Leaked AMD Zen Engineering Sample Benchmarks?

Pieter3dnow said:
Yeah this does not work for me when you type this:


No one would buy AMD stuff that is outdated by 5 years as hardware upgrade , this suggests that you have an opinion of people that do want to buy this "instantly outdated" Zen ..
Also very good insight as to what the Zen retail performance would be.
Click to expand...
Ah, fair point. I personally have no idea what Zen's performance will be like - the "5 year old hardware" bit was based on what other people on the forum were saying about it not keeping up with Ivy Bridge and being competitive with Sandy Bridge. I personally don't have enough knowledge of the topic to guess whether or not it will be competitive with Sandy Bridge, Haswell, or something else entirely. But no one has jumped on the folks making those comparisons, so I figured they were somewhat on point and used that information to frame my question.

Zarathustra[H] & OFaceSIG - thanks for the responses.
 
  • Like
Reactions: NKD
like this
good lord. The damn processor isn't even out yet. and some people are fuckin out to call it a fail already. Guess what for what amd has right now if they can be competitive its not a fail, lol. Even if its slower than intel I doubt it will matter in every day computing or gaming. Lets wait and see to call it a fail. To nail it in a coffin already its just plain ignorant.

And for people that think turbo boost works at full speed when all cores are loaded I have a bridge to sell you. I have a 6700k it does not boost full speed when I load all threads, it only boosts if all cores are not stressed so it can push remaining cores to turbo staying within the target power envelope. If AOS is stressing all cores you can bet your ass the chip is not running at full turbo. I bet its around 3ghz max.

What amd needs is to target the top part at 3.5ghz base minimum. That will get them pretty decent bump.
 
I don't understand those that want this product to fail. If AMD goes under the hardware industry will be like Microsoft and Windows. Love Intel and it's decrees or eat a dick.
 
NKD said:
What amd needs is to target the top part at 3.5ghz base minimum. That will get them pretty decent bump.
Click to expand...

I already said that two pages ago. I said this would be a failure unless they can get it up to those speeds at launch. I'm hopeful they will correct this deficiency, and not make the mistake of launching too early (like Bulldozer). But I'm not going to pull any punches on the hardware as-it-stands now - it's weak, and is no faster than Piledriver on mixed integer-heavy code (the vast majority of processing loads).

Glad to hear you're adding such insightful commentary to this thread.
 
Last edited:
defaultluser said:
I already said that two pages ago. I said this would be a failure unless they can get it up to those speeds at launch. I'm hopeful they will correct this deficiency, and not make the mistake of launching too early (like Bulldozer). But I'm not going to pull any punches on the hardware as-it-stands now - it's weak, and is no faster than Piledriver on mixed integer-heavy code (the vast majority of processing loads).

Glad to hear you're adding such insightful commentary to this thread.
Click to expand...

It's an es. Who knows they tend to be clocked lower for testing. I am sure there are few floating around ranging from different clocks. I won't worry too much about that. Final silicon tends to be more targeted for speed when es
 
NKD said:
good lord. The damn processor isn't even out yet. and some people are fuckin out to call it a fail already. Guess what for what amd has right now if they can be competitive its not a fail, lol. Even if its slower than intel I doubt it will matter in every day computing or gaming. Lets wait and see to call it a fail. To nail it in a coffin already its just plain ignorant.

And for people that think turbo boost works at full speed when all cores are loaded I have a bridge to sell you. I have a 6700k it does not boost full speed when I load all threads, it only boosts if all cores are not stressed so it can push remaining cores to turbo staying within the target power envelope. If AOS is stressing all cores you can bet your ass the chip is not running at full turbo. I bet its around 3ghz max.

What amd needs is to target the top part at 3.5ghz base minimum. That will get them pretty decent bump.
Click to expand...


I think if you look at my posts I repeatedly say IF what we suspect is true, it will be a failure.

All of this is conjecture at this point. We have leaked results from an Engineering sample which may or may not be reflective of final silicone, or may even be a plant to throw people off, or misinformation from a competitor or fanboy to try to persuade people who are on the fence to buy Intel now and not wait for Zen. it could be anything. We just don't know.

