jbltecnicspro
[H]F Junkie
- Joined
- Aug 18, 2006
- Messages
- 9,547
Oh I agree. Intel is so fucked.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
AMD: Zen will offer 40% faster performance per clock than Carrizo
- Thread starter Mr. Bluntman
- Start date
Oh I agree. Intel is so fucked.
Dayaks
[H]F Junkie
- Joined
- Feb 22, 2012
- Messages
- 9,774
I am hoping Zen does hit Q3-Q4 2016. We promised the offspring a laptop and would love to use a competitive AMD product with HBM on the die. Make it happen AMD!
I will make a series of statements. You can agree or disagree.
1. IPC isn't the end all of CPU performance. It is when you are talking about pure single thread. Multi-core scaling is still very important. It is very possible for a AMD cpu with 16 threads to be beaten in single thread and perform better in multi threaded and vice versa.
2. 40% IPC gain at the same clock rate, could be a number of things. It can be 40% better in this one benchmark, it could be on average 40% better, and lastly it could be at least 40% better over Excavator in all areas. Making it really pointless to say only 40%. Another thing to note is AMD themselves say the IPC is 40% better than excavator, so carrizo? which is only a mobile processor without l3. The piledriver Fx cpu IPC is higher than kaveri or carrizo, thanks to l3 and higher clocks.
3. Everyone is talking about the CPU only, honestly I'm more interested in what its chipset can do how many pci-e lanes, channels of ram, sata ports, etc etc,
4. People are saying AMD Zen core will be lucky to match Sandy Bridge. I woefully disagree, keep in mind Sandy Bridge was produced on 32nm tech. AMD with Zen will be on 14nm. If the Zen core is as fast as sandy bridge in IPC then AMD would be entirely too naive to bring it to market and expect good things. Anything less than Haswell will be a disappointment.
5. Honestly the biggest excitement is on the mobile side of things. AMD laptops have lacked serious cpu horsepower for quite some time now. Intel has huge price gaps between their i3, i5 and i7 lines, with the introduction of a more competitive Zen chip these price gaps will be smoothed out. Intel will no longer be able to command a 200$ increase in price for each segment.
1. IPC isn't the end all of CPU performance. It is when you are talking about pure single thread. Multi-core scaling is still very important. It is very possible for a AMD cpu with 16 threads to be beaten in single thread and perform better in multi threaded and vice versa.
2. 40% IPC gain at the same clock rate, could be a number of things. It can be 40% better in this one benchmark, it could be on average 40% better, and lastly it could be at least 40% better over Excavator in all areas. Making it really pointless to say only 40%. Another thing to note is AMD themselves say the IPC is 40% better than excavator, so carrizo? which is only a mobile processor without l3. The piledriver Fx cpu IPC is higher than kaveri or carrizo, thanks to l3 and higher clocks.
3. Everyone is talking about the CPU only, honestly I'm more interested in what its chipset can do how many pci-e lanes, channels of ram, sata ports, etc etc,
4. People are saying AMD Zen core will be lucky to match Sandy Bridge. I woefully disagree, keep in mind Sandy Bridge was produced on 32nm tech. AMD with Zen will be on 14nm. If the Zen core is as fast as sandy bridge in IPC then AMD would be entirely too naive to bring it to market and expect good things. Anything less than Haswell will be a disappointment.
5. Honestly the biggest excitement is on the mobile side of things. AMD laptops have lacked serious cpu horsepower for quite some time now. Intel has huge price gaps between their i3, i5 and i7 lines, with the introduction of a more competitive Zen chip these price gaps will be smoothed out. Intel will no longer be able to command a 200$ increase in price for each segment.
horrorshow
Lakewood Original
- Joined
- Dec 14, 2007
- Messages
- 9,443
horrorshow
Lakewood Original
- Joined
- Dec 14, 2007
- Messages
- 9,443
True, (IPC * clock rate) * (core count * thread scaling) is what I use
The first part is per-thread performance, the second part is relative core count scaling, together we have the whole CPU performance.
