Discussion: AMD Ryzen

[Kulasko]

http://

Less die space used gives AMD flexibility to throw more on a die by update.

That is not die space. Moreover the Zen design looks terribly unoptimized. 44mm^2 vs 49mm^2 is a 11% difference on favor of AMD, but one must recall that Zen is a 2x128bit core, whereas Intel is 2x256bit. I.e. Intel has 11% higher size but about 2x more FLOPS.

There is much more to a CPU core design than just the maximum cache bandwith or FPU throughput. Yes, Zen has half the maximum performance in AVX workloads, but otherwise it looks like a solid competitor to the current gen of Intel CPUs. I would be suprised to see other common workloads with Kaby Lake besting Zen 100% or more.
Maybe Zen is slower, maybe even less efficient but we can hardly call a quad core with double the L2 and 11% smaller size on an 10-25% bigger node a tebbibly unoptimised design.

 

[sarinaide]

Zens issues will be process maturity which is a massive step forward given that it will not be the performance of the CPU. Hopefully the new bios updates and there is many stabilize clock speeds and allow boost to be activated. Apparently boost hasn't really worked because AMD hadn't focused on it while tweaking performance related stability.

It may take the first batch to iron out the process and once glofo get experience with the node, then improve silicon quality. That said 3.5-4ghz window would be sufficiently good to warrant competitive performance.

 

[jdwii]

"Maybe Zen is slower, maybe even less efficient but we can hardly call a quad core with double the L2 and 11% smaller size on an 10-25% bigger node a tebbibly unoptimised design."

Well also remember that it only has 4ALU with 2AGU(Still think that is a odd design, even a slightly concerned design). I'll never forget when Amd engineer's said the phenom II could never use the third ALU unit having 4ALU really makes me wonder how Ryzen's SMT scaling is.

Just can't wait to see benchmarks i'd be happy if it had haswell or better IPC

 

[8350rocks]

"Maybe Zen is slower, maybe even less efficient but we can hardly call a quad core with double the L2 and 11% smaller size on an 10-25% bigger node a tebbibly unoptimised design."

Well also remember that it only has 4ALU with 2AGU(Still think that is a odd design, even a slightly concerned design). I'll never forget when Amd engineer's said the phenom II could never use the third ALU unit having 4ALU really makes me wonder how Ryzen's SMT scaling is.

Just can't wait to see benchmarks i'd be happy if it had haswell or better IPC

David Kanter is expecting slightly better than haswell performance...I forget where I read that article, but it was quite interesting. He speculated in heavily threaded workloads it might rival skylake regularly, but workloads bound by a single thread will likely be slightly above haswell performance.

We shall see...

 

[sarinaide]

"Maybe Zen is slower, maybe even less efficient but we can hardly call a quad core with double the L2 and 11% smaller size on an 10-25% bigger node a tebbibly unoptimised design."

Well also remember that it only has 4ALU with 2AGU(Still think that is a odd design, even a slightly concerned design). I'll never forget when Amd engineer's said the phenom II could never use the third ALU unit having 4ALU really makes me wonder how Ryzen's SMT scaling is.

Just can't wait to see benchmarks i'd be happy if it had haswell or better IPC

David Kanter is expecting slightly better than haswell performance...I forget where I read that article, but it was quite interesting. He speculated in heavily threaded workloads it might rival skylake regularly, but workloads bound by a single thread will likely be slightly above haswell performance.

We shall see...

From more information it seems like it beats a 5960x on maximum throughput at low clocks. AMD's Zen+ update will likely feature major IMC updates as apparently the IMC is not quite as far along as initial sentiment, high but should be better is the word out.

I am also told the clockspeed issues will improve as glofo process matures and new stepping comes out.

Zen+ will likely focus more on IMC and clockspeed but i was assured that performance at low clocks is very good, need a very high overclocked 2600k to keep up.

 

[juanrga]

David Kanter is expecting slightly better than haswell performance...I forget where I read that article, but it was quite interesting. He speculated in heavily threaded workloads it might rival skylake regularly, but workloads bound by a single thread will likely be slightly above haswell performance.

Are you sure he said that?

Overall, our estimate for Zen is fairly close to the score for Intel’s Ivy Bridge but short of Haswell’s and Skylake’s. AMD argues that the Intel products receive an unfair 5–10% boost because the company compiles to 32-bit x86 code, which is unrealistic for many applications. Adjusting by 10% would put Zen about halfway between Ivy Bridge and Haswell for SPECint_rate2006. Either way, Zen delivers a tremendous improvement in per-core performance.

