Why is AMD losing the next gen race to Nvidia?

There are lots of games where the 390X outperforms the 480, by 5 ~ 20%, despite the 390X being 3 years old and having the same flop count on an older architecture.

This is just the latest example.
 
So now we're pretending the R390X doesn't have ~15% higher compute throughput, twice the ROP count and theoretical fillrate, 50% higher memory bandwidth, is 90% larger and consumes twice the power of a RX 480?
Ok, whatever fits the circlejerk's narrative. Yup, zero achievements between Polaris 10 and Hawaii.


As for the ridiculous thread title, last I heard AMD is confirmed to have won the semi-custom SoC design for both the Scorpio and Plus. Unless the pc market all of a sudden became larger than both consoles together, I'd say AMD already won the next-gen race.

But what do I know? It's not like assuring that all meaningful developers are inevitably working for your architecture is a good thing because let's face it: this thread is all about stating otherwise.

And this thread dedicated to saying that every. single. DX12. benchmark that appears and shows better results for GCN GPUs (all of them, lol) needs to have some serious faults in order to not be considered as "valid".

And this thread which is about saying how limited Fiji GPUs are because they have 4GB of RAM even though they consistently come out on top of anything similarly priced from the competition in the latest benchmarks.



It's ridiculous how 5/6 members of the forum have turned the "architecture" forums into a huge anti-AMD circlejerk and continue to do so unattended.
 
Last edited by a moderator:
So now we're pretending the R390X doesn't have ~15% higher compute throughput, twice the ROP count and theoretical fillrate, 50% higher memory bandwidth, is 90% larger and consumes twice the power of a RX 480?

(64×44×2×1050)/(64×36×2×1266)=1,013691416535018430753027909426

I'd say at least this number's in the same ballpark.

So now we're pretending the R390X doesn't have ~15% higher compute throughput, twice the ROP count and theoretical fillrate, 50% higher memory bandwidth, is 90% larger and consumes twice the power of a RX 480?
Ok, whatever fits the circlejerk's narrative. Yup, zero achievements between Polaris 10 and Hawaii.
You seem to have forgotten to figure in the 20% clock advantage from RX 480, the improved performance per CU (AMD says 15%, Press Deck p11) and the Delta C Compression (AMD claims ~35% more effective memory bandwith, cannot tell exactly from the deck, p 12), that Hawaii lacks.

But other than that …
 
Last edited:
So now we're pretending the R390X doesn't have ~15% higher compute throughput, twice the ROP count and theoretical fillrate, 50% higher memory bandwidth,, is 90% larger and consumes twice the power of a RX 480?

You are embarrassing yourself.

The question was whether the marked drop in performance / flop was due to the reduction in ROPs and memory bandwidth. That was the actual question. To respond with "So now we're pretending the R390X doesn't have ...twice the ROP count and theoretical fillrate, 50% higher memory bandwidth..." is simply utterly bizarre. It makes no sense as a response.

You're wrong about compute throughput btw. And no-one mentioned power.

Ok, whatever fits the circlejerk's narrative. Yup, zero achievements between Polaris 10 and Hawaii.

Nobody claimed this. You are pretending this happened, then getting angry about it.

It's ridiculous how 5/6 members of the forum have turned the "architecture" forums into a huge anti-AMD circlejerk and continue to do so unattended.

Which is strange, because I'm directly asking an architecture based question, and you are responding with an unappealing mixture of rage and tears.
 
Sure, even though you answered the question yourself by acknowledging the huge difference in fillrate on the post where you made the "technical question", you certainly did not want this post to present a new negative thingie about AMD cards. Unlike pretty much every post in this thread.

As if all the "likes" from the BFFs in your posts weren't double-confirming that you were doing just that.
 
did ya get up on the wrong side of the bed this morning? Just asking cause you are really getting pretty personal and actively avoiding a valid question he presented.
 
Sure, even though you answered the question yourself by acknowledging the huge difference in fillrate on the post where you made the "technical question", you certainly did not want this post to present a new negative thingie about AMD cards. Unlike pretty much every post in this thread.

I don't know that it's the cause. It's the most obvious difference that I can see, looking at tables of specs on Wikipedia, but I don't know that it's the reason hence my question.

And a drop in the utilisation of CUs for a 4th gen part compared to a 2nd gen part is worth noting and talking about. AMD are striving for the opposite, hence their work on the Geometry Processors.

As if all the "likes" from the BFFs in your posts weren't double-confirming that you were doing just that.

What does this even mean?

And if you can't talk about a drop in CU utilisation what's the point in this place? It's something that's happened. There will be a reason for it.

Those underutilised CUs are burning power and leaving performance on the table. That can't be desirable.
 
As for the ridiculous thread title, last I heard AMD is confirmed to have won the semi-custom SoC design for both the Scorpio and Plus. Unless the pc market all of a sudden became larger than both consoles together, I'd say AMD already won the next-gen race.

How much money does AMD do with consoles again? Judging by their income figures, LESS than nVIDIA does in the PC market. A graphics card can be way more expensive than a PS4 / XBO alone and AMD only supplies the APU, plus Sony is making a profit on every console sold since the beginning. Oh yes, the supposed superiority of AMD in the PC Gaming market due to ports from consoles. Nope, PS4 / XBO launched nearly 3 years ago and those effects have yet to materialise. So how is AMD winning again??? Looking at your attitude, it seems to be you seeing reality with red tinted glasses, not some green conspiracy. You are the only person here constantly attacking people, instead of contributing with anything meaningless for the debate.
 
I can't believe im waisting my Saturday on you, man im a glutton for pain.

You yourself stated it, you don't know where Zen performance will fall, it will be better to Bulldozer, but you don't know where. You want to see the counter argument to that? all you have to do is read what I posted. You took the positive, I took the negative.
Thats not even remotely the same, an actual negative view would be to make some kind of INFORMED
evaluation, things like:
1- I think the one less AGU will be an issue ( looks like the stack engine makes this a non issue) but typical server workloads are unlikely to care
2- I think the lack of AVX 256 throughput(512/256LS) vs > haswell will really hurt the Zen for the Server Market
3- I can't think of a third point but etc.

Even if Zen was a failure you wouldn't be right, you'd just be a broken clock. But as you continue to ignore that we already know enough about Zen to a level of details to produce an article to the level of detail that you would find at somewhere like RWT. At this point your view is formed my emotion not evaluation of known Evidence. As i have presented evidence you have shifted the goal posts without acknowledgement, first it was 8 cores Zen vs 4 core intel, now look at what you are arguing.

So go ahead prove me wrong, provide us a detailed back down of things to be concerned about With Zen's micro architecture and platform architecture. ( please i would actually like this, discussing counter positions if how we all learn and grow!).


If Zen comes out and doesn't match Xeon's core for core performance, no person in their right mind would put Zen CPU's in a server or HPC even if they cost 25% less, because if they do, Intel will try to match the price on performance. So Intel's faster chips will still cost more but there will be equilibrium.
You keep saying this like its backed with some level of fact, it isn't, enterprises , cloud and web 2.0 buy things like Xeon-D and middle of the road cpu's like E5-2670's ( seriously go look at Azure , AWS etc). You keep ignoring what matters to servers at this point you are nothing but a TROLL, So actually backup your words that even a 20% Performance deficit matters. I've been in my current job for the last 6 years in that time i have worked with 50-60 customers largely on Datacentre Infrastructure I have rarely seen a Hypervisor that was CPU bottlenecked. Go look at current market directions in the enterprise space things like NFV, NSX network services models etc. These functions are high CPU low memory usage, if everyone was so smashed for CPU time would they be pushing these services that perform so poorly on x86 vs ASIC/FPGA to the x86 edge?

Then you have completely ignored the potential for Zen to clock higher at the same power etc. You have to consider approach, intels large die approach allows for better cache coherency across a large amount of cores. Intels process advantage allows the manufacture of these large dies at highest possible costs, lowest possible defects.

Now look at AMD's approach, small dies interconnected together, reduces the process deficit across the board, better yields, clock and power compared to making a 32core chip on LLP . But they will suffer increased latency when they go across a MCM interconnect its just physics and total die size will be a bit bigger.

But guess what, the model for the last 5 years has been smaller number of Vcpu VM's and just spin up more, hybrid cloud, cloud busting etc drives this for internal applications as well as external applications. The model the vast majority of cloud and enterprise use eliminates that interconnect disadvantage, the last AMD server farm chip i saw in production was mangy-cours and MCM didn't matter in VM's then and it wont matter now.

On the business side its not so simple as you make out, its not 2006 anymore and all other markets are basically shrinking for Intel they just wont go around slashing Server Revenue. Theoretically if AMD can produce enough chips to take 10% market share, the SKU is 10% behind in per core performance but either is 25% cheaper or 25% more cores and 25% more memory capacity and throughput at the same price point what does Intel do. The way they make the most money is to do nothing, if the make price comparable they lose 10% GP, if they get aggressive they lose even more. That will hit the bottom line more and thus have a greater effect on market cap and they will have to explain that to angry share holders.

Of course Intel can adjust there product development to match or exceed but that will take years to make it to market.
 
Continued last post

You are saying what Intel does doesn't matter (and no I'm not talking about 10% IPC, I'm talking about 10% performance overall and the difference of Zen to Kaby will likely be even more than that), I think not, they have more pull, more influence, more leverage, better products for now and most likely in the future with Zen too, 99.2% of server infrastructure is based on Intel right now too! you totally dismissed all of this. Just because Zen performance mitigates the gap they have now, doesn't mean Jake in the server market, you know that!

First of all when is Kaby lake be making it to a large core count Server CPU, Broadwell just got released in June Haswell got released Sept 2014, SO when is SKYLAKE EX,EZ getting released let alone something like Kaby lake ( unlikely to ever make it to an EX,EZ anyway.)

Second In my very first post on this I said the Area where AMD would find it hardest to complete is the exact area where Kaby lake will be strongest, which is the high clocking consumer desktop 4 core parts.

In the laptop space AMD has a far better chance because the intel process advantage will be less and AMD Superior GPU will give an advantage, up until Zen AMD has been so far behind on perf/power that GPU advantage hasn't mattered.

Third, I never said AMD would take all of intel market share over night, of course its going to take time to take market share, But the market is very large and there is plenty of opportunity for AMD with a competitive product, any server sales AMD make is Revenue that they previously didn't have.

What you are doing in your post, is you took out all the business aspects of the server market. Things don't change on a dime. If you look at the original Opterons, how long did it take them to make inroads? And that was with a sizable performance advantage. I don't understand why you can't understand if Zen doesn't match Xeon's core for core, what would the use be for current Xeon servers to be upgraded to Zen? Power envelopes looks to be similar, All it will come down to is cost, and that cost is meaningless because Intel will price appropriately to what AMD prices.

I did no such thing, I never once made a prediction of sales or revenue all i talked about it what actually matters to the average large server farm environment. I also never said someone running Haswell or broadwell would upgrade to Zen, but there lots of Servers older then that, . Look at when Facebook started selling E5-2670's on ebay for example, there will still be millions upon millions of those Sandy/Ivy bridge CPU's in production around the world mid next year. When it comes time to refresh haswell and broadwell AMD (if they are still around) wont be competing with Zen, it will be at a minimum 48core Zen+ hopefully they scale memory controllers at least linearly to match.
So what you are left with is people buying new servers. Well guess what Intel is still a better option, with the volume they have been purchasing before, I'm sure they were getting discounts before.....

Things don't change that fast.

All the talk about ARM server chips, AMD gaining back server marketshare, if the performance isn't there to Intel's Xeon's, its all a pipe dream.
And yet you continue to ignore whats important, Why do companies buy massively expensive 4 or 8 proc servers is it for all the single thread performance, no it fucking isn't. Its it for aggregate throughput no it fucking isn't, you just spin up another docker image in real time and add it to the Load balance pool for that. It is for the massive memory throughput and massive memory capacity for things like HEC and HANA. Its the same all the way down the enterprise stack, its amazing you continue to blindly ignore, not counteract , no counter points just ignore the massive importance of memory capacity and throughput.

Why is 10% more single thread performance more important then 25% more memory throughput and capacity, AGAIN I ASK YOU BE SPECIFIC!
 
Continue last last post
This is a strawman; his logic does not extend in this way unless you're purposefully forcing the subject.


I'm the Enterprise Architect for Infrastructure [compute, storage, network, virtualization and cloud] for a US-based Fortune 250 organization, and "Retail" price means exactly jack. I can easily explain to you why processor cost is the least of my concerns, let's use our newest eCommerce virual environment to explain:
completely agree that is why i used percentages

Our entirely new hardware stack for the "public" version of our eCommerce platform consists of:
Two HP C7000 Chassis, each with:
Qty 4: BLc FlexFabric 20/40 F8 Module
Qty 6: 2650W Platinum PSUs
Qty 1: Six "Active Cool" 200 FIO Fan Brackets
Qty 1: BLc Intelligent Power FIO
Qty 1: 3" LCD Enclosure
Three years of prepaid support: 24/7 with 4hr MTTR
In each C7000 Chas sis, there are ten BL460g9 blades, each with:
Qty 2: Intel E5-2698v3
Qty 16: 32GB 2Rx4 DDR4-2133 DIMMS
Qty 2: FlexFabric 20Gb 650FLB Adapter
Qty 1: SmartArray P244br/1G Controller
See Razor middle of the Road CPU's , middle core count, more expensive 32gb Dimms because you cant hit 512gb memory with 16gb dimms with 12 memory channels.
You know the most expensive part of these twenty blades and two chassis? The 10TB of memory, that's what. Want to guess the second most expensive part of these twenty blades and two chassis? All the operating system licensing (VMware, RHEL, Windows.) The line-item cost of those processors was significantly less than half of the MSRP printed on the Intel ARK site, and if their cost was ZERO, the total affect on the price of the new cluster is still less than a 5% difference. I actually saved money by purchasing fewer blades and mitigating with higher cores-per-socket XEONs.

You have missed a step in your logic, if you migrated to fewer PROC's you reduced your VMware, Windows etc licencing costs. But they are also Opex costs so a different budget geranted it can be a major consideration but unless you shift app/middleware etc stake your going to pay for anyway.

Also consider what CPU cost is already factored into the blade/1ru chassis cost, I have access to pricing for a particular vendor we do full maintenance and support for, the costs are per line item manufacture cost + a tiny little bit. If i looked at the cost of a blade or a 1ru chassis (without PSU's) the CPU's you have picked are about 2.5 times(each) the price factores into hardware cost more then you think.

But then also consider with 16 memory channels you would have been able to reduce your memory cost by around 20% by going 16*16 dimms to reach 512gb a host.

There will be people who care about individual 1P socket/core prices. For those of us who do high-density virtualization with carrier-class equipment, your price mninutae means nothing. When I'm stacking VMs, I will go as dense and as fast as I can reasonably buy. The only reason I didn't buy the 18's is because they aren't available in the half-height blade form factor I wanted.
And across your data centre vm farms what is closest to max capacity IO, memory or CPU. You would be very atypical if CPU came ahead of either memory or IO.

Yep pretty much what my friend said who is the IT manager server side, for all of the northeast for Cargill, not as descriptive, but he won't switch off of Intel cpu's unless AMD can offer much better performance then they are currently seeing.
And ask him what is VM vs physical server mix is ( i was shocked just how low it is in many US org's vs Australian org's, in Australia its been and upward trend from around 70% when esx 3.5 dropped in 2008 ) and how much he could reduce his server footprint if he had more memory capacity at comparable performance.

Razor You have continued to Ignore Zen will have more core's and more memory even if it has less performance. My assumption has been that Zen will be close enough single thread performance, i might be wrong, but nothing we have seen so far says that it wont be.

And I am now Done on this topic!
 
Last edited:
Continue last last post

completely agree that is why i used percentages


See Razor middle of the Road CPU's , middle core count, more expensive 32gb Dimms because you cant hit 512gb memory with 16gb dimms with 12 memory channels.


You have missed a step in your logic, if you migrated to fewer PROC's you reduced your VMware, Windows etc licencing costs. But they are also Opex costs so a different budget geranted it can be a major consideration but unless you shift app/middleware etc stake your going to pay for anyway.

Also consider what CPU cost is already factored into the blade/1ru chassis cost, I have access to pricing for a particular vendor we do full maintenance and support for, the costs are per line item manufacture cost + a tiny little bit. If i looked at the cost of a blade or a 1ru chassis (without PSU's) the CPU's you have picked are about 2.5 times(each) the price factores into hardware cost more then you think.

But then also consider with 16 memory channels you would have been able to reduce your memory cost by around 20% by going 16*16 dimms to reach 512gb a host.


And across your data centre vm farms what is closest to max capacity IO, memory or CPU. You would be very atypical if CPU came ahead of either memory or IO.


And ask him what is VM vs physical server mix is ( i was shocked just how low it is in many US org's vs Australian org's, in Australia its been and upward trend from around 70% when esx 3.5 dropped in 2008 ) and how much he could reduce his server footprint if he had more memory capacity at comparable performance.

Razor You have continued to Ignore Zen will have more core's and more memory even if it has less performance. My assumption has been that Zen will be close enough single thread performance, i might be wrong, but nothing we have seen so far says that it wont be.

And I am now Done on this topic!


Well he is sitting here having a beer with me now as we are about to head out. 25% to 75% mix, virtual machines are not that useful for them. They only use virtual machines to collate some inventory data.

He says its not a question of performance if they are close. Its not even a question of cost. This is a billions of dollar a year company, private company, the largest beef/chicken/pork (think they dropped pork recently though) processor and distributor in the WORLD. for them even 10% difference in over all performance per chip doesn't matter they will just get another stack and be done with it. These guys aren't crunching numbers.

Now If this was an investment bank, I know for a fact that if 10% more performance per chip will be something they are looking at, they will pick up AMD in a heart beat if that was the case, but they need that performance, because that time then need to save on some of the financial models can save millions of dollars if not more. RAM needs are there too, but they need processor performance as well.

He says more memory is good but it comes down to what they already have, and what contracts they have, he won't tell me the details but you get the picture? He says to break those contracts or get better a better deal, there has to be a great deal of pressure on Intel's capabilities from the competing company (AMD) to even look at them. They didn't even RFP Intel or AMD for the current contracts, there was no negotiations involved. They just asked Intel, give us your best price through Dell and HP, and that was it. They don't care about the server footprint either...... they have enough space and more to expand two fold if need be.

Bottom line is its not a question of price. Its not a question of performance if they are close, the only way AMD can get their business is if they can show Cargill they can improve operational efficiency with their products and it has to be enough that they can't ignore.

Any case have a good weekend! Drink a nice cold one down for me :)
 
Last edited:
I can understand why companies do that, AMD hasn't been competitive in a long time; But with Zen looking like it will be very strong competition it would be pure incompetence to not use that as leverage to at the very least, force Intel to lower prices.
 
Yeah Intel will match anything AMD will give as price, I mentioned that already. Intel isn't selling these chips at MSRP to these companies, they are already heavily discounted, and as I mentioned before AMD can't go into a price war with Intel. So how do you think AMD will be able to beat Intel on price per performance?
 