All we can do is make predictions based on what we DO know which is:

1.) AMD has stated that they expect a 40% increase in IPC over (presumably) Kaveri.

2.) Global foundries process is optimized for low clock speeds, and apparently does not perform well above 3Ghz.


#1 is probably optimistic, because most companies, and especially AMD are typically optimistic about performance before launch, but if we assume it is true, and they are able to hit a base clock of 3.8Ghz and a turbo of 4.0Ghz on launch, that will give us a chip with single threaded performance (which is really what matters, at least to me) equivalent of a Sandy Bridge Core i5-2500k. This is based off of linear interpolation of frequencies and IPC's and existing benchmarks in Single Threaded Cinebench R11.5. This would be a little disappointing, but it would be usable in an enthusiast rig.

If #2 is true as well, and we assume max turbo is 3.2Ghz this is more troubling, as it would result in single threaded performance in line with a low end old Intel Pentium G850 or a Pentium G2030. If this is the case their big last hurrah of a Zen launch will be totally inadequate for enthusiasts, unless they do more mental gymnastics to try to convince themselves that the core count makes up for the poor single threaded performance, which it absolutely doesn't in 99% of workloads.

These are educated guesses at this point, and AMD could still surprise us with some positive revelations, but this is what we can project based on current knowledge.
 
It's a good policy to keep your expectations low with AMD. Their hype machine, whether fans overblowing things they say or their execs making outright lies about things (JF-AMD, every single exec at the fury launch), combined with AMDs history of grossly overstating strengths while completely omitting weaknesses (BD was touted as being high core count with high clocks, the fact that IPC was weaker than previous AMD chips was suppressed as much as possible, now zen touts high core count and IPC improvements, and they're not talking at all about clocks) should give pause for concern. Add to the fact they have a lack of a 2C/4T part for the low end, meaning they believe they need a 4C/8T part to compete with intel's low end is also extremely ominous, damning even. The Zen launch has clear AMD warning signs that not everything will be as good as the few things they have talked about might otherwise indicate, to not recognize that after all the previous disappointing AMD launches since BD is just being obtuse.
 
cybereality said:
I like having a mix of different hardware: Intel, AMD, Nvidia, components from different OEMs, just to play with and experiment.

Typically I try to switch between Intel/Nvidia and AMD/AMD for every other rig. Kind of broke that trend recently (since I needed a VR rig and AMD's current CPU offerings aren't enough) but I plan my next rig to be all AMD.

For me, it's just fun to test out different types of hardware and different brands.
Click to expand...

As long as Zen is > Sandy Bridge-E I'll be using it for my next rig. I have 6 sandy bridge cores now. 8 Ivy bridge cores level would still be an upgrade. Then I'll also have PCI-E 3, m.2 etc for an overall platform upgrade. That's all I really want. I want all around upgrade.
 
OFaceSIG said:
As long as Zen is > Sandy Bridge-E I'll be using it for my next rig. I have 6 sandy bridge cores now. 8 Ivy bridge cores level would still be an upgrade. Then I'll also have PCI-E 3, m.2 etc for an overall platform upgrade. That's all I really want. I want all around upgrade.
Click to expand...

I have a hexacore Sandy-E as well.

At stock clocks Ivy-E and Haswell-E are faster per core, but since I have no need for 8 cores, and my Sandy-E actually overclocks higher than Ivy-E and Haswell-E, the resulting absolute speed per core overclocked to overclocked actually winds up being higher on my Sandy-E.

So for me, Sandy-E is actually better than Ivy-E and Haswell-E. The extra clocks I get when overclocked result in more performance than the meager Ivy and Haswell-E IPC increases do.

Only downside is power consumption and heat. At 5ghz it really heats up my office.
 
I used to overclock my SB-E. Then one day for the hell of it I set it back to stock clocks. I played BF4 for years in 3 x 1080p eyefinity overclocked. Then after going back to stock for the hell of it and not running out of CPU and I haven't gone back to overclocked. No need for the extra heat and power consumption. So 8 stock Ivy-ish cores will be a nice upgrade for me. I play the BF series and Doom which all can use 8 cores or greater so I'm really looking forward to Zen.
 
-Dragon- said:
It's a good policy to keep your expectations low with AMD. Their hype machine, whether fans overblowing things they say or their execs making outright lies about things (JF-AMD, every single exec at the fury launch), combined with AMDs history of grossly overstating strengths while completely omitting weaknesses (BD was touted as being high core count with high clocks, the fact that IPC was weaker than previous AMD chips was suppressed as much as possible, now zen touts high core count and IPC improvements, and they're not talking at all about clocks) should give pause for concern. Add to the fact they have a lack of a 2C/4T part for the low end, meaning they believe they need a 4C/8T part to compete with intel's low end is also extremely ominous, damning even. The Zen launch has clear AMD warning signs that not everything will be as good as the few things they have talked about might otherwise indicate, to not recognize that after all the previous disappointing AMD launches since BD is just being obtuse.
Click to expand...
Just want to state 2 things. BD was supposed to clock in at 6-8Ghz which makes the IPC and Memory controller a bit more realistic as far as previous expectations.

But the bigger point I wanted to speak to is the 2C/4T part. They suck. No matter who makes it. They suck for gaming other than Farmville and candycrush. I have stated more than once that NO ONE WANTS an i3 (or in this scenario a 2C/4T part), although they may have no choice but to get one. In a prebuilt I can understand but to go out and purchase on for desktop gaming... (want so bad to say some things here but I don't like being that guy--- looking superior and what not).

Now add to that the consoles that every PC Masterrace poster gripes about holding us back and you can see an easy parallel to low core CPUs. 4 Core is the minimum, with or without HT. 2 Cores need to die for desktop usage even in prebuilts.
 
Like 90% of the mobile devices out there are 2C/4T parts, whether they have i3, i5, or i7 in front of them. The low power mobile market is the biggest market AMD needs to penetrate and one that is currently dominated by intel's 2C/4T parts because they're just fine for media consumption, web browsing, most office productivity work, and light gaming. I'm not just talking desktop usage here, I'm talking in the overall "does Zen have what it takes to save AMD as a company" scenario and if they can't strongly compete in the mobile market, which is really the only segment of PCs still growing, they're straight fucked. A 4C/8T being their low end seems to indicate to me that they won't be able to compete very well. AMD bragged on about the power savings of Polaris and we saw what garbage that was relative to nvidia, have they even said anything about Zen's power efficiency? Again with AMD you can tell what will suck ass by what they neglect to mention in marketing slides pre-launch.
 
Thanks for sharing Tyns , that is an amazing article. If those rumors are true, AMD may be making a real comeback.

That would be sick if they could compete with the Intel 8-core offerings but at a reasonable cost. October also is not that far away. GET HYPE!!!!
 
If the rumors of the 12CU apu are true that should give close to 460 performance with on the on chip graphics. Color me prepared to spend money :)
 
There's not a lot of new information aside from the power usage.
The eight core Zen CPU clocked at 2.8Ghz managed to considerably outperform AMD’s current eight core offering, the 4.0GhzFX 8350 despite a huge clock speed disparity. In fact, if we account for the clock speed disparity Zen would outperform the FX 8350 by 98% at the same clock speed. Offering nearly double the perforormance with the same number of cores, at the same clock speed. Which puts it squarely in the territory of Intel’s $1000 8 core i7 6900K and 5960X CPUs.

Read more: AMD Zen 8 Core 16 Thread & 4 Core 8 Thread CPUs Leaked - Impressive Performance At 95W & 65W
Click to expand...

Probably mostly a lot of hype, but I'm excited.
 
Last edited:
Darakian said:
tl;dr notes:
AMD is working on a new cpu called zen
it will come in 4 and 8 core versions
it'll be pretty ok to pretty good
Click to expand...

hahaha, okay. okay. I was hoping there would be some sort of new objective information.
 
KazeoHin said:
hahaha, okay. okay. I was hoping there would be some sort of new objective information.
Click to expand...
It's anything but, they seem to be claiming that Zen is so amazing it will compete with 8 core 140W broadwell chips at 95W, which from everything we know is patently ridiculous.
 
-Dragon- said:
Like 90% of the mobile devices out there are 2C/4T parts, whether they have i3, i5, or i7 in front of them. The low power mobile market is the biggest market AMD needs to penetrate and one that is currently dominated by intel's 2C/4T parts because they're just fine for media consumption, web browsing, most office productivity work, and light gaming. I'm not just talking desktop usage here, I'm talking in the overall "does Zen have what it takes to save AMD as a company" scenario and if they can't strongly compete in the mobile market, which is really the only segment of PCs still growing, they're straight fucked. A 4C/8T being their low end seems to indicate to me that they won't be able to compete very well. AMD bragged on about the power savings of Polaris and we saw what garbage that was relative to nvidia, have they even said anything about Zen's power efficiency? Again with AMD you can tell what will suck ass by what they neglect to mention in marketing slides pre-launch.
Click to expand...

I have my doubts about that if you look at the ARM eco system there plenty 4 small core solutions or others featuring more then 2 cores. The market you describe are people who buy their hardware not based on specs but rather looks and or price.
You forgetting the funnelling contra revenue funds Intel implies for the laptop market where AMD gets shafted , still today you find okay/nice laptops strangled by stupid specs (screen resolution 1366*766) same model Intel you can choose screen size sometimes even up to 4K and that is somewhat surprising since AMD does have a way better GPU.

Without Intel "paying" OEM there is no change for AMD to participate there even if their product was 10 times better ...
 
Pieter3dnow said:
I have my doubts about that if you look at the ARM eco system there plenty 4 small core solutions or others featuring more then 2 cores. The market you describe are people who buy their hardware not based on specs but rather looks and or price.
You forgetting the funnelling contra revenue funds Intel implies for the laptop market where AMD gets shafted , still today you find okay/nice laptops strangled by stupid specs (screen resolution 1366*766) same model Intel you can choose screen size sometimes even up to 4K and that is somewhat surprising since AMD does have a way better GPU.

Without Intel "paying" OEM there is no change for AMD to participate there even if their product was 10 times better ...
Click to expand...
IIRC Contra revenue only applied to lowest end Atoms, like mobile/tablet-tier Atoms, not craptop chips. Also, one of the best performing ARM chips is a dual core, last time i checked.
JustReason said:
Just want to state 2 things. BD was supposed to clock in at 6-8Ghz which makes the IPC and Memory controller a bit more realistic as far as previous expectations.

But the bigger point I wanted to speak to is the 2C/4T part. They suck. No matter who makes it. They suck for gaming other than Farmville and candycrush. I have stated more than once that NO ONE WANTS an i3 (or in this scenario a 2C/4T part), although they may have no choice but to get one. In a prebuilt I can understand but to go out and purchase on for desktop gaming... (want so bad to say some things here but I don't like being that guy--- looking superior and what not).

Now add to that the consoles that every PC Masterrace poster gripes about holding us back and you can see an easy parallel to low core CPUs. 4 Core is the minimum, with or without HT. 2 Cores need to die for desktop usage even in prebuilts.
Click to expand...
Reality is that i3s are actually competent for the cost in real world, less so in [H]'s userbase though. And yes, NO ONE WANTS an i5 either, in the same sense no one wants an i3.
 
-Dragon- said:
Like 90% of the mobile devices out there are 2C/4T parts, whether they have i3, i5, or i7 in front of them. The low power mobile market is the biggest market AMD needs to penetrate and one that is currently dominated by intel's 2C/4T parts because they're just fine for media consumption, web browsing, most office productivity work, and light gaming. I'm not just talking desktop usage here, I'm talking in the overall "does Zen have what it takes to save AMD as a company" scenario and if they can't strongly compete in the mobile market, which is really the only segment of PCs still growing, they're straight fucked. A 4C/8T being their low end seems to indicate to me that they won't be able to compete very well. AMD bragged on about the power savings of Polaris and we saw what garbage that was relative to nvidia, have they even said anything about Zen's power efficiency? Again with AMD you can tell what will suck ass by what they neglect to mention in marketing slides pre-launch.
Click to expand...
As far as Nvidia they have for years gimped their compute hence the lower power usage. So straight comparing pascal to Polaris is not as simple a comparison as most would like you to believe.