Between Ivy and Haswell, and his estimation is for a nonexistent 3.7GHz 4C/8T Zen.

 

[juanrga]

Last rumor about pricing

1800X: 8/16 @ 4GHz = $599
1700X: 8/16 @ 3.8GHz = $469
1700: 8/16 @ 3.7GHz = $389

Reported clocks are turbo

 

[cdrkf]

Last rumor about pricing

1800X: 8/16 @ 4GHz = $599
1700X: 8/16 @ 3.8GHz = $469
1700: 8/16 @ 3.7GHz = $389

Reported clocks are turbo

$389 for the 1700 sounds very appealing if the performance is where it appears to be (especially if all parts are unlocked as hinted). That is *a lot* of processor for that money. Still, will take it with a grain or 10 of salt until we get figures from AMD themselves (and even then, I expect retailers to mark up the prices if there is high demand- AMD can't force retailers to stick to it's pricing after all).

 

[sarinaide]

August 2016 before anything was even known, since then a LOT has changed, so much so that my mock trail on CB was below the real number.

• 
  AMD at low frequency has the performance (in respect of single threaded), performance is not a concern for the architecture.
  
  Silicon binning will improve per stepping with progressively better clocks and stability.
  
  Zen+ to focus mostly on improve IMC.

 

[juanrga]

Last rumor about pricing

1800X: 8/16 @ 4GHz = $599
1700X: 8/16 @ 3.8GHz = $469
1700: 8/16 @ 3.7GHz = $389

Reported clocks are turbo

$389 for the 1700 sounds very appealing if the performance is where it appears to be (especially if all parts are unlocked as hinted). That is *a lot* of processor for that money. Still, will take it with a grain or 10 of salt until we get figures from AMD themselves (and even then, I expect retailers to mark up the prices if there is high demand- AMD can't force retailers to stick to it's pricing after all).

Pricing seem to be confirmed, but I think the 1700 is a 6C/12T instead.

 

[truegenius]

54% price increase for just 8% turbo clock increase. 😛
And people think intel is greedy with their partially unlocked i5/7 vs k series fully unlocked i5/7 cpu for 10-20$ premium 😗
If these prices are right then either amd is being greedy here or is having serious issues achieving high clock without skyrocketing tdp. Just hope amd is being greedy else latter will be a little disastor ( if zen's ipc is around ivy bridge ).

 

[Embra]

I pretty sure the 1700 is a 8c/16t part. It should be targeted more for gamers.

https://elchapuzasinformatico.com/2017/02/precio-amd-ryzen-r7-1800x-r7-1700x-r7-1700/

Translation:

AMD Ryzen R7 1800X
We started with the AMD Ryzen R7 1800x (or Ryzen 7 as the name it ). Processor 8 cores and 16 threads we know comes at a frequency of 4.00 GHz . No further data was revealed, so we will not name any already filtered specs, as we would be joining information already confirmed with rumors. Thus, the CPU, according to AMD testing, exceeds the Intel Core i7-6900K in optimized applications. While Intel CPU worth about 1,200 euros, the cost of AMD as the rumors pointed, almost half, 599.99 euros .

AMD Ryzen R7 1700X
With the AMD Ryzen R7 1700X we face the second processor 8 cores and 16 threads at a frequency of 3.80 GHz , so we would essentially to the same processor with a lower frequency Base and Turbo (the latter is not mentioned). Both the R7 and R7 1700X 1800x, are focused more professional use , so its price will also be high, although "economic" regarding Intel processors for that segment, this time we talk about 469.99 euros .

AMD Ryzen R7 1700
The most interesting of all model, the AMD R7 Ryzen 1700 , which is intended to be the top-end CPU company thinking of gamers . He uses 8 cores and 16 threads at a frequency of 3.70 GHz , the most notable being that it is the only model that indicates its TDP, which is set as 65W , when it was expected to be 95W, so perhaps the rumored 95W are for the two models above. To which we, price, 389.95 euros , and here we do not know whether it is cheap or expensive, until we know their performance over a Core i7-7700K is impossible to categorize.

If surrender as high - end models of Intel would be expensive, but if Ryzen R5 (6 - core) are the direct rival of these Core i7, things change, you'll have to wait.

Models of 6 - core and 4 - core , not a word, do not know if it's because they have not yet reached the first distribution channels , or they arrive somewhat later than 8 - core models, perhaps the first. At least we have already managed to uncover an unknown, the price, will now play wait for performance.



