LinkBack URL
About LinkBacksRead more.Canard PC tested an 8-core Ryzen processor with 3.15/3.4GHz base/boost.
Read more.Canard PC tested an 8-core Ryzen processor with 3.15/3.4GHz base/boost.
Some more details - it is an A0 sample,and the motherboard has an old AGESA firmware version,meaning it can't run later ES and Turbo is not fully functional(apparently). The CPU basically did not actually Turbo past 3.3GHZ at maximum.
Regarding the games,4 to 5 of them don't really scale beyond 4 cores(apart from BF4 OFC). 4 of the games(maybe 5) show no gains with SMT.
The Core i7 6900K has a base clockspeed of 3.2GHZ but will tend to maintain Turbo of between 3.5GHZ to 3.7GHZ looking at reviews.
I guess these are sort-of interesting, but given the caveats ....
- ES not final shipping processor
- "conservatively" clocked
- no pricing
.... I'm not sure what practical, real-world value they have.
It also seems the motherboard is an earlier BIOS and firmware version that the one AMD used with a Titan X. The motherboard would not boot with a GTX1080 so they used a Fury X and AMD drivers are more single core limited,which also hints at decent IPC.
They have a use,since we can estimate Ryzen IPC,which in gaming seems relatively close to BW-E level,and between Haswell and BW-E level for non-gaming stuff and these are the worst case scenarios too.Originally Posted by Saracen
The games are easier to estimate as 3 of them don't scale beyond 4 cores,and seem to be more dependent on single core performance,one seems to show 90% of of its performance on the first 4 cores,and uses a 5th core slightly but is also dependent on high single core performance - the AMD FX8350 is beaten by a Core i3 4330 in those 4 games.
None of them scale with SMT,which means AMD improvements in performance are mostly down to core IPC improvements,not a better form of SMT over Intel.
One game is really more GPU limited,once you hit a certain performance level of a CPU and the other scales to around 8 cores.
The other factor is the A12 9800 quad core is only slightly slower than an FX8350. Both have similar core performance,but the A12 9800 can suffer under more multi-threaded situations due to lack of L2 cache. The fact that they are so close together also hints at the gaming suite being more favourable of single core performance than a large amount of cores or threads.
Last edited by CAT-THE-FIFTH; 27-12-2016 at 01:11 PM.
But "estimate", when not knowing how final production chips will compare, or how price-competitive they'll be.
Academically interesting, perhaps, but real-world useful? I certainly wouldn't base commercial decisions on it.
Look at my updated previous post - IPC performance is generally almost BW-E level and AMD CEO Lisa Su is on record saying the retail CPUs will be having a 3.4GHZ base clockspeed.
At that level of clockspeed,it is incredibly close to a £1000 Core i7 6900K over both gaming and non-gaming benchmarks. Like I said in 4 of those games,a Haswell dual core beats an eight core FX8350.
The sample tested had a 3.1GHZ base clockspeed - the AMD retail samples will starting at nearly a 10% higher clockspeed.
The Intel CEO is on record as saying Ryzen will be disruptive during 2017,and from that article I read it is strongly hinting Intel is betting on 10NM to fight Ryzen.
Edit!!
Also the editor behind CanardPC who leaked these results,leaked the first 1.4GHZ Athlon 64 results months before launch.
We can kind of see where that one went!
What is the bigger unknown is whether AMD is launching 4C/8T and 6C/12T models and how high they will be clocked at,or whether AMD is skipping the 4C/8T models and moving to the Zen APU.
Every leak so far has only shown 8C/16T IIRC and no lower core count versions. AMD still needs a counter to the 4C,4C/8T and 6C/12T Intel CPUs which are probably their biggest selling desktop performance CPUs.
Last edited by CAT-THE-FIFTH; 27-12-2016 at 01:15 PM.
Depends on pricing when you think about it... if AMD price their 8c/16t at a similar price point to 4c/8t i7's (it's possible based on past comments by AMD)then I'd say that might make a difference lol.
As to the benchmarks.... price will be a big factor I'd say now, those performance figures look promising for my usage (3D rendering etc) so if the price is right...
Personally I think this is what is going to happen,if AMD gets around BW-E level IPC and decent enough clockspeeds,and is partially based on some other hints I have see around the interwebs:
1.)Top bin Ryzen 8C/16T being around £500 to £600,with an AIO water cooler against the £1000 8C/16T Core i7 6900K which lacks a cooler.
2.)Lower bin Ryzen 8C/16T being around £400,with the Wraith cooler against the similarly priced 6C/12T Core i7 6800K which lacks a cooler.
3.)6C/12T Ryzen being around £300 to £350 with higher clockspeeds than the lower bin Ryzen 8C/16T SKU,with the Wraith cooler against the similarly priced 4C/8T Core i7 6700K/Core i7 7700K which lacks a cooler.