Yeah Intel will match anything AMD will give as price, I mentioned that already. Intel isn't selling these chips at MSRP to these companies, they are already heavily discounted, and as I mentioned before AMD can't go into a price war with Intel. So how do you think AMD will be able to beat Intel on price per performance?
Your friend answered for you, small performance differences doesn't matter and he wants more memory but is bound to contractual obligations.

So a contract from Dell/HP with more RAM would do, but AMD needs to fight an uphill battle against corporate inertia. Nothing new.
 
yep at the end we won't see the server market as a upswing for AMD for probably a year and that is if they are competitive price and performance. This was probably the same reason we didn't see them make much market penetration back with the original Opterons, that and Intel's snide tactics. And it was about a full year before their marketshare started going up.

Intel knows this, they know if AMD can get a foot hold in the server market its much harder to get any lost marketshare back, its going to be a hard fought battle and they are not going to just give it up if their products are just as capable.
 
With Opteron AMD had the distinct advantage that Intel was playing silly buggers - trying to keep x86 as 32-bit whilst pushing anyone who wanted 64-bit towards Itanium. Intel was refusing to give a lot of customers what they really wanted (64-bit x86), and this was an open door for AMD. Even then, as has been said, it took a while for Opteron to really gain penetration.

I don't think that open door exists any more - Itanium is dead and Intel are on their A-game with x86. Is there anything that customers are clamouring for that Intel is refusing to deliver for "political" reasons?
 
yep at the end we won't see the server market as a upswing for AMD for probably a year and that is if they are competitive price and performance. This was probably the same reason we didn't see them make much market penetration back with the original Opterons, that and Intel's snide tactics. And it was about a full year before their marketshare started going up.

Intel knows this, they know if AMD can get a foot hold in the server market its much harder to get any lost marketshare back, its going to be a hard fought battle and they are not going to just give it up if their products are just as capable.

Its not aimed on you, but this make me push in a more large reflexion: I dont know why everyone take it this way: When AMD said they will, want have some GPU marketshare back, peoples seems think they plan to get 70% marketshare in the next 6 months.

Same goes for the server,or even consumers CPU markets, they dont plan to have 60% of the marketshare back in 2017.
Specially on the professional, entreprise front, the only way will be that they are alone in a specific sector who suddenly become the important one. ( and this implied that other company will seat back and do nothing ).

Thoses things take time... and as you said, the inertia of the professional markets ( whatever it is workstations, image industry ( VFX, 3D etc ), sever, cloud is incredible, without saying it is like that for good reasons.
 
With Opteron AMD had the distinct advantage that Intel was playing silly buggers - trying to keep x86 as 32-bit whilst pushing anyone who wanted 64-bit towards Itanium. Intel was refusing to give a lot of customers what they really wanted (64-bit x86), and this was an open door for AMD. Even then, as has been said, it took a while for Opteron to really gain penetration.

I don't think that open door exists any more - Itanium is dead and Intel are on their A-game with x86. Is there anything that customers are clamouring for that Intel is refusing to deliver for "political" reasons?
Intel sadly are still playing silly buggers to a certain extent by trying to claw development back with the failed (in terms of its complications and missed schedule) Broadwell, if they wanted they could release the enthusiast Skylake probably now and the Xeon HPC centric CPUs end of the year/early 2017, but no we all have to put up with Broadwell instead (context enthusiast and Xeons).
I think the release of Zen will possibly expedite Skylake Xeon, but if AMD do not release the higher core models early next year and start to creep closer to summer then unfortunately they start to fall under the shadow of Skylake Xeons rather than Broadwell for comparison anyway from businesses.
For AMD it is a bit like Polaris, they just missed the perfect launch window that would had given them massive opportunities with less of a fight, which is gutting as I feel Intel are deliberately sandbagging and deserve to be taken by surprise.
Cheers
 
Back
Top