And as far as mobile low power devices, AMD has been working on their own ARM tech whish I think got pushed back to 2017 maybe further so I don't think they are as concerned in the x86 space for now. Besides they are focusing at least at first on desktop and server markets likely then following with mobile/low power after the process get hammered down.
 
JustReason said:
As far as Nvidia they have for years gimped their compute hence the lower power usage. So straight comparing pascal to Polaris is not as simple a comparison as most would like you to believe.
Click to expand...
You can say that if you want but most of us compare the two in games we play and not some arbitrary compute benchmarks that show AMD in a favorable light but isn't really any fun (AotS). So when and where it matters nvidia pascal takes a messy dump all over polaris. Shit the 480's power usage was so bad it lost PCI Express certification, which could do serious damage to their hopes of getting into OEM machines.
JustReason said:
And as far as mobile low power devices, AMD has been working on their own ARM tech whish I think got pushed back to 2017 maybe further so I don't think they are as concerned in the x86 space for now. Besides they are focusing at least at first on desktop and server markets likely then following with mobile/low power after the process get hammered down.
Click to expand...
ARM? LOL you think ARM can save AMD? I mean ARM devices are EXTREMELY low margin and most large consumers of ARM chips make their own already anyway so what exactly does AMD expect to do with ARM? And it's not like AMD ARM chips are going to displace intel low power x86 chips, so saying ARM is AMDs answer to mobile devices seems extremely short sighted (then again that seems to be AMD's fetish, play the long game while the floor crumbles under you, but enjoy your "ha I told you so" a few years after you're out of business and the market is finally ready for what you tried 10 years ago. As far as servers, servers are pretty much as picky about perf/W these days as mobile devices, again if Zen can't compete in the mobile market it can't compete in the server market because server operators care about lifetime expenses in cooling and power and shaving every penny they can off there. As far as focusing on the desktop market, the desktop market continues to collapse and is on the verge of just being a niche market for more powerful gaming machines, and business machines that are basically laptop parts in a small desktop case (see NUCs, etc...), but those don't care about power usage as much so I guess AMD might have a chance there? AMD's strategy seems to be a mess.
 
ARM? LOL you think ARM can save AMD? I mean ARM devices are EXTREMELY low margin and most large consumers of ARM chips make their own already anyway so what exactly does AMD expect to do with ARM? And it's not like AMD ARM chips are going to displace intel low power x86 chips, so saying ARM is AMDs answer to mobile devices seems extremely short sighted (then again that seems to be AMD's fetish, play the long game while the floor crumbles under you, but enjoy your "ha I told you so" a few years after you're out of business and the market is finally ready for what you tried 10 years ago. As far as servers, servers are pretty much as picky about perf/W these days as mobile devices, again if Zen can't compete in the mobile market it can't compete in the server market because server operators care about lifetime expenses in cooling and power and shaving every penny they can off there. As far as focusing on the desktop market, the desktop market continues to collapse and is on the verge of just being a niche market for more powerful gaming machines, and business machines that are basically laptop parts in a small desktop case (see NUCs, etc...), but those don't care about power usage as much so I guess AMD might have a chance there? AMD's strategy seems to be a mess.
Click to expand...

ARM is a potentially huge market, and one both AMD and Intel flat out missed. NVIDIA's actually way ahead on that front, since the entire auto-industry is moving toward NVIDIA's platform right now. For other non-specialized uses, ARM is VERY low margin, high competition. Frankly put, AMD isn't going to be making money there.
 
-Dragon- said:
ARM? LOL you think ARM can save AMD? I mean ARM devices are EXTREMELY low margin and most large consumers of ARM chips make their own already anyway so what exactly does AMD expect to do with ARM?
Click to expand...

Put them in servers. That is what K12 is all about as I recall.

ARM in the datacenter is probably going to be the next "big thing" as datacenters try to reduce their massive power use.

That, and AMD is used to low margin. Just look at all those console contracts :p
 
ARM got no chance in servers. ARM works in phones due to the cost factor. And K12 seems gone and all focus on Zen.

AMD and Intel never had a proper chance in phones, and its much more about the ecosystem and volume cost than performance. Not to mention both got a conflict of interest due to higher margin segments.

Just look what happened to Bobcat successors. Krishna and Wichita got cancelled.
 
lolfail9001 said:
IIRC Contra revenue only applied to lowest end Atoms, like mobile/tablet-tier Atoms, not craptop chips. Also, one of the best performing ARM chips is a dual core, last time i checked.

Reality is that i3s are actually competent for the cost in real world, less so in [H]'s userbase though. And yes, NO ONE WANTS an i5 either, in the same sense no one wants an i3.
Click to expand...


Well, in the all important "single threaded performance" measure even the lowest end i3 beats out the highest end AMD FX chip... And that is pretty sad.

I was really hoping AMD would make a strong comeback with Zen. I still hope it will happen, but the chances are looking more and more remote.
 
Looks like a giant step in the right direction for AMD. Been waiting forever to see improvements. The issue is, I have basically stopped gaming.
 
Zarathustra[H] said:
Put them in servers. That is what K12 is all about as I recall.

ARM in the datacenter is probably going to be the next "big thing" as datacenters try to reduce their massive power use.

That, and AMD is used to low margin. Just look at all those console contracts :p
Click to expand...

Nevermind there's no software for servers that runs ARM code, ARM performance is a fraction of X86/SPARC/POWER, and the minute ARM starts to add all those bits that add performance their power draw is going to be a lot like other processor architectures. ARM is the server is a pipe dream. There'sa reason why you haven't heard about K12 in about 18 months, and that's likely because AMD already killed it.
 
gamerk2 said:
Nevermind there's no software for servers that runs ARM code, ARM performance is a fraction of X86/SPARC/POWER, and the minute ARM starts to add all those bits that add performance their power draw is going to be a lot like other processor architectures. ARM is the server is a pipe dream. There'sa reason why you haven't heard about K12 in about 18 months, and that's likely because AMD already killed it.
Click to expand...


Nah. ARM is more efficient at performing the same tasks than X86 is. X86 is plagued by legacy compatibility issues that inevitably make its performance per watt lower.

You are wrong about the software availability too. Open source has already embraced ARM. Versions of many common server tools have already been ported. Having actual ARM server hardware will only accelerate this trend.

I think K12 has lots of potential to be a market disruptor, but its more of a medium to long term thing than it is a overnight thing. I very much doubt K12 is dead. They dd juggle aroudn the priorities to move Zen up in the timeline, but I feel fairly certain K12 will be the next to come down the pike.
 
Zarathustra[H] said:
Nah. ARM is more efficient at performing the same tasks than X86 is. X86 is plagued by legacy compatibility issues that inevitably make its performance per watt lower.
Click to expand...

PPW is actually a lot worse for ARM, by virtue of it being a good 5x slower at most workloads. And yes, I can provide benchmarks. ARMs problem is the minute they start adding the hardware to add performance, their low-power advantage vanishes. Basically the same thing that happened to the POWER arch after it split from PPC.

Secondly, please stop it with OSS; OSS doesn't, and never will drive the business community.

Intel will be providing foundry services to people using ARM processors. So Intel won't be flat out missing that market.
Click to expand...

Intel is producing for other people, not producing ARM designs themselves. Decent chunk of change, but nothing major.
 
gamerk2 said:
PPW is actually a lot worse for ARM, by virtue of it being a good 5x slower at most workloads. And yes, I can provide benchmarks. ARMs problem is the minute they start adding the hardware to add performance, their low-power advantage vanishes. Basically the same thing that happened to the POWER arch after it split from PPC.
Click to expand...

