Discussion AMD Ryzen MegaThread! FAQ and Resources

Page 64 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar


Could be a diamond chip, exception to the rule. However, I don't think it's impossible for process maturity to offer slightly better overclocks soon.

I know the info is that Ryzen 3 will also be a 2+2 CCX setup like the R5, but is it confirmed? Is there any chance the R3 will be a 4+0 CCX, and that's why it is scheduled for 2H 2017, considering the other chips arrived so close to each other?
 
Those links just take me to the pages....

Also, what do you think about this post?

c4652052229d456192ca900a9fc75776.png

 

If i had to guess i would bet the APU series would be only one CCX cluster as the extra space would be needed plus i'm still guessing we will see parts with even less L3 cache then the 1400.

 


I feel like this is something i'm going to prove once i get my chip

When a program is optimized for throughput vs more latency focused Ryzen gets Broadwell like performance if not even a bit more. However when a program is latency focused we get a lower number more around ivy-haswell.

Not to be mean but these reviewers need to test more chips all we get is Skylake-kabylake with sandy and no haswell chip or they will use a I5 haswell but not the Haswell-e chip and compare both ryzen and that chip at the same frequency with many applications. I'd basically pay money to see a Haswell vs ryzen IPC comparison among many applications and games funny enough that is one reason why i jumped to Ryzen.

"i7-5960X vs R7 1700" would be a fair comparison as both have the same clock speeds 1700 only has a 200mhz higher turbo

https://www.pcgamesn.com/amd/amd-ryzen-7-1700x-review-benchmarks

This would be one site they do have ryzen memory clocked at 2667 however
 


4.2GHz is a super diamond chip, I would say. Current average OC on watter for 1800X is 4067MHz from HWBOT database.

Why would the R5 series be 2+2 and the R3 series 4+0? It doesn't make sense to provide the better configuration for the cheaper worst chip. The likely reason for the delay of the R3 series is that those are chips with more stuff disabled in the die. And it takes more time to extract those chips from the production and pile them in enough quantities.
 


One reason I could think of is they needed to put out the 4-core R5's, so they just took the full 2-CCX chip and disabled half of it. Then, they could keep finishing the 1-CCX. The future APU's will use one CCX, one full cluster is better than two half-clusters, and AMD certainly has some background in making different chips with same names (The Phenom II X4 970 could be a full Deneb chip, or a 6-core Zosma with two cores disabled).
 


I believed you were asking me about my posts on IPC. Notre that directly linking to reviews from computerbase or HFR isn't allowed per forum rules.

You can find my comments on Arstechnica review in this post from mine

http://www.tomshardware.co.uk/forum/id-3327589/amd-ryzen-megathread-faq-resources/page-31.html#19592740

where I mentioned that their Haswell chip got slightly higher scores than the Broadwell chip in several of the benches, when it would be just the contrary because Broadwell has better IPC.

Said that, RyZen is faster that Broadwell in Handbrake. You can check this in the x264 benche in the 3GHz review graphs that I provided before and also in the PcPer review

clock-handbrake.png


I have never used Geekbench and I know some people that claims it is a broken and useless bench. I don't know.
 


There is only one Zen die on fabrication at this time: the Zeppelin die, which has two CCX. Any RyZen model, was it R7, R5, or R3 is derived from the same die; somewhat as all the Piledriver-based 4000/6000/8000 chips use the same four-modules Vishera die. The R3 models are R5 models without SMT and lower clocks. It is almost sure for me they will be 2+2 chips for reason stated above.

AMD will fabricate a different die for the Raven Ridge APUs, and this die has only one CCX, and yes, I expect AMD to launch an Athlon-like line of quad-cores --APUs with iGPU disabled-- but two comments here: (i) the APU are scheduled for 2018 and (ii) the APUs use a different CCX design with half the L3 cache. The improvements of having quad-cores in the same CCX will be lost by the smaller LLC.
 


