I would say though, that is a single benchmark- and from what I remember Intel's newer designs show little improvement at the same speed in that test. I think we'd need to take a selection of different single thread performance metrics and compare to get a better picture- in some areas Excavator is much closer to Intel than others... I'm unsure what type of processing Cinebench represents?
 


That is my point. The 40% is measured against a current implementation of Excavator. Zen (with L3) is 40% faster than Excavator (without L3).

Of course a single benchmark doesn't tell all the history. In some workloads Zen will be on pair with Skylake on another workloads Zen will be 70% behind Haswell. It depends. The reason why I am using CB is because AMD has been using CB to measure IPC and performance for a while. AMD gave CB scores when presented the IPC gains of Excavator over Steamroller about one year ago.

Slide%2012%20-%2015W%20Optimized.png


Recall also the last slide from AMD with CB scores of Kaveri, Carrizo, Bristol, Orochi, and Zen

AMD-Zen-Performance-Double-FX-83501.jpg


The reason why AMD uses CB is because this is a good benchmark that characterizes overall performance.
 


So that sounds like we'll see Zen this year, though either as a paper launch, or with very limited (and likely over priced) stock for die hard enthusiasts... Still, on a plus side means we should get the final specifications and benchmarks sooner rather than later if nothing else :) That will put an end to all the speculation, and instead go back to the 'AMD is doomed', 'no AMD will DOMINATE Intel with Zen+' discussions 😛
 


😀 Precisely I believe that the launch of Zen will be the starting point for unending discussion in forums about 'AMD is doomed' from one side and 'AMD will DOMINATE Intel with Zen+' from the other side.
 


I'm just hoping the reality is somewhere in-between- good enough to put AMD back on a steady footing. I think we've pretty well established Zen is going to be a big improvement but isn't going to be dethroning Intel.

What I think it *might* do is nudge 6 and 8 core parts down into the 'mainstream' with entry kit being quad, rather than the dual core (albeit with HT) parts that are still dragging along at the moment.
 
I'm mostly curious about what AMD can do in the generation after Zen. Remember the gains Intel saw with Sandy Bridge over Nehalem? It wasn't quite the same level of architectural redesign, but it was still substantial. Those performance gains immediately plateaued, Ivy Bridge felt like a disappointment, and they've focused more on power efficiency and features and IGP ever since.

Is the Zen architecture just getting started, or is it headed for the same incremental improvements?
 


Those specs are from a forum leak and previously discussed here

http://www.tomshardware.co.uk/forum/id-2699237/amd-cpus-soc-rumors-speculations-temp-thread/page-32.html

A resume is as follow. That leak is abut A0 stepping silicon. AMD could use A0 stepping as base for final product or wait to launch A1 stepping. If it is A0 then Zen will have the clocks mentioned of 2.8GHz base and 3.2GHz turbo. If AMD releases A1 silicon then clocks could improve about 200--400MHz.

A granularity of four-cores is perplexing. Most people including me was expecting 8-core and 6-core. The only good part seems to be the low idle consumption.
 


Zen+ will bring 5--10% gains over Zen, and Zen++ will plateau.
AMD-40-IPC-Zen-Zen-.jpg
 


But those are the same people that expected 12-cores, 4.1GHz base, faster than skylake, AVX512 support,... for Zen.

Therefore, I will maintain my claim that Zen+ will bring 5--10% IPC gains over Zen and will ignore what "they are expecting" for the next uarch.
 
I've been following Zen for quite some time, and although I don't expect it to totally dominate Intel (even with Zen+) I do expect Zen to bring AMD back into the high end CPU market. Like most others have speculated I fully expect Zen's IPC to be at the performance levels of Haswell, not Sandy Bridge. Heck my FX 8370 gets single core Cinebench R15 scores very close to Sandy Bridge, overclocked past 5Ghz. 40% IPC gains over Excavator will be roughly equal to Haswell performance.

With IPC roughly equal to Haswell but running 8 cores and 16 threads Zen should be able to make a nice "splash" into the CPU market. It should be able to give 4 core 8 thread Skylake processors a run for their money, but will of course be outperformed by Intel's top tier offerings. It will be interesting to see where the stock clock setting will be for Zen FX as well as where AMD will price Zen. Considering the uphill battle AMD will have in getting back into the high end CPU market it would be nice to see Zen priced between Intel i5 and 4 core 8 thread i7 Skylake processors. At that price point Zen would be highly affordable and desirable.
 
Is the IPC claimed for per core? Or it is total die? I think it is the later. So, 6 core Zen has 40% faster ipc than 6 core excavator.

They can also increase ipc just by giving each core a dedicated FPU.
 
Status
Not open for further replies.