Hear Hear. Roll on CES.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
*** AMD "Zen" thread (inc AM4/APU discussion) ***
- Thread starter Boomstick777
- Start date
Hear Hear. Roll on CES.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
why no battlefield 1 benchmarks ? the most optimized cpu benchmark that utilizes cpu cores currently ?you want to show cpu power for a large sector of the market but choose to not do that benchmark
so while i think it will be a decent product why not simply do all the normal benchmarks we see like cinebench like bf1 like you know everyday or ususal products ? you know why just like i do.
as for blender benchmark as clouds of doubt have sprung up already over build settings used and such you do have to ask is it a light stripped version or something tinkered with.you really dont know.as said by many wait for real every day benchmarks before jumping on the amd blender train.
What are you babbling on about. BF1 isn't a cpu test, cinebench isn't an optimised CPU test, it's an Intel owned, biased, useful for Intel only test. Blender is a real application that real people use, handbrake is a real application that people use. Cinebench is a stand alone separate install with it's own build that is LOOSELY based on a real world application but is still Intel controlled and Intel biased.
The only company that takes Cinebench seriously, is Intel. Intel would not preview a CPU by running a game benchmark and Intel would not preview a CPU by running a AMD biased benchmark either.
Suggesting that AMD should have done both is frankly ridiculous because BF1 would have told you nothing and when picking a single cpu benchmark to show on the day no one is going to pick an Intel biased one. Nvidia don't do benchmark slides using the game they are worst in and AMD is best in, and AMD don't do gpu benchmarks in the games they are worst in that Nvidia are best in.
what ? bf1 is one of the best gaming cpu benchmarks.
especially in 64 man games.
most people know this and is the biggest fps game of the moment surely you would rather see benchmarks on something everyone uses more than a small percentage benchmark like blender?
i mentioned cinebench as a eg.by showing all the main progs for cpus used we can gather actually data instead of hearsay which is basically this whole thread.
in closing no one actually knows anything about the new AMD cpus.its that simple.
especially in 64 man games.
most people know this and is the biggest fps game of the moment surely you would rather see benchmarks on something everyone uses more than a small percentage benchmark like blender?
i mentioned cinebench as a eg.by showing all the main progs for cpus used we can gather actually data instead of hearsay which is basically this whole thread.
in closing no one actually knows anything about the new AMD cpus.its that simple.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Also not sure how people have trouble with this.
Bulldozer comparison was a 8 core Bulldozer vs 4 core 2500k and 4 core with HT 2600k. In benchmarks that used lots of threads, the Bulldozer had a natural advantage and single threaded performance wasn't clear.
The current benchmarks are being shown between an 8 core 16 thread Zen and an 8 core 16 thread Intel Broadwell-E. If it had bad multithreading or hyperthreading, it wouldn't keep up with a 16 thread broadwell-e. If it had bad single thread performance or IPC, an 8 core 16thread Zen might do well against a quad core(with or without HT) Skylake as in the Bulldozer vs Sandy comparison, but it would fail to keep up with the 6900k which has the same number of cores.
If an 8 core 16 thread chip is keeping up with another 8 core 16thread chip with relatively known performance information, then the Zen can't be that far off in either single threaded performance or multithreaded scaling.
It might be 10% slower in single thread and it gains more in multithreading situations than the Intel chip, ie one core is 90% as fast as a Broadwell-E but it gains 5-10% more performance than intel does via HT or when loading 8 cores fully. It could be the reverse, it could have 10% better single threaded performance but scales slightly worse when fully loaded. But there really isn't room for it to be 30% slower, because Intel's all cores scaling isn't bad enough for AMD to be able to make it up.
Caporegime
- Joined
- 9 Nov 2009
- Posts
- 25,339
- Location
- Planet Earth
SMT tends to have greater performance improvements on weaker cores IIRC,and lower improvements on more complex cores - its mainly a throughput improvement technology to improve utilisation of a core,by replicating part of the front end. You can kind of see this in other CPUs which use it. Some of the biggest gains in SMT scaling are seen with smaller in-order type cores,which Zen isn't. Remember,early Atom chips had SMT and they gained far more from it overall than the Intel chips with more complex cores.
Bulldozer,etc used much more shared resources and is a relatively narrow core,ie,more of a "Speed Demon" design,designed to hit high clockspeeds.
AMD originally tried to go with a CMT-like design since it was actually an attempt to make more compact cores,which would be easier to make(IIRC,I remember reading somewhere it was done to try and negate some aspects of the Intel process node lead),plus they wanted to make a high clockspeed design to make up for some aspects of this.
To achieve the high clockspeeds and the shared nature of the resources,the design impacted on IPC it seems - in fact those Jaguar cores in the consoles,are around the same IPC,and they are a much simpler core than in Bulldozer and Piledriver.
From what has been released of the Zen uarch it seems much closer to a modern Intel core design in layout and resources.
Look at how much Intel gains from SMT after so many years of using it from the Pentium 4 days??
I don't expect AMD to do much better in that regard,so IPC of Zen is probably going to be probably more comparable to some of the new Intel cores,but that could be anything from IB to Skylake,so we will need to wait and see what that is in reality.
Edit!!
Other things too - if AMD has improved branch prediction a decent amount over Bulldozer and has improved uncore performance too,it will improve single threaded performance.
Previous AMD CPUs had relatively weak memory controllers,and poor cache performance which has been highlighted in numerous articles.
Second Edit!!
In fact I kind of expected more SB/IB level IPC with Zen going from the 40% claim,but after seeing the demo,and from various leaks we are getting,I think it will be more,but we will need to wait and see. Like I said I can't see AMD getting massively higher SMT performance from an Intel like large core,than Intel with many more years of experience of implementing it have achieved.
Bulldozer,etc used much more shared resources and is a relatively narrow core,ie,more of a "Speed Demon" design,designed to hit high clockspeeds.
AMD originally tried to go with a CMT-like design since it was actually an attempt to make more compact cores,which would be easier to make(IIRC,I remember reading somewhere it was done to try and negate some aspects of the Intel process node lead),plus they wanted to make a high clockspeed design to make up for some aspects of this.
To achieve the high clockspeeds and the shared nature of the resources,the design impacted on IPC it seems - in fact those Jaguar cores in the consoles,are around the same IPC,and they are a much simpler core than in Bulldozer and Piledriver.
From what has been released of the Zen uarch it seems much closer to a modern Intel core design in layout and resources.
Look at how much Intel gains from SMT after so many years of using it from the Pentium 4 days??
I don't expect AMD to do much better in that regard,so IPC of Zen is probably going to be probably more comparable to some of the new Intel cores,but that could be anything from IB to Skylake,so we will need to wait and see what that is in reality.
Edit!!
Other things too - if AMD has improved branch prediction a decent amount over Bulldozer and has improved uncore performance too,it will improve single threaded performance.
Previous AMD CPUs had relatively weak memory controllers,and poor cache performance which has been highlighted in numerous articles.
Second Edit!!
In fact I kind of expected more SB/IB level IPC with Zen going from the 40% claim,but after seeing the demo,and from various leaks we are getting,I think it will be more,but we will need to wait and see. Like I said I can't see AMD getting massively higher SMT performance from an Intel like large core,than Intel with many more years of experience of implementing it have achieved.
Last edited:
Soldato
- Joined
- 18 May 2010
- Posts
- 23,057
- Location
- London
So as far as we can tell the info we know and the performance shown by AMD is at least believable?
Soldato
- Joined
- 19 Feb 2011
- Posts
- 5,849
Yep, no reason to doubt the Handbrake and Blender tests they showed, or the DOTA test.
We just dont know price, final boost clock, date of availability and how it fares in the average game.
did you watch the demo?
they tested bf1 on ryzen @ 3.4 ghz and 6900k at stock clocks which are higher and ryzen had better fps.
dont forget the dota streaming test where amd won by a big margin.
https://www.youtube.com/watch?v=rwtH-VHIhrU
Last edited:
Update to the latest patch, they fixed the cpu bug.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
For those who want to know, Kyle on [H] who is fairly rabidly anti AMD most of the time, got confirmation of the settings from AMD for the Blender test, ran a 6900k locked to 3.5Ghz and got effectively the exact same score as AMD showed for the 6900k.
Secondly and most importantly, he ran the test in dual and quad channel and saw precisely no difference in performance as I've pointed out before. It's not at all a bandwidth limited test(you want something with far simpler processing to cause the need for more than dual channel on a CPU, as in previous examples, winrar/7-zip benchmarks are the only real type of test I can see any significant difference with quad channel(out of home use type testing, servers with a huge user count and minimal work being done for each user probably likes quad channel mem).
