" As reported by our sister website Laptop Mag, the HP Envy x2 with Raven Ridge was a repurposed chassis from HP’s catalogue, rather than something hyper optimized."
I think this sentence is missing a link to a Laptop Mag article.
What psxotaku said ... I tend toward low power these days so the embedded 12-watt AMD V1605B w/8 CUs floats my boat. (I hate being pragmatic and responsible.)
RR-2018 brings Vega 11 to the table along with the "Re-spin of Zen." The Internets opine (for years!) that Vega 11 has slightly higher performance at vastly better performance/Watt. Maybe Chipzilla helped to gate those Vega M CCN cores with their "Dynamic Power Sharing." HA!
In the past AMD has typically boosted clocks and efficiencies with new steppings, with no reason to think a second shot would be any different this time around.
I'm curious where the improvements for Raven Ridge 2018 are, I'd assume binning (probably for notebooks).
Zen 2 is taped out for Rome (server chips). No word on Matisse (Ryzen update), Picasso (Raven Ridge update), or Castle Peak (sounds like a threadripper that would almost certainly use either Rome or Matisse chips similar to current threadrippers). As far as I can tell, Rome was "always" (meaning they switched by the time AMD had to worry about how the process effected physical layout of transistors) while the rest may have had work done assuming that GF would manufacture them. Both rework and scheduling time on TSMC fabs may delay things.
Vega 2 won't have many customers, being a professional card in the shadow of nvidia. Machine Learning developers may certainly be interested as well as the traditional CAD users. Vega 2 is also almost certainly the GPU of picasso, so I certainly hope Anandtech gives plenty of coverage to Vega 2 (ignoring how few readers are likely to buy it) and focus more on how well AMD is working with 7nm TSMC and how well vega 2 will work in picasso.
I'm surprised that Navi didn't get a mention. I'd hardly bet on a 2019 debut (unless inside a PS5) but I'd expect it to bring competition to the GPU market. I'd at least expect it to force a 7nm 21x0 line from nvidia sooner than 2080(ti) buyers will expect.
Things to ask AMD: Is Vega 2 sampling? It should have sampled months ago. Are we going to see it before Jan 1,2019?
Are matisse, picasso, and navi taped out (I'm not so hopeful about navi). AMD typically takes a year from (official) tapeout to launch, so we need to see these things before the year (2018) is out if we expect to see them in 2019 (Q4, don't expect them any earlier). Note that "tapeout" might not be such an important thing for matisse (and to a lesser extent picasso) as they are probably backporting any issues found in Rome and won't officially "tapeout" until the Rome design is finalized.
Vega 7nm has been sampling for some time now. A figure of a 40 percent boost to performance over Vega 64 was put out there several months ago. The problem is that Vega is very much tied to HBM2 memory, and due to higher prices/lower availability, AMD is keeping the volume low for that very reason. If HBM2 memory were readily available and the price wasn't as high, then AMD would probably have Vega 7nm out there for gamers at the end of this year.
Ryzen third generation should also be in good shape at this point, but getting yields up at TSMC may be why AMD is not talking about it(to avoid wild speculation that would only hurt the company). Until AMD has at least 300,000 units ready, I don't think AMD wants to put any timeline on when it will officially launch the product. AMD may also be waiting to see what speed Intel i9 chips can actually run at so that AMD can properly price the third generation chips appropriately. If AMD can hit 5GHz on air and Intel can hit 5.3GHz on air, if AMD is able to improve IPC by 10-15 percent, that is the sort of thing AMD will want a big launch event to cover. If AMD can get third generation to the same 5.3GHz on air, that is the sort of thing that AMD management will want to really push.
So, decisions are being made, but until Intel actually ships the new i9 chips, AMD will probably stay silent.
Why is that when AMD makes slight changes they call it progress but when Intel makes the changes they call it Marketing or rebranded - they slides here clearly look like Marketing.
I can explain HStewert. The last few years Intel was not making the best product it can make but AMD is giving it all they have. It's simple as that and everyone that follows tech knows Intel has not put its best product it can make out in the consumer space. In servers sure, but not in the consumer / HEDT segment.
While Intel has surely procrastinated technologically (largely because they had no competition) one reason that their architecture has been basically frozen since Skylake is that they decided to tie their next architecture (Ice Lake) to their next node, and we all know their problems with their 10nm node. The second generation (10nm+) of their next node actually, since they originally planned to use 10nm for Cannon Lake, which is basically the Skylake design at 10nm plus an AVX-512 block.