Could you elaborate on that? What is a different CCX design: changes to the actual workings of the architecture, or only the L3 disabled?
 
You're right.

Thanks man. I found the graphs I was looking for in your posts. The french website that compared the Zen architecture to Broadwell.

 


APUs will likely be 1 CCX with L3 intact (since it is baked into the CCXs), and the remaining R7 die space will be consumed by CUs for iGPU.

I do not have 100% confirmation on this, but sources point to this being most likely. Disabling L3 on APUs this time around would be less ideal because it would not save die space for the CCX like giving up L3 did on construction cores.
 


Could be silicon lottery on that 4.2 OC unless we start seeing more and more of that.

As for R3 HEDT CPUs, I would expect those to run 2+2 as well. They can harvest far more dies for those chips by being less picky about what cores they need to disable. With a maximum of 8 cores on zeppelin dies, they should be able to pull out 2+2 configuration dies no matter how badly the chip comes out after manufacture.

I think the primary delay right now is due to trying to get the most money for the chips coming off the line. Why sell quad R3s when you can sell quad R5s for more money and still be competitive? From a business perspective it makes a lot of sense.
 
I said it before and i'll say it again R5 is a better deal then the I5 and i expect the R3 to repeat the same thing even more so after i saw "simulated performance of a R3 1200"

http://www.techspot.com/review/1386-amd-ryzen-5-1400/page6.html

212 coolermaster is still much better then a wraith spire
https://www.youtube.com/watch?v=VNjkDoYZYjU

But its still not to bad, gonna see for myself i bet i'll like it but want my water cooler real fast ha ha picky about sound.

Now on Amd and i have to say i only played Fallout 4 and i'm running my memory at 2933 just fine and at stock clocks and it does actually feel real nice.
 
As a note on IPC: Ryzen definitely has a stronger SMT implementation then HTT, which will certainly make IPC look better in any multithreaded benchmark (when comparing the same physical/logical setups). For pure single-threaded benchmarks, Intel is still 15-20% ahead.

This is probably an argument in favor of Intel beefing up HTT a bit.
 


Gamer i respect your statements why on earth am i seeing smooth gaming performance compared to my 4790K at 4.7Ghz in gaming?

This is weird i've been testing and re-testing but in Fallout 4 when a city or new location loads for example i feel a small stutter on my Intel build vs my 1700(at stock) at 2933mhz memory. The only guess i can come up with is 1866mhz that i have for haswell is to slow for my 1080.
 


It is all identical to the CCX design used in Zeppelin die, but it has half the L3 cache. The CCX used in the Raven Ridge die has 4MB of L3. The amount of cache was reduced for reasons of power consumption.

You bring a good question. I have always believed it is a new CCX, but it could be the old CCX with half the cache disabled. I don't know.
 


It is not a stronger SMT implementation, but a more distributed microarchitecture that targets a lower single-thread IPC and leaves more execution units ready to execute a second thread. I even draw a diagram to illustrate the idea

SMT%2Bvs%2BILP.png
 


Well remember Single thread is 10 - 15% (sorry 20% is stretching it a bit Gamer- the 20%+ defeicit in *average* game FPS is as much a result of clock speed as it is IPC when comparing against something like a 7700k) is vs Kaby Lake. Which itself is easily 10+% higher in IPC than Haswell. That means you aren't facing much of an IPC deficit moving to Ryzen.

I think the other point though is the fact that games *do use 8 threads* these days (i7's do show a good uptick from i5's now)- whilst you probably don't need *more* than 8 threads for the game, if a full 8 threads are being used any background tasks, windows functions and so on are likely to interfere a bit. That as well as the huge cache pool on Ryzen (and as you say the faster memory) are probably adding up to result in a smoother overall experience.

It's worth noting that whilst Ryzen doesn't tend to win the average fps numbers, it does generally have better minimums.
 


This is what a lot of people are reporting. Not sure how to quantify this qualitatively...
 


AMD's projections are 7% clock per clock behind skylake, kabylake is just faster skylake.
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom