Caporegime
my phone messed up, ignore.
Last edited:
-
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
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
I don't remember saying we did represent the majority of the market. Gamers represent a FAR larger portion of the market than those who buy x99 platform type chips. The richer people with more disposable income are very happy to buy chips with more cores and no iGPU, most of the rest of the gaming segment would be too, just without any affordable option.
It's cheaper to reuse existing cpus, always has been, always will be. But they'd have sold 3 chips to me since the 2500k if they had a non iGPU mainstream option, they haven't so I haven't bought them. There is more than enough profit to be made if they offered such chips. I'm actively discouraged from upgrading because HT itself isn't a monumental leap in performance in many circumstances and the jump from Sandy to Ivy to Haswell, to Broadwell, to Skylake has focused on iGPU performance, for almost no one using iGPU's to their full extend.
When you run 4k video or use powerpoint and display it, you're using sub 5% of the power of the iGPU available to you, you aren't really using any of the shader power at all. 99% of current gpu acceleration would run just fine on one or two blocks of shaders along with the existing video processing blocks and video output blocks. The massive majority of ALL users do NOT use the iGPU much at all. It could have 90% of the shading power removed and maybe sub 3% of people would notice the performance difference, the rest would notice cheaper chips and longer battery life.
I'd take a 6-7 core 4970k with a very small iGPU that let it run video/outputs/anything short of 'real' gaming in a heartbeat over a 4 core plus massive iGPU almost no one uses.
It's cheaper to reuse existing cpus, always has been, always will be. But they'd have sold 3 chips to me since the 2500k if they had a non iGPU mainstream option, they haven't so I haven't bought them. There is more than enough profit to be made if they offered such chips. I'm actively discouraged from upgrading because HT itself isn't a monumental leap in performance in many circumstances and the jump from Sandy to Ivy to Haswell, to Broadwell, to Skylake has focused on iGPU performance, for almost no one using iGPU's to their full extend.
When you run 4k video or use powerpoint and display it, you're using sub 5% of the power of the iGPU available to you, you aren't really using any of the shader power at all. 99% of current gpu acceleration would run just fine on one or two blocks of shaders along with the existing video processing blocks and video output blocks. The massive majority of ALL users do NOT use the iGPU much at all. It could have 90% of the shading power removed and maybe sub 3% of people would notice the performance difference, the rest would notice cheaper chips and longer battery life.
I'd take a 6-7 core 4970k with a very small iGPU that let it run video/outputs/anything short of 'real' gaming in a heartbeat over a 4 core plus massive iGPU almost no one uses.
Caporegime
Joy is right. In Cinebench pd only has one 128Bit FPU instead of the modular combined 256Bit FPU that its supposed to run with as single threaded, the result is much lower performance than its capable of.
In multithreaded it runs in 4 core 8 128Bit thread configuration.
In multithreaded the 8 thread FX-8300 @ 5Ghz scores about 800, the 8 thread Haswell about 1000. That's 20%, its right there in your own link.
Oh and stop trolling. Its pathetic.
Soldato
Dude you've got it wrong. 1000 points is a 4790k which is quad core + HT so not a full 8 threads. Take the score of a 4690k and double it for a true representation of Haswell in Cinebench.
Soldato
I think everyone's hoping we get better than a 40% improvement based on full-module performance. Given that the 4c/4t Haswell i5 matches the FX-8 in that benchmark at matching clocks, it would be disappointing to say the least if the big 8c/16t Zen was only equivalent, at full throttle, to a hypothetical 6c/6t Haswell.
A 40% boost in single-core performance would be great, and would get AMD right back in the game. A 40% boost in comparable "module" performance would not.
Caporegime
Piledriver never operates as an 8 core CPU in FP work, In multithreaded its a 4 core 8 thread similar to a 4 core i7, the difference being the i7 has 4 big threads and 4 really small ones, pd has 8 medium ones. 128Bit.
In single threaded its supposed to run its 2 128Bit threads combined to make one 256Bit thread, tho in the case of cinebench thats not happening, hence the disparity in percentage scalling single vs multi.
Zen is supposed to have the same layout as intel, with 8 independent cores with 2 theads per core.
That with the +40% should make an 8 core Zen not far from 3x as fast as an FX-8300.
Last edited:
Soldato
- Joined
- 27 Feb 2012
- Posts
- 6,586
Gotta love your wild speculation, you sure you're not hired by AMD?