Cannon Lake looks like it will remain a beta node, perhaps with only the sole Core-i3 15W CPU with disabled iGPU that was released in low volume. Ice Lake is planned for HVM in mid 2019 and wide release (from top to bottom consumer) in late 2019. On the other hand it appears that they do not have too much confidence in Ice Lake and its successor Tiger Lake, which is why they hired Jim Keller to design Ocean Cove, which is set to be their first post-Core architecture since the original Core CPUs were released in 2006.
The first Ocean Cove CPUs are expected to be released in 2021 at the earliest (~15 years after the original Core was released), either at a late 10nm node or an early 7nm node.
"...which is why they hired Jim Keller to design Ocean Cove, which is set to be their first post-Core architecture since the original Core CPUs were released in 2006." And post-Lake, of course, since it is surely no accident, that they are moving from lakes to coves and oceans.
Everyone has opinions - and I disagree with your opinion - I just believe some people have limited focus on gaming market and don't see the entire picture.
The issue that I see is that Intel could have had 5.3GHz chips out for several years now, but used their horrible TIM to artificially limit how fast the chips could be clocked to. You have had people doing a delid on their Intel chips for this reason, to get the most they can out of their processors.
So, if IPC did not go up all that much, and true potential for clock speeds has not gone up by all that much(as I said, a delid fixed the problems with the bad TIM that Intel had been using), then, what real improvement has there been on the Intel side? If AMD is able to hit 5GHz on air with Ryzen 3rd generation, even if Intel still holds a lead for potential, that would be seen as a huge jump forward from AMD.
Until Ryzen, Intel was still pushing 2 and 4 core chips on the desktop, and many people have felt for years that dual-core SHOULD have died years ago.
Of course you disagree. It's an opinion that paints Intel in a bad light, so you can't possibly have that. I've never in my life met someone so emotionally attached to a manufacturer, and so devoid of reason and being able to consider that their beloved mascot can possibly be flawed.
And where the hell does gaming come into this? I personally use Ryzen products in a professional environment that has nothing to do with gaming, and it absolutely spanks Intel in this area.
Give it up; you grabbing at straws to justify your emotional investment in a company that doesn't give a damn what you think.
Is there somewhere that states how much of HBM production Nvidia are using for Titan/Tesla cards? Im not saying its a bad thing, its all business, but HBM production is not huge and if Nvidia can afford to buy the bulk why not, the margins on the tesla cards and servers are big right? Just my perception and probably very uninformed as usual :)
Intel's initial 10nm products will be slower than it's 14nm products. It's Core I9 series tops out at 5.0 GHz for the 8 core version, and that is on 14nm.
Many people here forget that it took Intel several generations to get 14nm to where it is today. Even with 10 nm rolling out holiday 2019 they are going to face some stiff competition from AMD.
You can't base this on low end products - I don't see this is much AMD growth during Holiday 2019 - just go to local BestBuy and you see why - AMD people live in blind world of desktop gaming.
It doesn't matter how well AMD will perform. You're so blinded by your Intel fanboi-ism that every article that pops up that puts Intel in a bad light in _any_ way gets you triggered to defend them. I hope they're paying you big bugs for shilling; you certainly are earning it.
BTW, are you admitting that AMD are now superior in gaming? If so, good, go AMD :)
Now amd has to make Also 7nm cpu and 7nm gpu with TCMC production instead of using Global foundaries to do some of it. It means that amd cpu and gpu compete to het same production capasity... witch one is losing. Very good Zen or somewhat ok Vega... I think that gpu will get delayed so that They can get unm zen first. And what gives best money... workstation cpu and workstation gpu so those Are first and then later normal customer versions, when there is enough cheap production capasity to produce those...
I think a lot is being expected of 7nm, and a lot of people are going to be very disappointed when it arrives.
Like with every process since 28nm, price/transistor will continue to march up, frequency limits (stemming from voltage limits, stemming from gate oxide thickness hitting the 1nm limit a decade ago) will be the same - or possibly less due to tighter packing reducing thermal headroom, idle power will drop as will power at lower frequencies. i.e. just like every CPU for the past few years, mobile devices will get faster, while dies hitting the performance limits will only creep up a little with improved architecture. 'Scaling out' by adding more cores onto larger dies (or more dies on package) only works for workloads that scale with parallelism, and on the consumer and workstation side that's heavily into Amdahl's law (unlike Gustaffson's for HPC). It also does nothing for price/perf due to the cessation of price/transistor improvements.