No doubt, it could also be 0% in five benchmarks, 100% in a few, and 70% in another. I expect this to be the pattern we see, with much less contrast, of course.
Zen will be using a new data fabric made by the same designer who created the amazing HyperTransport technologies. This is supposed to be 100GB/s per link, and we have no idea how many links, but we should expect Zen to provide one heck of a platform. In addition, having partnered with AsMedia for their IP and chipset implementation, we should actually expect AM4 platforms to match or exceed Intel's platform as AsMedia parts have been mostly beating out Intel solutions, even when hamstrung by inferior interconnects.
Agreed, my detailed analysis based on actual benchmarks shows that Zen should be very very slightly faster than Haswell on an average basis assuming the 40% over Excavator claim is distributed as I'd expect. By slightly, I'm saying Haswell is 139.39% faster than Penryn overall, and Zen will be 140.03% faster than Penryn overall - both in terms of IPC.
On #5, I don't care about the mobile world
I really do hope AMD makes a comeback with Zen. Though I'll probably be going with Skylake-E personally (Broadwell-E if my current system craps out...) for my main system, I'll be very happy to support AMD if they are competitive with performance and price for other builds.
In short it is too soon to tell. How about delays? Process? Respin delays? New platform delays? Some rather huge changes both for the cpu and the apu variants.
As for performance it still comes down sometimes perf/$ and what you are doing with it.
As for performance it still comes down sometimes perf/$ and what you are doing with it.
It isn't the end all be all, but it's an important factor. Even if that 40% number represented across the board improvements over Carrizo it wouldn't be enough to catch Intel. As was already pointed out that's not until 2016 when Intel will have either Broadwell-E or a Skylake-E variant by then. Keep in mind they'd still be behind Skylake.
Well let's analyze this statement. IPC at the same clock rate could be a number of things depending on how it's reported. I'll agree with that. I'd disagree that AMD would use the average number. AMD has a history of using the best case scenario to drum up hype for it's CPUs. As we mentioned earlier AMD touted a 50% increase over the Core i7 950. It achieved this in one benchmark. And again the article says Zen will offer 40% more performance than Carrizo. So that means we could be looking at a smaller improvement compared to Piledriver which wouldn't put AMD in a very good place.
I touched on this before. As someone with a keen interest in motherboards I can see where you are coming from. I don't quite agree with your statements though. First off, more PCIe lanes won't equate to more performance in games. It won't help anywhere that really matters. If you have a hard on for storage then AMD has a chance to do something like provide direct access for NVMe M.2 drives through the PCIe controller without going through a PCH and a DMI bus. They could leverage RAID 0 performance better in such an event. While enticing you have to keep in mind that it wouldn't make any difference in game or application performance. I/O intensive applications would benefit from it but you can already do that with PCIe based SSDs.
Let's say AMD went with more RAM channels. That would force memory makers to make special kits for Zen based systems. Not out of the question but it could lead to increased costs or availability issues in the short term. What would AMD gain? We know RAM performance is largely meaningless in gaming and general application performance. Haswell-E has far more memory bandwidth than Skylake, but that doesn't equate to better gaming performance. Most applications won't benefit from it either. Lets not forget having quad-DDR4 channels takes up a ton of motherboard real estate. This equates to cost and limitations for various form factors. Not a smart move.
SATA ports you say? Well that hardly matters. You've got up to 10 SATA 6Gb/s ports on Intel chipsets. Few people will make use of so many. SATA Express doesn't mean jack shit right now. There is hardly any incentive to support it other than it's backwards compatible with SATA. I've seen up to three SATA Express ports on motherboards all of which will generally go unused.
We are up to a ridiculous amount of USB 3.0/3.1 ports now. This is hardly a selling point to have more. AMD won't likely go with more as more logic and connections go onto the board raising costs. AMD needs to avoid that. I'd like to see native USB 3.1 support, Thunderbolt 3.0 support, HDMI 2.2, DisplayPort 1.3 support and more PCIe lanes native to an integrated PCIe controller onboard the CPU. If they have to go through a PCH I would like to see a transport for that which is an 8x link instead of the 4x PCIe link DMI 3.0 uses. I do expect relative parity with the platform vs. Intel, but I don't expect anything revolutionary.