Lee más en https://elchapuzasinformatico.com/2017/02/precio-amd-ryzen-r7-1800x-r7-1700x-r7-1700/

 

[TechyInAZ]

FYI: This page is going to be closing pretty soon. The new Ryzen megathread is in the works and will be published very soon.

 

[juanrga]

54% price increase for just 8% turbo clock increase. 😛

Plus two extra cores.

 

[illegaloperation]

What's going on with Raven Ridge?

Last I heard Raven Ridge is suppose to be released in (H2 2017), after Summit Ridge (Q1 2017) and Naples (Q2 2017).

Why is Raven Ridge low priority for AMD?

Since Summit Ridge is desktop only, we probably have to wait until Raven Ridge for laptop processors.

 

[Nope 1151]

First post, please tell me if I broke any rules.

Take a look the link from a reddit post below:
http://www.shopblt.com/search/order_id=%21ORDERID%21&s_max=25&t_all=1&s_all=AMD+AM4&search=Search

It seems that the Chinese leak was at least semi-accurate. I have also read the Chinese tech site before it was taken down- I can draw a couple conclusions on it.

1)It used simplified Chinese, therefore it is highly likely to be in Mainland China.
2) Written in a very formal Mandarin context and slang, probably means someone living (or grew up in) major cities.
3) The news "site" that broke the story had that leak as it's only post (That I can find)
4)Most workers in factories have phones but not PCs, that means
a) They published it in an internet cafe (Unlikely)
b) They wrote it on the phone (Unlikely)
c) Someone with access to a computer at home wrote it (Manager, or someone with a higher position)

 

[juanrga]

Pricing seem to be confirmed, but I think the 1700 is a 6C/12T instead.

Correction: the 1700 has 8 cores.

 

[renz496]

i haven't followed CP since sandy generation. so which AMD upcoming CPU going to compete head to head with 7600K at stock?

 

[sarinaide]

i haven't followed CP since sandy generation. so which AMD upcoming CPU going to compete head to head with 7600K at stock?

Likely the 4/8 R5 1400/1300 series parts if they can push their clocks to 3.5-3.8/4 then they will be very relevant against the 7600K though unlikely beat them outside price/perf the margins will be slight enough though.

SR7 - directly competitive with the 6900K and 5960X with the lower model more competitive with the 6850K

SR5 (1600 family) - directly compete with the 5820K, 5930K, 6800.

SR3 (1400/1300) - i7 4770,4790(non K), 6600, 6500, 7500, 7400 all Haswell i5 (gen2 haswell) and Broadwell 5000 series.

SR3(1100 series) - i3 and Pentiums

Performance

Haswell < Ryzen > Broadwell

IMC = Ivy bridge, was rather underwhelming but given where AMD's IMC was this is a massive leap, though will hold back the throughput a little. Zen+ features major updates to IMC.

 

[renz496]

so assuming the game is very cpu bound can those R5 1400/1300 compete directly with 7600K performance? personally i'm fine with more cores but i don't want single threaded performance being held back just so they can "enable" more cores on the CPU. we often see those intel 4 core always end being faster in games because they have much high base clock speed than those 6 core CPU. also i was gunning at least 4.5Ghz OC.

 

[sarinaide]

so assuming the game is very cpu bound can those R5 1400/1300 compete directly with 7600K performance? personally i'm fine with more cores but i don't want single threaded performance being held back just so they can "enable" more cores on the CPU. we often see those intel 4 core always end being faster in games because they have much high base clock speed than those 6 core CPU. also i was gunning at least 4.5Ghz OC.

Given that Kaby Lake will have a IPC advantage, for Ryzen to hold its ground against a 7600K it will need to have base+turbo around 3.6-4Ghz minimum and 3.8-4.2 to realistically be close as clockspeed is the limiting factor. Ryzen's market is more against the Haswell/Broadwell families, a top end 4/8 Ryzen can compete with a 4770/4790, I dont' really think that they market them against SKL/KBL but is a 4790 not a good CPU?

 

[TechyInAZ]

Thank you all for participating in this thread! But unfortunately (or fortunately), this thread will have to come to a close as our Official AMD Ryzen MegaThread is now up and running.

Warning: There has been quite a bit of moderation in this thread over the months from a few of you not listening to the rules applied here at TH. I want to warn you that the official Ryzen MegaThread is going to be moderated even more strictly. Remember at all times to NOT get involved or create a flame war and do not turn your replies into personal attacks.