There is no second coming of FinFET without radical process changes. GAAFET is the closest things to an improvement on Silicon (EUV will be great for process simplification and yeilds - though offset by startup costs - but won't fundamentally change any limits of the process itself), everything else will need new substrates and/or radical changes to IC fundamentals (e.g. Ovonics on Chalcogenides).
Let's see, AMD got 300MHz from the jump from 14nm to 12nm at Global Foundries, so 4.9GHz would be the minimum I would expect from the jump to 7nm. More significant design improvements to the Zen design will also result in an improved IPC.
A new design always has the "low hanging fruit", and Ryzen is no exception. Zen2 cores will be a significant improvement. It isn't just the move to 7nm, it is the combination of 7nm plus design improvements in the same generation that makes me(and others) feel that 2019 will be a great year for AMD.
14nm and 12nm were actyally same size, just upgraded version. The size difference is just for marketing. 7 nm is next redurece the size upgrade and that can mean low clocks, but better effiency!
I do expect a good bump from the 7nm node just for the simple fact the 14nm LPP process from GloFlo is not great for high performance parts. Will have to see but I think it significantly held back frequencies.
There will be no Vega 2, you rather mean Vega 20. If Navi is released in 2019 (it's possible that Sony wants a PS5-first release, so it might slip into 2020, though since PS5's Navi will be semi-custom that might not be an issue) it will not compete with Nvidia's high end graphics cards. At best it will compete with RTX 2070, minus the ray-tracing and DLSS, though it should be quite more power efficient and presumably cheaper*.
Dual Navi was canned due to the game developers' reluctance to target multi-GPU cards (according to AMD), but they might release dual-Navi professional cards, since in that market multiple GPUs are a requirement rather than an issue.
*Nvidia's RTX series (or, at least, the top three cards of the series) looks like it is going to be a disaster, so all bets are off. If their consumer sales slump, as is expected, AMD might gain an opening with Navi and Nvidia might be forced to move to 7nm within ~6 months to get back on track.
I feel like moving the next gen chipset to something a bit smaller than 55nm might help with platform power. Maybe something like the GF 28nm bulk process they've used in the past for inexpensive parts. This would probably help reduce power and hopefully keep manufacturing costs low, minus the dev costs of taping out a new chipset at a new node.
I know there's a tradeoff between surface area (specifically perimeter) vs I/O, but I'd expect that this should at least be theoretically possible since Intel manufactures the 1151 chipsets at 22nm and the 100/200/300 series chipsets offer more I/O than AMD's chipsets.
Of course the part of this I'm pretty ignorant about is how this would impact the overall platform. I feel like this could break backwards compatibility, at least with existing board designs, if not with the existing batch of CPUs. Since that seems to be what people think will happen with the switch to PCI-E 4.0, maybe we won't see a smaller chipset until then.
GF 12nm offers modest power/performance improvements over GF 14nm. Just by moving RR to Picasso 12nm, if they keep their frequency scaling modest, they can realize non-trivial power/performance savings and stay at least somewhat on the scale until 7nm APUs arrive. The same can be applied to their chipsets as they are largely on 14nm at the moment. As it still looks like AMD will have production comittments at GF in the near term, using the 12nm node as much as possible makes the most sense for them.
It's tricky. Is the goal best peak single-threaded performance? If so, quad-core will likely be a better bet. Or do they want improved idle power (being able to shut off more of the area if there are only a few threads running) and perhaps increased multi-core performance? If so, six core may be it.
It's partly a bet on the software ecosystem. Browsers have come a long way, many games and CPU-heavy graphics operations were still there, but other software is still catching up. Still, if you're stuck looking for ways to improve a single-threaded task in 2018 it's an obvious win to find some way to parallelize it. And if nothing else we all have "more" running in the background nowadays.
My bet would be on the six-core design and relying on binning or disabling cores to cut it down to four, letting them run at a higher power to make up. It's clear RR cores *can* use lots of energy, and they'll probably have to turn them down for most use-cases, just as we've seen with Huawei's mobile CPUs.
How they arrange six cores might be interesting - a hexagon around the L2 cache? Would be pretty, but probably not feasible. ^_^
Considering how Zen cores are packaged, they'll probably just create it so there's e.g. 2x4 compounds and just turbo the one compound for single threaded performance while putting the unneeded ones into low power or sleep state, allowing the high performance compound to perform well.
Due to the design, you could maybe make it do that each CPU gets one compound that clocks really high for single threaded and use a preferred core system that's set at manufacturing.
I hope Zen2 lives up to expectations, I'm due an upgrade at work and need the multithread capabilities, though the 2700 already looks pretty good.
Current Rumours suggest AMD going all in, and doing double the core count with 7nm, i.e going up to 64 Core. ( And I hope they charge a premium for it, )
I am not sure if they will do more core for same price, given 7nm is an expensive node. But if Zen 2 could do higher frequency and higher IPC, they could sell the same core count for same price and you still get 10 - 20% improvement.
This is it, IMO. 2018 numbers lead to 880 cinebench score. So all they need to do for 2020 is double the cores, and either up performance a little, or reduce power a little. Both seem feasible, my guess for this is perf. With 7nm and generational improvements, 2000 perf should be achievable.
So to hit this they I assume plan to combine a Zen2 and Navi core to make the APU? That does say a little something as to what they expect from the respective products efficiency wise.
Now let me think... didn't Intel bin 1 of their 28 core chips a while back and claim a huge push in speed vs. efficiency? If AMD can do even 20x improvement in 2020 I'll be very impressed. They should probably start with the Infinity Fabric...
AMD has Zen2 cores due in 2019 combined with the move to the TSMC 7nm fab process. Then you have other fun things that I believe will get announced in the next two years, such as AMD officially moving to supporting Gen-Z(Gen-Z consortium). If the chips in 2020 will support Gen-Z, the need for dedicated PCI Express and memory channels on the CPU may go away with motherboards that support it. That could provide a big improvement in overall system performance and also saving on wasted power.
afaik there is a significant difference between ryzen mobile laptops' idle power draw, conpared to intel systems'? so is optimizing on that front part of AMD's plan?
I wonder how custom APU made for ZhongShan Subor Z+ console play into this graph. 24 CU GPU could potentially increase C value by half (doubling GPU performance) and using embedded GDDR5 instead of DDR4 SODIMMs could be pretty power efficient.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
47 Comments
Back to Article
DanNeely - Wednesday, September 5, 2018 - link
" As reported by our sister website Laptop Mag, the HP Envy x2 with Raven Ridge was a repurposed chassis from HP’s catalogue, rather than something hyper optimized."I think this sentence is missing a link to a Laptop Mag article.
psxotaku - Wednesday, September 5, 2018 - link
Nice article, but I believe the rumored Ryzen 7 2800H and Ryzen 5 2600H are simply based on the Ryzen V1000 embedded series:https://www.amd.com/en/products/embedded-ryzen-v10...
Smell This - Wednesday, September 5, 2018 - link
What psxotaku said ...I tend toward low power these days so the embedded 12-watt AMD V1605B w/8 CUs floats my boat. (I hate being pragmatic and responsible.)
RR-2018 brings Vega 11 to the table along with the "Re-spin of Zen." The Internets opine (for years!) that Vega 11 has slightly higher performance at vastly better performance/Watt. Maybe Chipzilla helped to gate those Vega M CCN cores with their "Dynamic Power Sharing." HA!
In the past AMD has typically boosted clocks and efficiencies with new steppings, with no reason to think a second shot would be any different this time around.
iwod - Wednesday, September 5, 2018 - link
Raven Ridge 2018, Zen 2, Vega 2. Lots of good thing coming from AMD.wumpus - Wednesday, September 5, 2018 - link
I'm curious where the improvements for Raven Ridge 2018 are, I'd assume binning (probably for notebooks).Zen 2 is taped out for Rome (server chips). No word on Matisse (Ryzen update), Picasso (Raven Ridge update), or Castle Peak (sounds like a threadripper that would almost certainly use either Rome or Matisse chips similar to current threadrippers). As far as I can tell, Rome was "always" (meaning they switched by the time AMD had to worry about how the process effected physical layout of transistors) while the rest may have had work done assuming that GF would manufacture them. Both rework and scheduling time on TSMC fabs may delay things.
Vega 2 won't have many customers, being a professional card in the shadow of nvidia. Machine Learning developers may certainly be interested as well as the traditional CAD users. Vega 2 is also almost certainly the GPU of picasso, so I certainly hope Anandtech gives plenty of coverage to Vega 2 (ignoring how few readers are likely to buy it) and focus more on how well AMD is working with 7nm TSMC and how well vega 2 will work in picasso.
I'm surprised that Navi didn't get a mention. I'd hardly bet on a 2019 debut (unless inside a PS5) but I'd expect it to bring competition to the GPU market. I'd at least expect it to force a 7nm 21x0 line from nvidia sooner than 2080(ti) buyers will expect.
Things to ask AMD:
Is Vega 2 sampling? It should have sampled months ago. Are we going to see it before Jan 1,2019?
Are matisse, picasso, and navi taped out (I'm not so hopeful about navi). AMD typically takes a year from (official) tapeout to launch, so we need to see these things before the year (2018) is out if we expect to see them in 2019 (Q4, don't expect them any earlier). Note that "tapeout" might not be such an important thing for matisse (and to a lesser extent picasso) as they are probably backporting any issues found in Rome and won't officially "tapeout" until the Rome design is finalized.
Targon - Wednesday, September 5, 2018 - link
Vega 7nm has been sampling for some time now. A figure of a 40 percent boost to performance over Vega 64 was put out there several months ago. The problem is that Vega is very much tied to HBM2 memory, and due to higher prices/lower availability, AMD is keeping the volume low for that very reason. If HBM2 memory were readily available and the price wasn't as high, then AMD would probably have Vega 7nm out there for gamers at the end of this year.Ryzen third generation should also be in good shape at this point, but getting yields up at TSMC may be why AMD is not talking about it(to avoid wild speculation that would only hurt the company). Until AMD has at least 300,000 units ready, I don't think AMD wants to put any timeline on when it will officially launch the product. AMD may also be waiting to see what speed Intel i9 chips can actually run at so that AMD can properly price the third generation chips appropriately. If AMD can hit 5GHz on air and Intel can hit 5.3GHz on air, if AMD is able to improve IPC by 10-15 percent, that is the sort of thing AMD will want a big launch event to cover. If AMD can get third generation to the same 5.3GHz on air, that is the sort of thing that AMD management will want to really push.
So, decisions are being made, but until Intel actually ships the new i9 chips, AMD will probably stay silent.
HStewart - Wednesday, September 5, 2018 - link
Why is that when AMD makes slight changes they call it progress but when Intel makes the changes they call it Marketing or rebranded - they slides here clearly look like Marketing.FreckledTrout - Wednesday, September 5, 2018 - link
I can explain HStewert. The last few years Intel was not making the best product it can make but AMD is giving it all they have. It's simple as that and everyone that follows tech knows Intel has not put its best product it can make out in the consumer space. In servers sure, but not in the consumer / HEDT segment.iwod - Thursday, September 6, 2018 - link
Exactly this.Where are the slight changes? I am looking at Skylake, Kaby Lake, Kaby Lake+ , Coffee lake, and I cant remember what lake is it now they are doing.
Than I am looking at Zen and Zen +, and coming Zen 2.
So you thought after all these years Intel has been milking their products, they should have something up their sleeves? No, none at all.
Intel has been marketing their product as a New "Generation". Which is basically redefining what generation means in my dictionary.
Then there is the Intel IGP.
Santoval - Monday, September 10, 2018 - link
While Intel has surely procrastinated technologically (largely because they had no competition) one reason that their architecture has been basically frozen since Skylake is that they decided to tie their next architecture (Ice Lake) to their next node, and we all know their problems with their 10nm node. The second generation (10nm+) of their next node actually, since they originally planned to use 10nm for Cannon Lake, which is basically the Skylake design at 10nm plus an AVX-512 block.Cannon Lake looks like it will remain a beta node, perhaps with only the sole Core-i3 15W CPU with disabled iGPU that was released in low volume. Ice Lake is planned for HVM in mid 2019 and wide release (from top to bottom consumer) in late 2019. On the other hand it appears that they do not have too much confidence in Ice Lake and its successor Tiger Lake, which is why they hired Jim Keller to design Ocean Cove, which is set to be their first post-Core architecture since the original Core CPUs were released in 2006.
The first Ocean Cove CPUs are expected to be released in 2021 at the earliest (~15 years after the original Core was released), either at a late 10nm node or an early 7nm node.
Santoval - Monday, September 10, 2018 - link
"...which is why they hired Jim Keller to design Ocean Cove, which is set to be their first post-Core architecture since the original Core CPUs were released in 2006."And post-Lake, of course, since it is surely no accident, that they are moving from lakes to coves and oceans.
HStewart - Thursday, September 6, 2018 - link
Everyone has opinions - and I disagree with your opinion - I just believe some people have limited focus on gaming market and don't see the entire picture.Targon - Friday, September 7, 2018 - link
The issue that I see is that Intel could have had 5.3GHz chips out for several years now, but used their horrible TIM to artificially limit how fast the chips could be clocked to. You have had people doing a delid on their Intel chips for this reason, to get the most they can out of their processors.So, if IPC did not go up all that much, and true potential for clock speeds has not gone up by all that much(as I said, a delid fixed the problems with the bad TIM that Intel had been using), then, what real improvement has there been on the Intel side? If AMD is able to hit 5GHz on air with Ryzen 3rd generation, even if Intel still holds a lead for potential, that would be seen as a huge jump forward from AMD.
Until Ryzen, Intel was still pushing 2 and 4 core chips on the desktop, and many people have felt for years that dual-core SHOULD have died years ago.
iwod - Saturday, September 8, 2018 - link
no idea why you think it has anything to do with gaming.sa666666 - Saturday, September 8, 2018 - link
Of course you disagree. It's an opinion that paints Intel in a bad light, so you can't possibly have that. I've never in my life met someone so emotionally attached to a manufacturer, and so devoid of reason and being able to consider that their beloved mascot can possibly be flawed.And where the hell does gaming come into this? I personally use Ryzen products in a professional environment that has nothing to do with gaming, and it absolutely spanks Intel in this area.
Give it up; you grabbing at straws to justify your emotional investment in a company that doesn't give a damn what you think.
sa666666 - Saturday, September 8, 2018 - link
And I believe some people have limited focus on blind fanboi-ism and don't see the entire picture. Guess where you fit into that scenario?WatcherCK - Wednesday, September 5, 2018 - link
Is there somewhere that states how much of HBM production Nvidia are using for Titan/Tesla cards? Im not saying its a bad thing, its all business, but HBM production is not huge and if Nvidia can afford to buy the bulk why not, the margins on the tesla cards and servers are big right? Just my perception and probably very uninformed as usual :)iwod - Thursday, September 6, 2018 - link
Because Nvidia made much better margin on those Tesla cards and they can afford to pay premium to get what ever HBM volume they need.eek2121 - Wednesday, September 5, 2018 - link
Intel's initial 10nm products will be slower than it's 14nm products. It's Core I9 series tops out at 5.0 GHz for the 8 core version, and that is on 14nm.Many people here forget that it took Intel several generations to get 14nm to where it is today. Even with 10 nm rolling out holiday 2019 they are going to face some stiff competition from AMD.
HStewart - Thursday, September 6, 2018 - link
You can't base this on low end products - I don't see this is much AMD growth during Holiday 2019 - just go to local BestBuy and you see why - AMD people live in blind world of desktop gaming.sa666666 - Saturday, September 8, 2018 - link
It doesn't matter how well AMD will perform. You're so blinded by your Intel fanboi-ism that every article that pops up that puts Intel in a bad light in _any_ way gets you triggered to defend them. I hope they're paying you big bugs for shilling; you certainly are earning it.BTW, are you admitting that AMD are now superior in gaming? If so, good, go AMD :)
Legdotus - Thursday, September 27, 2018 - link
You are pathetic. Yout live in a blind world where everybody is a gaming nerd. Wake up kid!Trixanity - Wednesday, September 5, 2018 - link
AMD already said Vega 7nm launches in 2018. It was shown at Computex. I would assume it had been sampling during or shortly after the showcase.Targon - Friday, September 7, 2018 - link
Vega 7nm is for the professional market, AI and such, not for gaming or typical consumer use.haukionkannel - Wednesday, September 5, 2018 - link
Now amd has to make Also 7nm cpu and 7nm gpu with TCMC production instead of using Global foundaries to do some of it. It means that amd cpu and gpu compete to het same production capasity... witch one is losing. Very good Zen or somewhat ok Vega...I think that gpu will get delayed so that They can get unm zen first. And what gives best money... workstation cpu and workstation gpu so those Are first and then later normal customer versions, when there is enough cheap production capasity to produce those...
edzieba - Wednesday, September 5, 2018 - link
I think a lot is being expected of 7nm, and a lot of people are going to be very disappointed when it arrives.Like with every process since 28nm, price/transistor will continue to march up, frequency limits (stemming from voltage limits, stemming from gate oxide thickness hitting the 1nm limit a decade ago) will be the same - or possibly less due to tighter packing reducing thermal headroom, idle power will drop as will power at lower frequencies. i.e. just like every CPU for the past few years, mobile devices will get faster, while dies hitting the performance limits will only creep up a little with improved architecture. 'Scaling out' by adding more cores onto larger dies (or more dies on package) only works for workloads that scale with parallelism, and on the consumer and workstation side that's heavily into Amdahl's law (unlike Gustaffson's for HPC). It also does nothing for price/perf due to the cessation of price/transistor improvements.
There is no second coming of FinFET without radical process changes. GAAFET is the closest things to an improvement on Silicon (EUV will be great for process simplification and yeilds - though offset by startup costs - but won't fundamentally change any limits of the process itself), everything else will need new substrates and/or radical changes to IC fundamentals (e.g. Ovonics on Chalcogenides).
edzieba - Wednesday, September 5, 2018 - link
This applies equally to 7nm EUV, 7nm SAQP or 10nm SAQP.Targon - Wednesday, September 5, 2018 - link
Let's see, AMD got 300MHz from the jump from 14nm to 12nm at Global Foundries, so 4.9GHz would be the minimum I would expect from the jump to 7nm. More significant design improvements to the Zen design will also result in an improved IPC.A new design always has the "low hanging fruit", and Ryzen is no exception. Zen2 cores will be a significant improvement. It isn't just the move to 7nm, it is the combination of 7nm plus design improvements in the same generation that makes me(and others) feel that 2019 will be a great year for AMD.
haukionkannel - Thursday, September 6, 2018 - link
14nm and 12nm were actyally same size, just upgraded version. The size difference is just for marketing. 7 nm is next redurece the size upgrade and that can mean low clocks, but better effiency!FreckledTrout - Wednesday, September 5, 2018 - link
I do expect a good bump from the 7nm node just for the simple fact the 14nm LPP process from GloFlo is not great for high performance parts. Will have to see but I think it significantly held back frequencies.Santoval - Monday, September 10, 2018 - link
There will be no Vega 2, you rather mean Vega 20. If Navi is released in 2019 (it's possible that Sony wants a PS5-first release, so it might slip into 2020, though since PS5's Navi will be semi-custom that might not be an issue) it will not compete with Nvidia's high end graphics cards. At best it will compete with RTX 2070, minus the ray-tracing and DLSS, though it should be quite more power efficient and presumably cheaper*.Dual Navi was canned due to the game developers' reluctance to target multi-GPU cards (according to AMD), but they might release dual-Navi professional cards, since in that market multiple GPUs are a requirement rather than an issue.
*Nvidia's RTX series (or, at least, the top three cards of the series) looks like it is going to be a disaster, so all bets are off. If their consumer sales slump, as is expected, AMD might gain an opening with Navi and Nvidia might be forced to move to 7nm within ~6 months to get back on track.
MrCommunistGen - Wednesday, September 5, 2018 - link
I feel like moving the next gen chipset to something a bit smaller than 55nm might help with platform power. Maybe something like the GF 28nm bulk process they've used in the past for inexpensive parts. This would probably help reduce power and hopefully keep manufacturing costs low, minus the dev costs of taping out a new chipset at a new node.I know there's a tradeoff between surface area (specifically perimeter) vs I/O, but I'd expect that this should at least be theoretically possible since Intel manufactures the 1151 chipsets at 22nm and the 100/200/300 series chipsets offer more I/O than AMD's chipsets.
Of course the part of this I'm pretty ignorant about is how this would impact the overall platform. I feel like this could break backwards compatibility, at least with existing board designs, if not with the existing batch of CPUs. Since that seems to be what people think will happen with the switch to PCI-E 4.0, maybe we won't see a smaller chipset until then.
Trixanity - Wednesday, September 5, 2018 - link
With the exception of Z370 (which is 22nm), the 300 series chipsets are actually 14nm.lightningz71 - Wednesday, September 5, 2018 - link
GF 12nm offers modest power/performance improvements over GF 14nm. Just by moving RR to Picasso 12nm, if they keep their frequency scaling modest, they can realize non-trivial power/performance savings and stay at least somewhat on the scale until 7nm APUs arrive. The same can be applied to their chipsets as they are largely on 14nm at the moment. As it still looks like AMD will have production comittments at GF in the near term, using the 12nm node as much as possible makes the most sense for them.WatcherCK - Wednesday, September 5, 2018 - link
With a shrink to 7nm would Zen 2 get a core jump? 6 core (maybe even at 4GHz all core) plus some more compute cores for around $250 US?GreenReaper - Wednesday, September 5, 2018 - link
It's tricky. Is the goal best peak single-threaded performance? If so, quad-core will likely be a better bet. Or do they want improved idle power (being able to shut off more of the area if there are only a few threads running) and perhaps increased multi-core performance? If so, six core may be it.It's partly a bet on the software ecosystem. Browsers have come a long way, many games and CPU-heavy graphics operations were still there, but other software is still catching up. Still, if you're stuck looking for ways to improve a single-threaded task in 2018 it's an obvious win to find some way to parallelize it. And if nothing else we all have "more" running in the background nowadays.
My bet would be on the six-core design and relying on binning or disabling cores to cut it down to four, letting them run at a higher power to make up. It's clear RR cores *can* use lots of energy, and they'll probably have to turn them down for most use-cases, just as we've seen with Huawei's mobile CPUs.
How they arrange six cores might be interesting - a hexagon around the L2 cache? Would be pretty, but probably not feasible. ^_^
RSAUser - Friday, September 7, 2018 - link
Considering how Zen cores are packaged, they'll probably just create it so there's e.g. 2x4 compounds and just turbo the one compound for single threaded performance while putting the unneeded ones into low power or sleep state, allowing the high performance compound to perform well.Due to the design, you could maybe make it do that each CPU gets one compound that clocks really high for single threaded and use a preferred core system that's set at manufacturing.
I hope Zen2 lives up to expectations, I'm due an upgrade at work and need the multithread capabilities, though the 2700 already looks pretty good.
iwod - Thursday, September 6, 2018 - link
Current Rumours suggest AMD going all in, and doing double the core count with 7nm, i.e going up to 64 Core. ( And I hope they charge a premium for it, )I am not sure if they will do more core for same price, given 7nm is an expensive node. But if Zen 2 could do higher frequency and higher IPC, they could sell the same core count for same price and you still get 10 - 20% improvement.
Mil0 - Monday, September 17, 2018 - link
This is it, IMO. 2018 numbers lead to 880 cinebench score. So all they need to do for 2020 is double the cores, and either up performance a little, or reduce power a little. Both seem feasible, my guess for this is perf. With 7nm and generational improvements, 2000 perf should be achievable.FreckledTrout - Wednesday, September 5, 2018 - link
So to hit this they I assume plan to combine a Zen2 and Navi core to make the APU? That does say a little something as to what they expect from the respective products efficiency wise.ballsystemlord - Wednesday, September 5, 2018 - link
Now let me think... didn't Intel bin 1 of their 28 core chips a while back and claim a huge push in speed vs. efficiency?If AMD can do even 20x improvement in 2020 I'll be very impressed. They should probably start with the Infinity Fabric...
Targon - Thursday, September 6, 2018 - link
AMD has Zen2 cores due in 2019 combined with the move to the TSMC 7nm fab process. Then you have other fun things that I believe will get announced in the next two years, such as AMD officially moving to supporting Gen-Z(Gen-Z consortium). If the chips in 2020 will support Gen-Z, the need for dedicated PCI Express and memory channels on the CPU may go away with motherboards that support it. That could provide a big improvement in overall system performance and also saving on wasted power.RSAUser - Friday, September 7, 2018 - link
Interesting, thanks for mentioning this.hanselltc - Thursday, September 6, 2018 - link
afaik there is a significant difference between ryzen mobile laptops' idle power draw, conpared to intel systems'? so is optimizing on that front part of AMD's plan?GreenReaper - Friday, September 7, 2018 - link
I'm sure that they're hoping the move to 7nm will help them achieve that.vithrell - Thursday, September 6, 2018 - link
I wonder how custom APU made for ZhongShan Subor Z+ console play into this graph. 24 CU GPU could potentially increase C value by half (doubling GPU performance) and using embedded GDDR5 instead of DDR4 SODIMMs could be pretty power efficient.Kalpana123 - Sunday, September 9, 2018 - link
RR-2018 brings Vega 11 to the table along with the "Re-spin of Zen." The Internets opine (for years!) that Vega 11 has slightly higher performance at vastly better performance/Watt. Maybe Chipzilla helped to gate those Vega M CCN cores with their "Dynamic Power Sharing." HAhttp://generaltreatments.xyz/fever-treatment-sympt...
http://generaltreatments.xyz/asthma
http://generaltreatments.xyz/upper-respiratory-tra...