The problem is AMD's 40% number and how they arrived at that figure. If they keep within the tradition of previous bullshit estimates, is a pipe dream at best. AMD likes to take the best case scenario and publish that as the numbers to expect. When the part launches we find that number was valid for one or two scenarios and that was it. Ultimately this is just speculative so we really have no idea at this point. But Zen being 14nm doesn't have jack shit to do with it's IPC in comparison to Sandy Bridge. There are more optimizations in the newer CPUs than just a simple die shrink which result in the performance gap between Sandy Bridge and Skylake. I suspect that Zen will fall in line right behind Haswell / Haswell-E. Unfortunately for AMD we'll probably have or be close to the release of Skylake and Broadwell-E / Skylake-E by that point. Matching Haswell at the right price point would be ideal as a minimum. That would put AMD back in the game. Falling short of that won't be good.
If AMD reported 40% as an average number with highs above that, they'd be golden. I just don't think that's the case based on the company's previous behavior.
You make a good point. That could end up being the market where Zen shines. If Zen isn't a home run, there is still K12. So as a business they could continue. If Zen isn't a success then AMD's finished in the consumer CPU market. If K12 fails too then AMD's done for good.
TheBlueChanell
Supreme [H]ardness
- Joined
- Apr 15, 2005
- Messages
- 4,658
Not to nit-pick but IPC can't be higher because of clocks that negates IPC, instructions per clock-cycle. IPC comparisons need to be done at the same clock-speed.
Not 100% certain either and I'm too lazy to check but I was almost certain during the Kaveri review Anand compared it's IPC with that of Bulldozer/Piledriver and it was indeed higher even with the missing/disabled cache. I could be wrong there and it also may of been Godavari but I recall reading that and thinking it was interesting.
TheBlueChanell
Supreme [H]ardness
- Joined
- Apr 15, 2005
- Messages
- 4,658
lmao. I can't believe anyone thinks anything AMD or any ARM partner could do would put Intel out of business.
At one point, DEC, MIPS and others never thought Intel could dethrone them in the server market. ARM based servers are now filling the low to medium end roles once occupied by Intel servers. Intel has more cost flexibility than DEC or other platforms had back in the day. So Intel isn't vulnerable in the same way, but I've seen the seemingly impossible happen in a few short years in the two decades I've been in the industry.
But I'd hardly call Intel fucked at this point.
Will Zen compete with Haswell in October 2016 and be only (let me count, Haswell, Broadwell, Skylake, Kaby Lake) only four generations behind Intel, assuming it doesn't get delayed into Cannonlake territory in 2017? What's stopping Intel from lowering the prices on whatever Zen does end up competing with, since that will be anywhere from two to four generations behind anyway?
MrGuvernment
Fully [H]
- Joined
- Aug 3, 2004
- Messages
- 21,823
You may be surprised how many people still buy AMD cause "OMGZZZZ 8 COREZZZ" for gaming...
and you tell them how much power and heat and they have no clue cause mommy and daddy pay the bills.
Mr. Bluntman
Supreme [H]ardness
- Joined
- Jun 25, 2007
- Messages
- 7,088
$200 or less for motherboards at the very top, yes it would be a win.
Nehalem is past its prime, so this shouldn't be a difficult hurdle for AMD to climb. Phenom II was already between Conroe and Penryn performance most cases. The general consensus is that Zen should at least be as fast as SB/IB up to Haswell levels at best. Happy upgrading next year
Last edited:
Probably true to a point... but some people just want raw power. They might not be leaving their computer on and only use it for gaming. So it could be a hog, but the hogging would be for like an hour or 2.
Mr. Bluntman
Supreme [H]ardness
- Joined
- Jun 25, 2007
- Messages
- 7,088
If they're buying a Bulldozer based PC for raw power today (unless we're talking about strictly power consumption), they're doing it wrong. Even an i5 3570 is faster in 90-95% of programs and uses less than two thirds the power.
We don't know what it will compete with. If it matches Haswell there is nothing stopping Intel from slashing prices and destroying them. They won't but they could do so now if they wanted to.
While ARM is starting to push in the server market, it will probably hurt AMD more than Intel.
Intel now has mid/low end Xeon Cpus to take on ARM while AMD has nothing. Plus Intel's reliability is second to none, its like Intel lasts forever. ARM has to prove itself in that regard.
It took decades for Intel to become the leader in the server market. Surely they won't let it go without a fight.
![Zarathustra[H]](https://cdn.hardforum.com/data/avatars/m/9/9298.jpg?1707758471){kind=avatar}
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 38,881
Agreed.
Even an i3 is faster than a max overclocked FX-8350 in many applications, especially where per thread performance is important.
It's difficult to express how far behind AMD really is with their current line of CPU's.
I mean, we are talking about them making a leap of 40% over their current designs over a year from now, and still only catching up with where Intel was 5 years ago...
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
Zarathustra[H];1041875701 said:
Well let's not get carried away. My wife's rig sports a i3 and both my rig (sig) and my boy's rig running a Phenom II x6 1090 run circles around it. I don't run synthetic benchmarks so maybe an i3 will really shine in those but in real world use of a PC (web surfing, photo editing and word processing often at the same times while listening to iTunes), I'll take an 8350 over an i3 any day. I totally agree they're way behind the game right now but let's not go crazy saying they can't even keep up with a dual core.
Honestly, in pure FPS/Frame times they can hardly.
But in other things (normal desktop usage) i won't know.
I think CPUs should not really be compared that much for gaming at all, because they are not really needed to be powerful above a certain threshold.
And then FPS Benchmarks make it seem like i3 is only slightly slower than a 999$ cpu from intel
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
Yeah most things we do with our PC's are barely a workload for most all CPU's these days. I mean my 8350 is obviously pretty dated but my rig runs great. I'm getting all the frame rates I see other 290x's get and in everyday use it is very fast. Now if I were encoding huge video files or running dual Ti's pushing 4K resolution then it would really be nice to have an i7 but for 95% of the things we do with PC's, like single GPU gaming at 1080 for me or photos and word processing for the wife, 8350's and even i3's are plenty of muscle and I'm not sure you'd notice a faster CPU.
All that means that AMD doesn't have to come out with an Intel killer. Just bring something that comes close enough to be a viable alternative then price it cheaper and it'll sell plenty.
All that means that AMD doesn't have to come out with an Intel killer. Just bring something that comes close enough to be a viable alternative then price it cheaper and it'll sell plenty.
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 38,881
I would disagree.
My work laptop is a 1.8Ghz dual core (albeit with HT) i5-3427U with turbo up to 2.8Ghz. I can't tell any difference in desktop responsiveness between it, my desktop at home (i7-3930k, 6 cores with HT @4.8Ghz), the rig I built for my stepson from spare parts (FX-8350) or my old HTPC (A10-7850K)
(I do notice a difference on my new HTPC, a Chromebox Broadwell dual core Celeron @ 1.4Ghz though)
Desktop responsiveness really isn't that dependent on CPU power. it's more dependent on how much RAM you have and how fast your drives are. Slap 8GB of RAM and an SSD in most computers and they feel more or less the same on the desktop.
I don't buy dual core CPU's for desktops, but I do use single threaded benchmarks as a predictor for how capable an architecture is, as you really can't count on multithreading for all applications, but if you have fast threads, they will be fast in everything.
I recently did this comparison in another thread, after someone suggested that FX-6300 and FX-8320 chips were better bang for the buck chips than lower end Intel chips.
Zarathustra[H];1041863137 said:
I knew things were bad for AMD, but where I really was shocked was that the FX-8320 performed similarly to the Atom C2750 in the 3d particle computation test
I was not expecting that at all.
Mr. Bluntman
Supreme [H]ardness
- Joined
- Jun 25, 2007
- Messages
- 7,088
Since out of order execution was added Atom isn't the chump it used to be. Lest ye forget that the ALUs to Bulldozer and descendants are 2-issue wide, at least on the INT side.
Last edited:
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 38,881
Interesting. I haven't looked into Atom's in a long time.
I wonder then how a modern dual core atom would perfom comapred to - say - my dual core broadwell Celeron at 1.4Ghz.
Mr. Bluntman
Supreme [H]ardness
- Joined
- Jun 25, 2007
- Messages
- 7,088
There are quad core SoCs that are clocked as high as 2.16GHz, so I would imagine it's close
Exactly what i stated before, AMD cannot compete with Intel, and should not even aim to.
But Zen can definitely be a viable alternative if it has Sandy Bridge or Ivy Bridge Performance, with more cores than an i5 and priced a little below the new 6600k.
I don't think we will see ZEN high end PCs in best buy, but we could surely see people thinking about it before buying the 6600k.
I think most people don't even consider AMD anymore because of the FX Chips
But Zen can definitely be a viable alternative if it has Sandy Bridge or Ivy Bridge Performance, with more cores than an i5 and priced a little below the new 6600k.
I don't think we will see ZEN high end PCs in best buy, but we could surely see people thinking about it before buying the 6600k.
I think most people don't even consider AMD anymore because of the FX Chips
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 38,881
I know what you mean.
I don't necessarily NEED the fastest CPU on the block. I'd be OK with one that can run the stuff I want (my favorite game happens to also be one that is unusually CPU heavy) and right now, that simply rules out anything AMD.
Zen doesn't have to be the fastest chip on the block to breathe new life into AMD. Simply being fast enough for consideration is enough.
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 38,881
Let's just hope AMD can avoid the Osborne effect.
Zarathustra[H];1041875920 said:
I bought and returned a fanless tablet running a quad core atom @ 1.8 Ghz. I think it was the ASUS T100. Got it for like 199 cad was a good deal
While it was not enough to be used as a laptop for me (screen/keyboard), it definitely surprised me how well it performed.
It was able to open folders, webpages almost instantly, and could even play Dota at 20 - 30 Fps.
Last time i owned an atom was when i was in high school and it was a fat ass "netbook" couldnt even play youtube. I think they have come a LOOOOONG way.
defaultluser
[H]F Junkie
- Joined
- Jan 14, 2006
- Messages
- 14,398
Zarathustra[H];1041875920 said:
Bay Trail not sucking is one of the reasons Windows tablets and convertibles took off so aggressively in the last year.
That, and Intel not charging anything for the processors, and Microsoft giving away the OS
The per-clock performance is similar to than AMD's Jaguar, but the clock speeds Intel can achieve are higher, so it's faster overall. Unlike the previous in-order Atom, you'd actually nnotice the difference if you plugged-in an SSD into one of these machines.
Agreed on all points. I'm just saying K12 could be good for AMD in that market.
Michaelius
Supreme [H]ardness
- Joined
- Sep 8, 2003
- Messages
- 4,684
Not really K12 was just another in long series of wunderwaffe desings that were suppossed to be shortcut to succes.
At least with Zen they seem to be back to roots.
At least with Zen they seem to be back to roots.
defaultluser
[H]F Junkie
- Joined
- Jan 14, 2006
- Messages
- 14,398
Yeah, people swore up and down that K12 was faster than Zen, but the reality is that it's not THAT much faster.
People always claim ARM will run amok in the server room. They point to all those RISC processors that Intel defeated back in the 90s. But the difference then is that Intel was not just better, they were compatible with software most people used every day IN THE OFFICE, AND AT HOME.
In a world where each major application family had it's own EXPENSIVE closed hardware ecosystem AND OS, x86 plus DOS plus Netware (and later Windows NT) had a lot going for it. And since Intel plus their competitors (AMD, Cyrix) had a large product lineup by 1995, you really could get a 32-bit PC at almost any price point.