It's funny because it's Intel's quad channel platform that would probably see the best utilisation of an iGPU for encoding and gaming due to the bandwidth available.
Every single benchmark I've seen of BF1 cpu testing hasn't shown realistic scaling from core or clock speed from cpu to cpu at all.
The reality is the massive majority of gamers buy a quad core i5 or an AMD chip rather than a Broadwell-E type chip. The majority of gamers will buy a Intel i5 or an Zen quad core igpu later in the year. Enthusiasts often buy stuff they really don't need. Gaming isn't and has literally NEVER been the best way to showcase CPU performance. A lot of reviews don't do significant gaming testing in CPU reviews, some ignore it completely others dedicate more time. THe majority of people buying x99 type platforms want the CPU performance for something other than gaming.
The majority of gamers are looking for a chip in the < £200 price bracket. It wasn't a launch, it was a preview, there is no reasonable expectation for them to go up on stage and spend 5 hours showing you slide after slide of dozens of different applications. AMD showed benchmarks significantly more relevant to the mass market for a Zen 8 core 16 thread chip, regardless of what gamers and enthusiasts believe(I'm both) something like rendering and full 16thread utilisation is significantly more relevant to the market these chips are addressing.
If AMD came out and tried to sell $300-500 cpu only chips intended for heavy usage to gamers and enthusiasts, the board would fire the CEO for being an idiot. She isn't, they didn't, they wouldn't, neither would Intel.
Secondly and most importantly, he ran the test in dual and quad channel and saw precisely no difference in performance as I've pointed out before. It's not at all a bandwidth limited test(you want something with far simpler processing to cause the need for more than dual channel on a CPU, as in previous examples, winrar/7-zip benchmarks are the only real type of test I can see any significant difference with quad channel(out of home use type testing, servers with a huge user count and minimal work being done for each user probably likes quad channel mem).
It's funny because it's Intel's quad channel platform that would probably see the best utilisation of an iGPU for encoding and gaming due to the bandwidth available.
Every single benchmark I've seen of BF1 cpu testing hasn't shown realistic scaling from core or clock speed from cpu to cpu at all.
The reality is the massive majority of gamers buy a quad core i5 or an AMD chip rather than a Broadwell-E type chip. The majority of gamers will buy a Intel i5 or an Zen quad core igpu later in the year. Enthusiasts often buy stuff they really don't need. Gaming isn't and has literally NEVER been the best way to showcase CPU performance. A lot of reviews don't do significant gaming testing in CPU reviews, some ignore it completely others dedicate more time. THe majority of people buying x99 type platforms want the CPU performance for something other than gaming.
The majority of gamers are looking for a chip in the < £200 price bracket. It wasn't a launch, it was a preview, there is no reasonable expectation for them to go up on stage and spend 5 hours showing you slide after slide of dozens of different applications. AMD showed benchmarks significantly more relevant to the mass market for a Zen 8 core 16 thread chip, regardless of what gamers and enthusiasts believe(I'm both) something like rendering and full 16thread utilisation is significantly more relevant to the market these chips are addressing.
If AMD came out and tried to sell $300-500 cpu only chips intended for heavy usage to gamers and enthusiasts, the board would fire the CEO for being an idiot. She isn't, they didn't, they wouldn't, neither would Intel.
Hum what?
we know a lot, it kicks Intels cpu.
anything else you want to know?
its that simple.
I built a system last August to play Doom. Had been out of gaming (using only laptops) for 10 years. When I started looking, I was surprised with the state of affairs. Everything seemed overpriced. Now this is coming from someone that has money to spend, but likes to buy within reason. Both CPUs and GPUs seemed to be absurdly priced.
On the GPU side I wanted top level and was looking at 1080 but the Founder's Edition seemed like a total rip-off. Same for the 1070 which I pondered for a while. I would have paired them with a G-Sync monitor and that's where I became infuriated (looking at prices) and gave up on them.
So I fell back to mid-range. At least there was competition there. I decided on an RX480 based on benchmarks and (educated) guess it would leapfrog ahead of the 1060 as more and more DX12 games come out. Even so I think they're a bit overpriced, but at least it keeps my options open. Didn't get a Freesync yet, but is Vega comes out offering competition at the top-end (like 480 vs 1060), it would seal it for me. I'll upgrade the monitor as well to a Freesync one.