This is not my area of expertise, but it does not jive with my personal experiences. The only way Intel was able to compete with their x86 technology in light phones and tablets (Merrifield and Moorefield) was by using their amazing process expertise to stay a few steps ahead in process node. The likes of qualcomm produced higher performance at lower power on larger process nodes than Intel's offerings here. If some of this power use success can be translated to the server farm / data center market, there would be billions in power savings.


gamerk2 said:
Secondly, please stop it with OSS; OSS doesn't, and never will drive the business community.
Click to expand...

Not your average corporate IT departments servers no. They care about Exchange servers, and use shitty windows server products. Massive data centers on the other hand? You bet. Do you really think that Googles server farms are running Windows? lol. They'd laugh you out of their building if you suggested that. Large scale datacenters are mostly custom in-house developed software running on top of an open source base.

The massive server farm is the target of K12. Google alone used 260 million watts of power for its data centers in 2014, and it's only gone up in the last 2 years.


gamerk2 said:
Intel is producing for other people, not producing ARM designs themselves. Decent chunk of change, but nothing major.
Click to expand...

I've seen that too. I agree.
 
gamerk2 said:
PPW is actually a lot worse for ARM, by virtue of it being a good 5x slower at most workloads. And yes, I can provide benchmarks. ARMs problem is the minute they start adding the hardware to add performance, their low-power advantage vanishes. Basically the same thing that happened to the POWER arch after it split from PPC.

Secondly, please stop it with OSS; OSS doesn't, and never will drive the business community.



Intel is producing for other people, not producing ARM designs themselves. Decent chunk of change, but nothing major.
Click to expand...

Right, ARM has a low-power advantage (even compared to Atom) because of the low-complexity decoders. On mobile, it's a LARGE fraction of total power. I/O costs are optimized to the absolute minimum the platform can handle, so the decoder difference can stand out.

But when you need higher performance in a single server, you need I/O to strap cores together (L3 cache on AMD K12), or more powerful cores. That I/O costs the same amount of power no matter what platform you're using, and the execution resources use the same amount of power no matter what architecture you're using. Those costs SKYROCKET as you add more cores, and overwhelm the decode power.

Just look at that 32w TDP for eight A57cores. Those A57 cores use under 1w each on your average mobile chip, but all that I/O added to make them actually DO SOMETHING increases power by 4x! That power would be identical from any other architecture.

The execution and I/O power vastly outweigh decoder power on bigger more powerful chips. So, much like PowerPC, ARM has no inherent advantage in the server world: it's just another architecture. RISC is no better than CISC.
 
Last edited:
Zarathustra[H] said:
This is not my area of expertise, but it does not jive with my personal experiences. The only way Intel was able to compete with their x86 technology in light phones and tablets (Merrifield and Moorefield) was by using their amazing process expertise to stay a few steps ahead in process node. The likes of qualcomm produced higher performance at lower power on larger process nodes than Intel's offerings here. If some of this power use success can be translated to the server farm / data center market, there would be billions in power savings.
Click to expand...

You cant use Atom to compare with due to internal competition and priorities. Apollo Lake for example is 50% faster than the previous Atom when resources are given somewhat free. If you want to, use Core M. But then you are more on a tablet size compare.

There is also nothing at all that benefits ARM in terms of perf/watt. x86 isn't an issue there.

Zarathustra[H] said:
Not your average corporate IT departments servers no. They care about Exchange servers, and use shitty windows server products. Massive data centers on the other hand? You bet. Do you really think that Googles server farms are running Windows? lol. They'd laugh you out of their building if you suggested that. Large scale datacenters are mostly custom in-house developed software running on top of an open source base.

The massive server farm is the target of K12. Google alone used 260 million watts of power for its data centers in 2014, and it's only gone up in the last 2 years.
Click to expand...

K12 would have to be the most exceptional ARM chip ever to even somewhat get some interest from the big 7. And we already seen what happens to ARM when it moves out of the phone. It becomes a huge powerhog. Zen is the product AMD is using here to try and get sales. And K12, is it not cancelled? AMDs ARM dreams died with Seattle.
 
You must log in or register to reply here.
Back
Top