AMD to Detail Zen 2, Navi Architectures Come Hot Chips in August

[jabbadap]

Hot Chips is not where you go to launch a consumer product like Ryzen or Radeon, it's an industry conference for technical deep-dives and architectural explanations with a high to extreme level of detail. It's not consumer oriented whatsoever.

Besides, the topics of the Hot Chips talks are confirmed by AMD and Hot Chips. Zen 2 is one, Navi is another, with a third keynote for ecosystem/HPC/compute more generally.


In all likelihood:
Ryzen 3000 will launch at Computex.
Navi will be teased at Computex and launch at E3.
AMD has confirmed keynotes at both, and the timing makes sense. They might not be hard launches, but they'll be launches. The level of inventory currently doesn't matter much - it'll mostly sell out in the coming months anyhow, and AMD is always good at discounting previous-gen products.

Well something what to expect of that navi presentation would be looking in the past what amd said about their last major new GCN mach update Vega at the Hotchips 2017. And at the that time Vega had been out since FE edition launch, though RX Vegas were launched just a few days before that presentation.

 

[oxidized]

Less for mainstream, much for for enterprise. But the fact remains Intels old Core arch has more problems. You may claim otherwise but the data shows that to be the case.

I'd say, not at all for mainstream, and much for enterprice. I agree intel's arch has more problems, that's a fact, but it doesn't have some problems Zen has, and it's much older

True but AMD got it's act together and started producing worthy products. Intel is yet to do so.

They just weren't expecting such a response after all these years, but i'm sure of one thing, when they'll answer, they'll answer in full force, and there won't be any Zen or Jim Keller to the rescue, intel has much more money, and they'll be more than happy to invest them in something that can put them in an even higher position in the market.

That is your experience. My experience and the reviews show otherwise. AMD seems to have better 0,1% and 1% lows. And better frametimes - especially when streaming.
The expensive memory thing is constantly brought up by Intel supporters. It may have had merit last year but these days you can get 2x8GB 3200 CL16 kit for less than 100$/€ which is more than enough for Ryzen. Or better yet get a 3000 kit and set it to 3200 for 50-70$. Literally changing one BIOS setting.
Yes GN audience is more tech savvy but these people also affect buying decisions of their friends and relatives. And when they say get AMD people will.

Fine, but what matters in the end is first person experience, not reviews or suggestions. The expensive memory thing is a fact, and it's especially true when we're talking about ryzen 1xxx, because infinity fabric latencies are directly tied to how fast memories are, besides for a big period of its life ryzen 1xxx has even problems with many types of memory and only digested a few types of brands and models. No that's just not an example to talk about, GN audience is GN audience, and doesn't directly or indirectly reflect the big market choice.

What was wrong with your 480?

Different things, pretty noticeable coil whine, plus while i was watching some tv series i instantly noticed its non existent antialiasing, and movies and tv series were pretty terrible to watch, so much i noticed it immediately and instantly tried switching back to my old GTX580 to be sure i wasn't going crazy, and i wasn't. After this i asked amazon to swap it, and they said they didn't have any sample to replace it with, the day after i found a greatly priced 1060 Gaming X and i bought that instead.

 

[Valantar]

I'd say, not at all for mainstream, and much for enterprice. I agree intel's arch has more problems, that's a fact, but it doesn't have some problems Zen has, and it's much older


They just weren't expecting such a response after all these years, but i'm sure of one thing, when they'll answer, they'll answer in full force, and there won't be any Zen or Jim Keller to the rescue, intel has much more money, and they'll be more than happy to invest them in something that can put them in an even higher position in the market.


Fine, but what matters in the end is first person experience, not reviews or suggestions. The expensive memory thing is a fact, and it's especially true when we're talking about ryzen 1xxx, because infinity fabric latencies are directly tied to how fast memories are, besides for a big period of its life ryzen 1xxx has even problems with many types of memory and only digested a few types of brands and models. No that's just not an example to talk about, GN audience is GN audience, and doesn't directly or indirectly reflect the big market choice.

Different things, pretty noticeable coil whine, plus while i was watching some tv series i instantly noticed its non existent antialiasing, and movies and tv series were pretty terrible to watch, so much i noticed it immediately and instantly tried switching back to my old GTX580 to be sure i wasn't going crazy, and i wasn't. After this i asked amazon to swap it, and they said they didn't have any sample to replace it with, the day after i found a greatly priced 1060 Gaming X and i bought that instead.

I have to ask: AA in video? What video player are you using? Which renderer? While I'm by no means an HTPC expert, I've never heard of or seen AA settings in a video player, nor heard anyone mention it as an issue. AFAIK the last thing you would want in video playback is a post-processing filter effectively removing detail and reducing sharpness.

Well something what to expect of that navi presentation would be looking in the past what amd said about their last major new GCN mach update Vega at the Hotchips 2017. And at the that time Vega had been out since FE edition launch, though RX Vegas were launched just a few days before that presentation.

That's true, but AMD didn't have a giant E3 presence that year, let alone one directly following a Computex keynote. And given that there's no high-end card for pros/prosumers coming (it's already out, it's the VII), what would they do at E3? AMD had full product reveals at E3 in both 2015 and 2016. That is of course not proof that there'll be a launch there this year too, but it's as good an indicator as any. And they really need a shot in the arm in the gaming market now, so it would be very, very weird PR wise to squander an opportunity like that.

 

[aQi]

I am surprised that amd will be using pci 4.0 standards and thats a catch for ssd market only. If im not wrong pci 4 standards took a leap and meanwhile where no body bothered to use these standards pci 5.0 got invented.
Since intel is just calm and looking forward i guess intel might be giving amd a chance to release pci 4.0 so that they can come over to their pci 5.0 standards. Both these speeds might co exist while only and only ssd will burst out revolutionary speeds. Iv even read amd wont provide pci 4 standard for navi gpus.

 

[TheoneandonlyMrK]

I am surprised that amd will be using pci 4.0 standards and thats a catch for ssd market only. If im not wrong pci 4 standards took a leap and meanwhile where no body bothered to use these standards pci 5.0 got invented.
Since intel is just calm and looking forward i guess intel might be giving amd a chance to release pci 4.0 so that they can come over to their pci 5.0 standards. Both these speeds might co exist while only and only ssd will burst out revolutionary speeds. Iv even read amd wont provide pci 4 standard for navi gpus.

I think everyone expects big platform change's to enable pciec5 (and am5),I wouldn't be surprised if Amd didn't realize a cross compatible Infinity fabric protocol to use on various pciex types.
To optimise platforms built out of their chips.

 

[Xuper]

https://twitter.com/i/web/status/1130123724043235329

That 5000/4400 MB/s (Read/write) is a lot.
16 Phase VRM probably Fake VRM? just 8 real Phase VRM?

 

[oxidized]

I have to ask: AA in video? What video player are you using? Which renderer? While I'm by no means an HTPC expert, I've never heard of or seen AA settings in a video player, nor heard anyone mention it as an issue. AFAIK the last thing you would want in video playback is a post-processing filter effectively removing detail and reducing sharpness.

I had no idea either, but it only happened when playing a video, i use potplayer and i tried also VLC, same problem with the RX480, no problems with the GTX580.

 

[jabbadap]

https://twitter.com/i/web/status/1130123724043235329

That 5000/4400 MB/s (Read/write) is a lot.
16 Phase VRM probably Fake VRM? just 8 real Phase VRM?

Most probably maxed out 8 -phase controller with doublers, like their other aorus extreme boards have. Watch Buildzoids youtube video about z390 aorus extreme, which too have 16-phases.

 

[Valantar]

I am surprised that amd will be using pci 4.0 standards and thats a catch for ssd market only. If im not wrong pci 4 standards took a leap and meanwhile where no body bothered to use these standards pci 5.0 got invented.
Since intel is just calm and looking forward i guess intel might be giving amd a chance to release pci 4.0 so that they can come over to their pci 5.0 standards. Both these speeds might co exist while only and only ssd will burst out revolutionary speeds. Iv even read amd wont provide pci 4 standard for navi gpus.

You seem to have a rather naive view of the time required for a PCIe standard to move from initial draft spec to ratification to implementation across segments. You're right that PCIe 5.0 was fast-tracked (partially due to PCIe 3.0 having a longer-than-expected lifespan and the need for more bandwidth suddenly exploding, while 4.0 had languished in committee for years and years), and there was speculation that consumer platforms might skip it because of that - but that obviously didn't come true. PCIe 3.0 was ratified in 2010, but didn't show up in consumer platforms until Intel Ivy Bridge (mid-2012). PCIe 4.0 was ratified in june 2017 and is now becoming available for consumers two years later. In other words, the time to market is identical. PCIe 5.0 is yet to be standardized, and at best will reach that point this year. From then we can expect another 2+ years before it reaches consumers, especially on CPUs/motherboards - development cycles for these products are easily 2+ years, and for the most part you can't just copy+paste in a new component late in the design.

I had no idea either, but it only happened when playing a video, i use potplayer and i tried also VLC, same problem with the RX480, no problems with the GTX580.

That's very weird. I'm guessing there was some sort of decoder or driver issue, but I have zero idea how to troubleshoot it. Did you DDU when installing the new card? Might have caused some issues if not.

https://twitter.com/i/web/status/1130123724043235329

That 5000/4400 MB/s (Read/write) is a lot.
16 Phase VRM probably Fake VRM? just 8 real Phase VRM?

People need to stop complaining about "fake VRM". Unless you're an LN2 overclocker running the components close to spec (or over) with a need for extremely low voltage variance, there is zero perceptible difference between a doubled VRM setup and one that isn't - except for the latter being far more expensive, of course, as you'll either need dual controllers or some extremely complex and expensive ones. Doubling just means each side of each phase gets switched at half the frequency. With modern high frequency switching this doesn't matter whatsoever. Besides, even 8 single phases with high quality MOSFETS and chokes can deliver hundreds of amps. Doubling then just means more headroom and less heat.

 

[Xuper]

People need to stop complaining about "fake VRM". Unless you're an LN2 overclocker running the components close to spec (or over) with a need for extremely low voltage variance, there is zero perceptible difference between a doubled VRM setup and one that isn't - except for the latter being far more expensive, of course, as you'll either need dual controllers or some extremely complex and expensive ones. Doubling just means each side of each phase gets switched at half the frequency. With modern high frequency switching this doesn't matter whatsoever. Besides, even 8 single phases with high quality MOSFETS and chokes can deliver hundreds of amps. Doubling then just means more headroom and less heat.

We recently spoke with Buildzoid about the new Gigabyte B450 motherboard that was covered on his channel, where Buildzoid discussed how marketing is beginning to further blur the lines between what defines a “phase” in a VRM. Typically, you’d define a phase as a driver, a hi-side MOSFET, a low-side MOSFET, and a choke, which has its own PWM signal out of phase with the signal from other phases. We’ve seen boards in the past that do doubling schemes, where they go with 2 hi-side, 2 low-side, and 2 chokes, all driven by 1 driver and called “2 phases.” That’s not new. What’s new is that Gigabyte has axed a hi-side MOSFET on B450, resulting in 1 hi-side, 2 low-side, 2 chokes, and 1 driver. This makes the motherboard look like an 8-phase board, and Gigabyte marketed it as an 8-phase board, but it’s really not.

Source :

We will see soon when Buildzoid discusses about X570 Aorus Xtreme but I would be gladly If gigabyte doesn't screw it up.point is that gigabyte lied about VRM on B450 series.now I would be caution What they say.

Most probably maxed out 8 -phase controller with doublers, like their other aorus extreme boards have. Watch Buildzoids youtube video about z390 aorus extreme, which too have 16-phases.

I really hope those VRMs are same Z390 aorus extreme .Gigabyte had a worse reputation before , that's why I chose Asus for beefy VRM.

 

[Valantar]

Source :

We will see soon when Buildzoid discusses about X570 Aorus Xtreme but I would be gladly If gigabyte doesn't screw it up.point is that gigabyte lied about VRM on B450 series.now I would be caution What they say.



I really hope those VRMs are same Z390 aorus extreme .Gigabyte had a worse reputation before , that's why I chose Asus for beefy VRM.

While that's a shady move, 4 decent MOSFETS can still deliver quite a few amps of power. Not suitable for high-end overclocking, but ought to work okay for anything else. Not ideal, and as I said, shady, but the board is in all likelihood still perfectly fine. It will be slightly better than a regular 4-phase board, after all.

 

[aQi]

You seem to have a rather naive view of the time required for a PCIe standard to move from initial draft spec to ratification to implementation across segments. You're right that PCIe 5.0 was fast-tracked (partially due to PCIe 3.0 having a longer-than-expected lifespan and the need for more bandwidth suddenly exploding, while 4.0 had languished in committee for years and years), and there was speculation that consumer platforms might skip it because of that - but that obviously didn't come true. PCIe 3.0 was ratified in 2010, but didn't show up in consumer platforms until Intel Ivy Bridge (mid-2012). PCIe 4.0 was ratified in june 2017 and is now becoming available for consumers two years later. In other words, the time to market is identical. PCIe 5.0 is yet to be standardized, and at best will reach that point this year. From then we can expect another 2+ years before it reaches consumers, especially on CPUs/motherboards - development cycles for these products are easily 2+ years, and for the most part you can't just copy+paste in a new component late in the design.

This isnt the argue anyway. As per market and a slice of per consumer demand it is essential for companies to introduce new architecture once it gets ready. If you remember it took time from PCIe 1.0 to PCIe 2.0. Worth the while if PCIe 3.0 was invented during the same phase wouldn’t it in the place of PCIe 2.0 ?
Its just that PCIe 5.0 standards are already ready and it cant be denied that it can be adopted in place of PCIe 4.0. Its just that PCIe 3.0 is enough for consumers, even that isnt the big catch as most people with PCIe 2.0 are comfortable and achieve their desirable tasks performed at good speeds. Where as PCIe 3.0 is only beneficial in terms of super speeds of NVME SSDs.
My intensions are simple if it was for PCIe 4.0 which isnt a major breakthrough why not use PCIe 5.0 instead. I presume its how the market trends to be rather then implementation.

 

[Valantar]

This isnt the argue anyway. As per market and a slice of per consumer demand it is essential for companies to introduce new architecture once it gets ready. If you remember it took time from PCIe 1.0 to PCIe 2.0. Worth the while if PCIe 3.0 was invented during the same phase wouldn’t it in the place of PCIe 2.0 ?
Its just that PCIe 5.0 standards are already ready and it cant be denied that it can be adopted in place of PCIe 4.0. Its just that PCIe 3.0 is enough for consumers, even that isnt the big catch as most people with PCIe 2.0 are comfortable and achieve their desirable tasks performed at good speeds. Where as PCIe 3.0 is only beneficial in terms of super speeds of NVME SSDs.
My intensions are simple if it was for PCIe 4.0 which isnt a major breakthrough why not use PCIe 5.0 instead. I presume its how the market trends to be rather then implementation.

But that's exactly where you're wrong - PCIe 5.0 isn't a finished specification yet. It is planned to become so this year. And, again, it takes a few years from the specification is finalized until an actual mass-producible hardware implementation of the standard is ready. This requires R&D, testing, QC, and integration into a CPU and/or motherboard design. This is not a trivial process, which is why there is always a waiting period between the certification of a PCIe standard and its implementation in consumer products.

So, yes, it can be denied that PCIe 5.0 can be adopted instead of 4.0, because it isn't ready or available yet. PCIe 4.0 hardware is just becoming available now. 5.0 won't be for another few years at best.

You're right that consumers only really "need" faster PCIe for SSDs, but more PCIe is more important than faster for most use cases. If a PCIe 4.0 x2 SSD is as fast as a PCIe 3.0 x4 one, that means you can make a simpler controller and route half as many lanes to its socket, making motherboards easier to design and make. That's a win-win scenario, including consumers buying off-the-shelf PCs. It also allows for attaching more SSDs to the same amount of lanes from a CPU or chipset. And more fast USB controllers, Thunderbolt controllers, and so on. So no, we don't have much use for faster, but the same speed with fewer lanes (=more devices with the same number of lanes) has a lot going for it.

 

[oxidized]

That's very weird. I'm guessing there was some sort of decoder or driver issue, but I have zero idea how to troubleshoot it. Did you DDU when installing the new card? Might have caused some issues if not.

I tried everything possible, believe me...

 

[Valantar]

I tried everything possible, believe me...

Weird. Makes me want to check video playback quality across the PCs in my household (two Intel laptops, an AMD dGPU desktop, an AMD APU HTPC, and an Nvidia dGPU desktop). No doubt there'll be differences in rendering, but enough to be noticeable without pausing the video? Sounds unlikely to me, but that's why it would be worth checking

 

[TheLostSwede]

This isnt the argue anyway. As per market and a slice of per consumer demand it is essential for companies to introduce new architecture once it gets ready. If you remember it took time from PCIe 1.0 to PCIe 2.0. Worth the while if PCIe 3.0 was invented during the same phase wouldn’t it in the place of PCIe 2.0 ?
Its just that PCIe 5.0 standards are already ready and it cant be denied that it can be adopted in place of PCIe 4.0. Its just that PCIe 3.0 is enough for consumers, even that isnt the big catch as most people with PCIe 2.0 are comfortable and achieve their desirable tasks performed at good speeds. Where as PCIe 3.0 is only beneficial in terms of super speeds of NVME SSDs.
My intensions are simple if it was for PCIe 4.0 which isnt a major breakthrough why not use PCIe 5.0 instead. I presume its how the market trends to be rather then implementation.

Considering that PCIe 4.0 already has issues in terms of how the PCB traces can be made while maintaining reasonable signal integrity, PCIe 5.0 is going to require much more costly PCB designs or a complete re-think with regards to board layout. PCIe 4.0 requires expensive re-timers/signal boosters. Have a look here for example https://www.intel.com/content/www/u...express-gen4-retimer-features-tech-brief.html
As PCIe 5.0 doubles the speed, it gets even more complex and you can read a bit about that here https://www.synopsys.com/designware...s-of-moving-to-32gts-pcie-designs-2018q1.html
It's not trivial stuff at all and considering how things have developed, I doubt we'll see PCIe 5.0 in the consumer market any time soon, simply because the ATX motherboard standard was never designed to take high speed interfaces like this into account.

 

[oxidized]

Weird. Makes me want to check video playback quality across the PCs in my household (two Intel laptops, an AMD dGPU desktop, an AMD APU HTPC, and an Nvidia dGPU desktop). No doubt there'll be differences in rendering, but enough to be noticeable without pausing the video? Sounds unlikely to me, but that's why it would be worth checking

I know it sounded weird to me too, but i made 2 screenshot of the same video, about at the same frame, and the screenshots don't even make it justice, watching it in person made it much more noticeable, even because everything was moving. Here are those 2 snaps i was talking about.

http://imgur.com/9V2H64s

http://imgur.com/QKku3QS

Look at pretty much all edges, but you'll notice it much easier looking at the counter, and above the mirror on the right, again, it was much more noticeable in person, because everything was moving, and this is a compressed image. I also tried with other videos, ofc, same result.

 

[aQi]

Considering that PCIe 4.0 already has issues in terms of how the PCB traces can be made while maintaining reasonable signal integrity, PCIe 5.0 is going to require much more costly PCB designs or a complete re-think with regards to board layout. PCIe 4.0 requires expensive re-timers/signal boosters. Have a look here for example https://www.intel.com/content/www/u...express-gen4-retimer-features-tech-brief.html
As PCIe 5.0 doubles the speed, it gets even more complex and you can read a bit about that here https://www.synopsys.com/designware...s-of-moving-to-32gts-pcie-designs-2018q1.html
It's not trivial stuff at all and considering how things have developed, I doubt we'll see PCIe 5.0 in the consumer market any time soon, simply because the ATX motherboard standard was never designed to take high speed interfaces like this into account.

Thats spares alot of time thinking why amd only provided pcie 4.0 support to x16 slot on previous motherboards. Clearly if thats the case in terms of hurdles achieving such bandwidth i must say amd might be using in house development to perfect and control such speeds. Apart from that Phison with its PCIe Nvme SSD controller achieved remarkable speeds none the less this increase in bandwidth whether PCIe 4 or PCIe 5 will only benefit SSD or perhaps might say OCuLink-2 connector or new Thunderbolt might show up.

 

[TheLostSwede]

Thats spares alot of time thinking why amd only provided pcie 4.0 support to x16 slot on previous motherboards. Clearly if thats the case in terms of hurdles achieving such bandwidth i must say amd might be using in house development to perfect and control such speeds. Apart from that Phison with its PCIe Nvme SSD controller achieved remarkable speeds none the less this increase in bandwidth whether PCIe 4 or PCIe 5 will only benefit SSD or perhaps might say OCuLink-2 connector or new Thunderbolt might show up.

PCIe 4.0 is likely to benefit a lot of things over time, but right now, NVMe based storage is the most logical devices that will gain benefits, especially as there will be PCIe 4.0 based NVMe drives in the not too distant future. If the numbers in the tweet in this thread are anything to go by, we should see some intimidate performance gains, even though they might not be useful in a practical sense.
I have a feeling the next thing we'll see is 10Gbps Ethernet cards, as we can then go down to a x1 interface.

 

[efikkan]

Have some/any of the existing AM4 motherboards gotten a similar feature as ASUS' BIOS flashback? Or can we expect another round of AMD having to ship out old CPUs to customers to allow them to upgrade BIOSes?

 

[EarthDog]

Higher-end boards typically come with flashback capabilities.... but this is a known issue for either camp when using older chipsets.

 

[Valantar]

As always, the safe alternative for anyone planning to buy an older-series motherboard with a new CPU is to wait until you can rely on stock to have the most recent BIOS updates - which takes a month or two. For anyone else, this isn't an issue - either you're buying a new-series motherboard, or you have an older CPU already that you're upgrading from which you can use for upgrading the BIOS.

 

[EarthDog]

which takes a month or two.

actually, it can vary wildly depending on where you source the board. A month or two isn't typically enough time.

Some places can order [insert # here] and sell [insert lesser # here] with the remaining quantity just sitting there. It can take multiple months and even then it is certainly not a guarantee to get one. It all depends on where you buy and its stock numbers for the board you are after and when it was refreshed.

 

[Valantar]

actually, it can vary wildly depending on where you source the board. A month or two isn't typically enough time.

Some places can order [insert # here] and sell [insert lesser # here] with the remaining quantity just sitting there. It can take multiple months and even then it is certainly not a guarantee to get one. It all depends on where you buy and its stock numbers for the board you are after and when it was refreshed.

That's true. I was (though in hindsight in no way clearly) talking about the most popular models at medium-to-large retailers with high inventory turnover. That is obviously not the case for every board or retailer out there.

 

[EarthDog]

It can be the same story with huge brick and mortars (I was intentionally generic with my words above). It just depends on how many are ordered, when, and how many are sold/leftover along with when you are buying. All around its a crapshoot, really. In general, the longer the better, no matter where you are purchasing it from.

EDIT: Also, those large B&Ms have to take whatever stock they get from AIBs... if AIBs have a bunch of old stock, chances are they aren't taking the time to flash them either.

PS - I also happen to know the Procurment Manager for components at Microcenter (he's based in Ohio where I live).