• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

EliteRetard

Diamond Member
Mar 6, 2006
6,490
1,019
136
"How did amd outdo intel in the power efficiency game? The 1700 @ 3ghz has a 65w tdp while a 3.2ghz 6900k is 140w... mind boggling i tell ya.
Currently the 6900k is $1050 on newegg with a $50 discount, lets see how fast this will drop."


Since quoting is broken for whatever reason for me....

this is what interests me personally as someone who owns X99 platform only for about half year.... if all these leaks about performance and prices are true and AMD is truly gonna sell their 8-cores for as little as 1/3 of the price of Intel´s ones, while performing pretty much the same, do you think we are in for massive price-drops of entire Intel´s HEDT line-up? Current BW-E i mean... No way i am going to sell my entire rig only to replace it with Ryzen, too much hassle with that for me, but i could think about at least upgrading my CPU (6850k) for something faster, if the price becomes right as the result of Ryzen´s competition. Ideally 6950x... what do you think are the chances its gonna drop to say 1000?
You'll be lucky if they drop the 6950x to $1,500. Remember Intel owns 90% of the market, and AMD taking away 1% of that isn't really a big deal. Over time if AMD can remain competitive and significantly cheaper then Intel will adjust their pricing down....I suspect eventually settling back into the $1k max CPUs. And for the foreseeable future Intel will be able to charge a noticeable premium on inferior (performance or otherwise) products, same as Nvidia.
 

EliteRetard

Diamond Member
Mar 6, 2006
6,490
1,019
136
And I want to suggest something for those confused about the TDP numbers.

Product segmentation.

Just like Intel has 5 to 90 watt i7's, or a 4c/8t 3.8GHz at 45w vs a 2c/4t 3.9GHz at 51 watt (i7-7700HQ vs i3-7100).
If AMD is employing XFR kinda like Intel's turbo boost, then the TDP of the chip makes a big difference in it's maximum stable clocks.
It makes sense to offer a variety of core configurations and TDPs, even if that means a 65w 4 core and a 65 watt 8 core (and a 95w 6 core).
At the same time you don't want a bazillion options for each and every possible combination, that's confusing and costly.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
My understanding is that this is similar to how Sandy Bridge was designed, as well:
Sandy Bridge allows 256-bit AVX instructions to borrow 128-bits of the integer SIMD datapath. This minimizes the impact of AVX on the execution die area while enabling twice the FP throughput, you get two 256-bit AVX operations per clock (+ one 256-bit AVX load).

Zen has two FPU ADD units and two FPU MUL units, each 128 bits wide. Presumably AVX will be handled the same way as on Sandy Bridge, and then should provide the same throughput. Haswell and up have 2x the theoretical AVX throughput of Sandy Bridge, but that doesn't mean double speed in real applications, not even x264.
SB and IVB did it this way. I think, this changed with HSW, when they added FMA, AVX2 for integer ops, and other stuff needing a redesign of the execution backend.

Zen just needs to handle 256b ops in an improved way as has been done in the BD family. There still are instructions which can go to only one execution unit. 256b ops would use it for 2 cycles then, like 128b on K8 with 64b units. There is no need to do some "flex FPU" 2x128b combining. So the seamingly easiest way was to split 256b front end uops in the dispatcher into 2x128b FPU uops and let their execution being sorted out by the FPU's scheduler.

On a side note: David Kanter mentioned in the RWT forum, that Zen's AGUs might be just needed once for a 256b mem access. Especially in the case of aligned data this would just mean setting a single bit in an already calculated address.

256b theoretical peak FLOPS can already get lost with a little more complex code due to necessary data shuffling. And in typical apps they might contribute IPC peaks (actually not true IPC, as 256b needs half the instructions ;)), while in any other cycle there is no benefit. So it averages down to something.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
It's just scary the SMT yield AMD achieved on their second try globally, and their first try for all pipelines. I think it's really a benefit of having INT and FP separated like AMD does from several generations ago. The Port scheme of Intel IMO seems more or less a nightmare to have a good yielding SMT going on.
They probably could hire more people which already did SMT in other uarchs to bring in more experience - if they didn't already for BD with SMT and vertical MT inside. As I once heard they abandoned SMT for K7 due to complexity. But now the simulators, FPGAs, other tools have improved enough that even a team that size and with those resources could implement it. And K12's ARM cores also were about to get SMT as it seems.

On Intel's defense: SMT is not only about saturating the execution units (hence they added more over time), but also for doing useful stuff while sorting out some stalls in one of the threads - mostly branch or memory access related. So it could have been their philosophy to improve that sorting out by adding mem related ports and one more branch unit.
 
  • Like
Reactions: Drazick

inf64

Diamond Member
Mar 11, 2011
3,145
2,162
136
I have been looking for some tests in Discrete gaming:
http://www.anandtech.com/bench/CPU/1358

I hope this test is interesting. $560 on the GPU is maybe to much today.
Best budget CPU looks to be i7-7700 for $303.
Where would you put the Ryzen alternatives?
I think better way to compare gaming performance is to use ie. hardware.fr's numbers:


I think that any 8T and above Ryzen that has 3.5Ghz base and 3.8+Ghz Turbo will be scoring above 120 pts in the chart above. I guesstimate that top SKU (1800X) should be getting between 130 and 135pts in the chart above, top 12T Ryzen SKU between 125 and 130 and top 8T Ryzen SKU between 120 and 125pts.


Yep that was expected. Goal of 40% and 52% on average achieved. R15 seems to be on the higher end of the IPC gain range as Zen is having 59% higher ST IPC than XV core at the same clock. In MT portion of the R15 Zen has 2.5x higher throughput per core than XV at the same clock(SMT effect + no more CMT penalty nets a total additional gain of ~57.6% or 1.576x over XV; CMT penalty is 16% in XV core while running 2T on a module in R15 benchmark).
 
Last edited:

OrangeKhrush

Senior member
Feb 11, 2017
220
343
96
I have to wait until middle of march to get my new toy.

My GF's son plays on the HTPC, it is showing its age now, this is a big upgrade from a i5 4460
 

Greyguy1948

Member
Nov 29, 2008
151
15
91

inf64

Diamond Member
Mar 11, 2011
3,145
2,162
136
So Videocardz made that slide up? that's a bit silly.
I doubt they made it up. The font and color scheme seems identical to what we saw in previous leak (CB 1800X and 1700X results). It likely came from the same presentation.
 

lolfail9001

Golden Member
Sep 9, 2016
1,056
353
96
I doubt they made it up. The font and color scheme seems identical to what we saw in previous leak (CB 1800X and 1700X results). It likely came from the same presentation.
They sort of wrote it right there: it is basically a reproduction of actual slide.
 

blublub

Member
Jul 19, 2016
135
61
101
"We are just getting started" - haha, hype train just accelerated to unstoppable mode until ZEN+ arrives.
Hell are they planning another 40℅ increase over Ryzen???
Ahh that would make Intel obsolete...lol
 

ASK THE COMMUNITY