4.)4C/8T Ryzen being around £200 to £250,with the Wraith cooler against the similarly priced 4C Core i5 6600K/Core i5 7600K which lacks a cooler.
AMD will also "sell" the fact that socket 2011 motherboards are £180+ and going from previous AMD pricing decent motherboards are £80+ so you also save on the motherboard side. OFC,the pricing is based on the current value of the pound which might be different next year(could be less,or it could be more).
Some more information from CPC:
https://twitter.com/InstLatX64/statu...39775944785921
SMT is bugged in the sample and so is part of the cache.
It makes me wonder how it will fare with these working properly and with Turbo working fine.
I can't remember where I saw it but it was claimed the 8C would launch initially, with the 4C and 6C versions following a few months later. That's not unlike what Intel does with their i5/7 first with i3 coming later.
Based on those very broad aggregated benchmarks alone, the Ryzen CPU seems to more or less match Intel's IPC. It's possible they won't have as much clocking headroom as Intel's 14nm process but I'm impressed even the ES chips and early motherboards/BIOS are clocking and performing as well as they are. They're only 300MHz off the 6900k and are drawing less power so I'd be surprised if there wasn't at least that much headroom for shipping CPUs.
The aggregate benchmarks seriously obfuscate the performance attributes of Ryzen (perhaps it's intentional so as to not give too much away?) Something else I seem to recall reading is that first generation Zen has concentrated less on things like AVX with a view to improving that with the next iteration, which makes a lot of sense given the budget/manpower/time constraints and the fact it's still not that widely used outside of niche applications - it makes sense to build a solidly-performing core and go from there as the need arises.
CAT-THE-FIFTH, you have a broad knowledge in computers, thanks for the info
I know where you're coming from, Cat, and I don't disagree with what you say. But .... there's a lot of assumption and supposition in there, as to how this pans out, how production chips relate to these ES chips, and what pricing, market positioning and strategy will be.
It may very well be that you're right in all that. It wouldn't surprise me. But it's still guesswork.
Bear in mind, I've been on the receiving end of ES chips, preproduction models and code for, well, it's in the decades. I've even been presented with ES chips by AMD, along with an example of every production speed of a given generation, to do exactly these kinds of tests. I've had the presentations, white papers, meetings with everybody from product managers to CEOs. In fact, my current "main" machine is still running a now rather elderly Intel ES chip.
My point is that while the speculation is interesting, it won't have real world value until we find out if it's correct or not and, as LSG501 pointed out, pricing makes a big difference.
Well its not entirely guesswork as the 8C/16T will be shipping at 3.4GHZ as a minimum(according to the AMD CEO) and we have a famous overclocker and Caseking employee mention the following:
https://www.reddit.com/r/AMD_Stock/c..._after_he_was/
He also mentioned if you had a Core i7 6900K already you would not find much gains too. Then we had 8Pack who is another well known overclocker and OcUK employee hint the top model will be bundled with a AIO water cooler.
So we can elucidate its similar speed to what Intel has in BW-E and it will have better price/performance,and certainly the gaming benchmarks hint at a massive jump in single core IPC over Piledriver,which is where AMD has struggled for quite a while.
One of the big questions at the AMD preview,was whether AMD single core IPC improvements,or AMD having better SMT was the reason why it was matching a Core i7 6900K in their demos. Increased SMT throughput would mean lower single core performance. However,looking at the games run,it pretty much says single core performance is the main reason for the improved scores.
There is also a Ryzen competition being run by AMD in China which ends on the 18th of January,which hints we might be seeing at least the top SKU launch somewhere in January or February.
Even with the pricing,going from prior history,even if AMD was being agressive with pricing I expect Ryzen will not be cheap for the top SKUs,especially if an ES is so close to a Core i7 6900K. Even if the ES was final performance,the Core i7 6900K is £1000,so even £500 would be considered "cheap" relative to that mental pricing.
Last edited by CAT-THE-FIFTH; 28-12-2016 at 03:59 PM.
Someone found on AT forums an Easter Egg in the CanardPC article and I confirmed it looking at a scan of the article(its on page 10).
The following code is seen in one page:
010110100110010101101110010011110100001101000000010000010110100101110010001111010011010101000111
Convert it on this page:
http://www.roubaixinteractive.com/Pl...ry_To_Text.asp
It gives you:
ZenOC@Air=5G
Wow that's...different?
Seriously though that's too long a string to be a coincidence so that's pretty interesting if true! 5GHz is a heck of a clock speed for an ES like this but I wonder if that's with all cores enabled or not?
Either way, it's a reassuring hint that shipping clock speeds likely have a lot of headroom.
This was posted last month on Hexus:
http://hexus.net/tech/news/cpu/99400...se-mobo-maker/
Last edited by CAT-THE-FIFTH; 29-12-2016 at 01:29 AM.
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
Reply With Quote