ARM can't be THAT fast or cheap because there's not enough of a performance overhead in x86, and they can't make inroads on the compatibility front because there is none. People don't use their smart phones to get real work done, just to communicate.
Once we standardized on x86 for everything, you'd pretty-much have to be compatible to have a chance to compete.
Last edited:
I think we are on the same page. I agree with you on the bulk if not all of your points. Having said that, I still see some potential in K12, at least for the short term.
From a pure sales standpoint, K12 could be a good product because ARM CPUs are popular right now in a lot of devices. If K12 performs well and the design scales to different usages nicely, and if its priced aggressively enough it could be a high volume product for AMD. No it won't let them win the server market, it won't get them the desktop market, or even the laptop market. It could at least generate some much needed cash. ARM CPUs are in a ton of things. If AMD can get the K12 into a lot of those devices then it's a home run. If AMD really wants to stay in the consumer desktop business then what AMD really needs is a solid consumer x86 CPU priced aggressively that can scale well in mobile and server markets. They need a solid platform to use with it which would allow aggressive competition with Intel in certain segments. If they can do that then they can get back in the fight. Which frankly everyone here should hope for exactly that.
MacLeod
[H]F Junkie
- Joined
- Jul 28, 2009
- Messages
- 8,280
Zarathustra[H];1041875873 said:I would disagree.
My work laptop is a 1.8Ghz dual core (albeit with HT) i5-3427U with turbo up to 2.8Ghz. I can't tell any difference in desktop responsiveness between it, my desktop at home (i7-3930k, 6 cores with HT @4.8Ghz), the rig I built for my stepson from spare parts (FX-8350) or my old HTPC (A10-7850K)
(I do notice a difference on my new HTPC, a Chromebox Broadwell dual core Celeron @ 1.4Ghz though)
Desktop responsiveness really isn't that dependent on CPU power. it's more dependent on how much RAM you have and how fast your drives are. Slap 8GB of RAM and an SSD in most computers and they feel more or less the same on the desktop.
I don't buy dual core CPU's for desktops, but I do use single threaded benchmarks as a predictor for how capable an architecture is, as you really can't count on multithreading for all applications, but if you have fast threads, they will be fast in everything.
I recently did this comparison in another thread, after someone suggested that FX-6300 and FX-8320 chips were better bang for the buck chips than lower end Intel chips.
I knew things were bad for AMD, but where I really was shocked was that the FX-8320 performed similarly to the Atom C2750 in the 3d particle computation test
Yeah I agree with you that in single threaded, synthetic benchmarks, AMD gets shredded. However I think that single threaded, synthetic benchmarks arent a true measure of how well a CPU will perform in real world use. Its kinda like how a 0-60 time isnt a true measure of how well a car performs. Dont misunderstand me though, Im not trying to discount IPC. Obviously its very important but in in this day and age when most things are not very CPU intensive in the least and most people use their PC's for several different things and often at once, 8 slower cores will be better to have than 2 fast cores and even in the single threaded applications most people use, a FX-8300 is going to have fast enough IPC that youre not going to notice.
So when you look at i3's going for around $120 and FX-8320's going for $140 and i5's going for $190 not to mention the FX is the only unlocked and overclockable chip in that bunch, their overall performance falls about in line. A 8320 should be able to be bumped to 4.5 at close to stock voltage if not at stock.
So yeah, AMD is outclassed by i5's and i7's by a long shot. But they're still a viable in the budget end. Like I said, my rig is outdated but I cant find any area where Im being held back. This thing still rips through games, zips through the light photo editing I do while running various other programs in the background. But if I were to step up to some heavier duty tasks like 1440p, SLI/Crossfire gaming, Id be moving over to Intel.
Either way, its still pretty sad that the only thing I can compare FX chips to in order make them look good are DUAL CORE, entry level chips. Sad indeed. Zen really needs to deliver.
Oh come on now. Almost all of the server space targeted by ARM is on Linux (web servers, etc.) and Linux / LAMP have been running on ARM for ages.
It's ultimately going to come down to who has the best product at the best price. Software compatibility isn't an issue.