On the CPU side however, I actually felt depressed. Sad state of affairs. I read about Zen but at the time it was all too far away and had very little info. I was encouraged though by knowing that AMD's moving to 14nm and decided it's worth waiting. For all the bashing AMD's CPUs are getting (consume a lot / are slow) people forget they're on an older process node. It was mostly FP performance and single-threaded performance that were the problem. Anyway, I figured Zen must be at least somewhat competitive just by the move to the new node. And surely AMD would try to catch up in those 2 areas it was lacking (definitely in FP; I mean they make darn GPUs, surely they know how to design FP arithmetic!)
I ended up getting an ultra-cheap FX-4350 (!) with an Asrock 970m motherboard. These let me play Doom @1440p and locked to 60FPS. This from a guy that will gladly pay $700 for the 8c/16t chip AMD demoed (or the 6900k for that matter), but I would pay neither of the two companies $1100 for it.
Having seen the Ryzen benchmarks I'm so glad I waited... I'm getting the 8c/16t as soon as it comes out. And if Vega delivers, I'm jumping on to Freesync.
On the GPU side I wanted top level and was looking at 1080 but the Founder's Edition seemed like a total rip-off. Same for the 1070 which I pondered for a while. I would have paired them with a G-Sync monitor and that's where I became infuriated (looking at prices) and gave up on them.
So I fell back to mid-range. At least there was competition there. I decided on an RX480 based on benchmarks and (educated) guess it would leapfrog ahead of the 1060 as more and more DX12 games come out. Even so I think they're a bit overpriced, but at least it keeps my options open. Didn't get a Freesync yet, but is Vega comes out offering competition at the top-end (like 480 vs 1060), it would seal it for me. I'll upgrade the monitor as well to a Freesync one.
On the CPU side however, I actually felt depressed. Sad state of affairs. I read about Zen but at the time it was all too far away and had very little info. I was encouraged though by knowing that AMD's moving to 14nm and decided it's worth waiting. For all the bashing AMD's CPUs are getting (consume a lot / are slow) people forget they're on an older process node. It was mostly FP performance and single-threaded performance that were the problem. Anyway, I figured Zen must be at least somewhat competitive just by the move to the new node. And surely AMD would try to catch up in those 2 areas it was lacking (definitely in FP; I mean they make darn GPUs, surely they know how to design FP arithmetic!)
I ended up getting an ultra-cheap FX-4350 (!) with an Asrock 970m motherboard. These let me play Doom @1440p and locked to 60FPS. This from a guy that will gladly pay $700 for the 8c/16t chip AMD demoed (or the 6900k for that matter), but I would pay neither of the two companies $1100 for it.
Having seen the Ryzen benchmarks I'm so glad I waited... I'm getting the 8c/16t as soon as it comes out. And if Vega delivers, I'm jumping on to Freesync.
Last edited:
Soldato
- Joined
- 11 Sep 2013
- Posts
- 12,488
what you on about.i have adequate fps
the point is bf1 actually uses your cpu and scales well when used. the benchmark they did wasnt even in action.
as said 64 man mp game in action.
it really tests cpus.
Does it ? My 6800K doesn't even sweat playing this game and at no points has the cpu been a limiting factor, it scales well with threads granted but lets put it this way i wouldn't see an improvement even with a 6950x as my titan xp is still the bottleneck at 1440p 144hz
Well yes because initial impressions were "zomg I've just run the AMD demo and Zen obliterates my own CPU it's amazing".
They've also not provided any verifiable benchmarks, like I said all we have is their own demo file for Blender which could artifically inflate Zen performance at the cost of other architectures. I mean look on this forum how people have gone mad over the AMD Ryzen Blender benchmark as though it's the be all and end all, if the code favours Zen architecture it could well be that 8 core Zen is closer to Intel 6 core in most normal situations.
It's like when Bulldozer was being hyped and AMD provided only encoding benchmarks that used full use of all of the cores and latest instruction sets to show it beating 2600K, how did that turn out?
You're aware Blender is an open-source renderer? There's no way it could be biased towards Zen - how would the devs have got access to tuning for an unreleased architecture? There would be evidence all over the commit logs, etc. The version linked in the other thread is just the bog-standard Windows x64 MSI installer for Blender 2.78.
Anyway the results aren't that amazing that anyone should expect cheating. Core-for-core and clock-for-clock it's about 5% quicker than the best CPUs from Intel. The next Intel revision will probably catch up. The big question now as SiDeards73 says is price and how quickly they can get them out of the door.