Caporegime
121x8 = 968.
But Nkata's FX83 gets 803.
188x4 = 752.
My 4670K got 731.
Doesn't this prove my point?
Yeah, I'm not trolling Humbug.. It's the same crap, different day.
EDIT : Hold on, I'm trolling, but you think almost 3 times FX83 performance is possible?! Holy Crapballs. Wow. Just Wow.
But Nkata's FX83 gets 803.
188x4 = 752.
My 4670K got 731.
Doesn't this prove my point?
Yeah, I'm not trolling Humbug.. It's the same crap, different day.
EDIT : Hold on, I'm trolling, but you think almost 3 times FX83 performance is possible?! Holy Crapballs. Wow. Just Wow.
Last edited:
I don't care at all about MT, that's a gimme. What I want is a cheap alternative to the 5960X with same or better ST (and lanes) for games, which will continue to require 90+% ST for years still. A couple of D3D12 titles in 2016 won't let AMD slide on raw horsepower.
AMD to Emphasize on "Generation" with Future CPU Branding
http://www.techpowerup.com/212353/amd-to-emphasize-on-generation-with-future-cpu-branding.html
http://www.techpowerup.com/212353/amd-to-emphasize-on-generation-with-future-cpu-branding.html
Caporegime
Your behaving like some teenaged emo. Constant moody swings
Of course not, multithreading is not linear scaling, your own multithreading scaling shows that.
With 8 128Bit theads and perfect scaling pd would get 880, it got 800 about a 10% difference,
If you look at the 8 thread i7 it has the same single thread performance as the i5,
Why, because cinenema 4D does not operated rendering in this single threading form, for the benchmark all its doing is shutting down all the other threads, in the case of Intel its the HT thread. In the case of AMD its the other 128Bit thread.
Cinebench is not representative of single threaded performance as its simply shutting of all other threads ignoring the HT in Intel and combined threads in AMD.
AMD have 8 128Bit threads for multithreading and one 256 bit in single always in 4 core conf. Again its not linier as it still operates from the same CU, the performance should be somewhat higher. The 128Bit thread from that CU is a bottleneck In cenebench gimped benchmark.
Last edited:
Caporegime
Humbug, do you actually know what you're saying? You were saying that the Single thread on Cinebench was the problem, so I changed the rules and you were wrong (Again)
My scaling is within 2%, the FX83's is 20% out. That's a MASSIVE difference. Hilariously 20% is the figure that AMD themselves tout (As per the BD Slide from 2011) as the performance hit from the module approach.
All I see is you just trying to change the rules to suit your own agenda (But because what you're doing is flawed, a rebuttal will just be as flawed)
The i7 having the same single threaded as the i5 is completely expected, it's meant to.
I don't pretend to know more than I do, and ultimately I end up wrong rarely. Whereas you talk the talk, but you don't have a clue.
If people like you didn't defend inferior products to the hilt like some type of religious cult, we'd probably have more parity.
The hilarity of the situation is I'm excited about Zen, but again you're going completely off the deep end (With the 3x higher performance claim) and no doubt when it falls short of that you'll pretend you never even said that (Not that it matters) my expectations of Zen are basically anywhere from 40-50% better performance across the board core for core clock for clock, and that'd be enough for me and many others to buy.
Cinebench is FPU performance, gaming is FPU performance. And I wasn't the one with the Cinebench comparison, that was Joey. I just mentioned (And correctly) that the IPC can be in excess of 50% (Which it can) but the argument is a bit of a fallacy (Which it is). Yet you persist.
My scaling is within 2%, the FX83's is 20% out. That's a MASSIVE difference. Hilariously 20% is the figure that AMD themselves tout (As per the BD Slide from 2011) as the performance hit from the module approach.
All I see is you just trying to change the rules to suit your own agenda (But because what you're doing is flawed, a rebuttal will just be as flawed)
The i7 having the same single threaded as the i5 is completely expected, it's meant to.
I don't pretend to know more than I do, and ultimately I end up wrong rarely. Whereas you talk the talk, but you don't have a clue.
If people like you didn't defend inferior products to the hilt like some type of religious cult, we'd probably have more parity.
The hilarity of the situation is I'm excited about Zen, but again you're going completely off the deep end (With the 3x higher performance claim) and no doubt when it falls short of that you'll pretend you never even said that (Not that it matters) my expectations of Zen are basically anywhere from 40-50% better performance across the board core for core clock for clock, and that'd be enough for me and many others to buy.
Cinebench is FPU performance, gaming is FPU performance. And I wasn't the one with the Cinebench comparison, that was Joey. I just mentioned (And correctly) that the IPC can be in excess of 50% (Which it can) but the argument is a bit of a fallacy (Which it is). Yet you persist.
Last edited:
Soldato
Yes! Finally after 10 years the AMD vs Intel flame wars have come out of retirement with the announcement of AMD's Zen CPU!!!!!!
Caporegime
If you just calm down for a minute you might stop to think and realize what's going on is obvious.
If cinebench doesn't shut down all other threads its rendering on 2 threads, IE 2 images at the same time, as it does 4 for the i5, 8 for the i7 and FX-8300, the fact that its not using the other thread from intel is not a big deal given that its only 5% of the total pipeline, on the AMD side its 50% of total, so the 128Bit thread from the CU is a massive bottleneck given that it should be 256Bit.
Caporegime
If you say so.
Last edited:
Please keep it civil chaps, sorry for opening a can of worms.
I think the point is that with SSE optimised code (which is all Cinebench R15 uses, no AVX or higher) both architectures have the same theoretical max (8 flops/cycle) per module/core. So it's not that surprising that we see adjacent results like this in the table:
(I just plucked these out, the rest of the table isn't so neatly ordered obviously, perhaps because the hyperthreaded CPUs are more efficient, and other component (memory mainly) differences?)
What's weird is the "single-threaded" test results, because there the gap between Intel and AMD is quite a bit larger.
Notice I picked the same users to try and do a fair comparison. Why would the Intel result now be 70% higher than the AMD one? We know that flat-out (4 FPUs each) these CPUs were posting the same numbers. Seems to me it must be an implementation problem, so I think humbug is onto something.
I think the point is that with SSE optimised code (which is all Cinebench R15 uses, no AVX or higher) both architectures have the same theoretical max (8 flops/cycle) per module/core. So it's not that surprising that we see adjacent results like this in the table:
(I just plucked these out, the rest of the table isn't so neatly ordered obviously, perhaps because the hyperthreaded CPUs are more efficient, and other component (memory mainly) differences?)
What's weird is the "single-threaded" test results, because there the gap between Intel and AMD is quite a bit larger.
Notice I picked the same users to try and do a fair comparison. Why would the Intel result now be 70% higher than the AMD one? We know that flat-out (4 FPUs each) these CPUs were posting the same numbers. Seems to me it must be an implementation problem, so I think humbug is onto something.
Last edited:
Caporegime
The difference in the scaling is still 20% for the AMD.
Although the i5 scaling is out by 4%. It's possible his RAM etc was better set up for his single run.
And it being 70% out isn't a massive difference from the 67% that I'd already shown. You're trying to make a point that isn't there, and far too much focus is being put on buzzwords.
Although the i5 scaling is out by 4%. It's possible his RAM etc was better set up for his single run.
And it being 70% out isn't a massive difference from the 67% that I'd already shown. You're trying to make a point that isn't there, and far too much focus is being put on buzzwords.
Last edited:
Soldato
Unless I'm missing something (I might very well be) that doesn't quite make sense.
It doesn't explain why in the multi-core bench 8 AMD cores are not performing 8 times as well as 1 (from the single-core bench). The i5 scales up 4 times, and the i7 gains 4x plus extra from the hyperthread cores.
What does the FX8 lose, core for core, going from single-core bench to multi-core, if it's not the single floating point per module handicap?
Caporegime
The CU can't push its calculations through a 128Bit pipe as well as it can a 256Bit pipe, an actual bottleneck.
Its only cinebench that operates in this way and it does to give you a single threaded performance option. But its flawed as even cinema 4D its self does not operate in this way. Its fully threaded.
Its only cinebench that operates in this way and it does to give you a single threaded performance option. But its flawed as even cinema 4D its self does not operate in this way. Its fully threaded.
Caporegime
Again, just buzzwords to make up excuses. But I guess it makes people think you know what you're talking about.
EDIT : Can someone show me a PD Module pic that shows it has an actual 1 256bit part? All images I find are two 128bit parts in the Module.
EDIT : Can someone show me a PD Module pic that shows it has an actual 1 256bit part? All images I find are two 128bit parts in the Module.
Last edited: