Cheaper Zynq Ultrascale+ and new Artix Ultrascale+ devices announced!

[SiliconWizard]

As I noted a good while ago already, the 7 series is still very under-represented on the market as far as dev boards go

What kind of devboard for 7 series would you like to have, which doesn't exist yet? I'm really curious.

Well, especially Spartan-7-based boards. Which are still extremely few. I know you designed one though!

[asmi]

Well, especially Spartan-7-based boards. Which are still extremely few. I know you designed one though!

OK, this is really interesting. What kind of device would you like to see used, what peripherals/connectors?
I actually thought hard about what kind of board would be interesting to design. I figured that my board covers the low end (if you delete the DDR2, you can do all the way down to the smallest S7-6) to low-ish middle, then there are Digilent's Arty-S7's, and finally SP701 which has pretty much everything you will ever need on a board. That's why I never actually went on to design anything with these devices. The only thing that I can think of as really missing is a board with SODIMM DDR3(L) slot so that you can load a crap ton of memory into it.

[SiliconWizard]

Well, the Arty S7 has relatively few IOs broken out. The Xilinx one is nice, but a bit expensive. But that's still very few boards compared to the tens of dev boards one can still find for the Spartan 6...
As I see it, the Spartan 7 just seems not to be taking off like the 6 did when it got released. Maybe it will happen over time, but it's slow for some reason. Maybe it competes with the Artix-7 in a way that makes it look slightly unattractive. Dunno. I do have a couple Artix-7 boards.

As for your own board, it should be fine for my needs. It's just that I haven't had time to build one as of yet. Otherwise, typically something with DDR3 and a comfortable number of IOs broken out would do.

Also am in sort of a dilemma right now, as Vivado stopped working on Win 7 since last version. But I still don't really want to switch to Win 10, besides I have some peripherals that are not supported in 10 either and that I want to keep. Solution would be to use Vivado on Linux. Are you guys all on Win 10 now?

[asmi]

Well, the Arty S7 has relatively few IOs broken out. The Xilinx one is nice, but a bit expensive. But that's still very few boards compared to the tens of dev boards one can still find for the Spartan 6...

If existing boards cover most needs, there is no need for that many. Like I said, when I investigated the market, the only two things I found missing were SODIMM DDR3(L) slot and SYZYGY-compliant connectors for cases when you need some amount of high-speed lines, but don't need so many as to use bulky and expensive FMC connector. And I didn't think that designing such a board would make a lot of sense.

As I see it, the Spartan 7 just seems not to be taking off like the 6 did when it got released. Maybe it will happen over time, but it's slow for some reason. Maybe it competes with the Artix-7 in a way that makes it look slightly unattractive. Dunno. I do have a couple Artix-7 boards.

Artix-7 and Spartan-7 have the same fabric and all hard IP blocks (aside from MGTs and PCIE obviously), so Artix-7 can do everything Spartan-7 can, which is yet another reason there aren't that many S7 boards. S7 excels over A7 when you need a crap ton of IO, which is why SP701 has got so many stuff on it. For example, there are enough pins in 484 package of S7 to implement TWO 64bit SODIMM slots, and still have an almost IO bank for other stuff! 676 one would leave you too full IO banks (100 IO balls) after two SODIMM slots. That's a lot of IO pins!

As for your own board, it should be fine for my needs. It's just that I haven't had time to build one as of yet. Otherwise, typically something with DDR3 and a comfortable number of IOs broken out would do.

My board as-is only has like 20 odd GPIOs, which of course it not very many. I would much rather prefer F256 package, but alas, S7 are not offered in that package.

Also am in sort of a dilemma right now, as Vivado stopped working on Win 7 since last version. But I still don't really want to switch to Win 10, besides I have some peripherals that are not supported in 10 either and that I want to keep. Solution would be to use Vivado on Linux. Are you guys all on Win 10 now?

I use both Win10 and Ubuntu. Upgraded my Win7 installation shortly after Win10 came out, and never looked back since.

[Bassman59]

As I noted a good while ago already, the 7 series is still very under-represented on the market as far as dev boards go

I dunno, as soon as it became clear that ISE and Spartan 6 was at the end of the line we moved to Vivado and Artix-7. We've got quite a few products in the field, and ... we never bought a single development board.

I don't understand the obsession with dev boards. You either design products or you're a hobbyist, and the hobby market is not interesting to Xilinx, certainly not to the point where they'll bother with an Ultrascale-based board to blink some LEDs.

[rounin]

I don't understand the obsession with dev boards. You either design products or you're a hobbyist, and the hobby market is not interesting to Xilinx, certainly not to the point where they'll bother with an Ultrascale-based board to blink some LEDs.

Think of the poor software engineers! I bought a ultrascale+ dev kit at my cost/risk last year to do some contract software development for another company. Never stepped foot in their office or saw their expensive secret prototypes. Cheaper dev kit = more margin.

[iMo]

.. I don't understand the obsession with dev boards. You either design products or you're a hobbyist, and the hobby market is not interesting to Xilinx, certainly not to the point where they'll bother with an Ultrascale-based board to blink some LEDs.

My new hobby board has got 2 User buttons and 8 LEDs for blinking too..

[asmi]

Xilinx is going to have a free webinar on April 6th about these things, hopefully I will be able to make it there and ask some questions.

[gnuarm]

We all will be lucky! I have designed a brand new board based on K325, and I just have ordered the PCB and The FPGA from JLCPCB and my chinese supplier for 100$ each, I hope every thing works as expected. Here you can take a look before I get my hands on this new puppy! I will share the results when I made progress.

It has 2 SFP, 2 Gigabit Ethernet, 2 USB host, 64Bit DDR3, one  SATA, one  4K HMDI, one  audio codec, and one LCD, and a lot of expansion connectors including 2.54mm and FPC's with 10Gbps SERDES for External ADC and DAC

What does it do?

[ali_asadzadeh]

If all things works at the first PCB design rev, It would be a nice General purpose High end toy, for evaluating things like, a custom made CPU with all regular IO, or maybe a DIY Scope, or testing connectivity like Fiber optics and Ethernet packet processing, evaluating a lot of ADC and DAC's before putting them into a new PCB, the possibilities are all with in your imagination.

what would you use it for if you get one?

[asmi]

You've got to stop hijacking this thread
Back to topic - I've made it to the webcast, and here is what they said (aside from a ton of marketing mumbo-jumbo of course):
- The motivation for Artix US+ is that while compute abilities are good, transceivers are out of date. Hence 12.5/16G transceivers.
- The smallest A10 part is going to be roughly at the level of A75T in terms of amount of logic, a15 and 20 will fit between A100 and A200.
- only smallest two parts will have PCIe 4.0 hardware support, top two members will only support PCIE 3.0x8 unless PCIE 4 is implemented via soft IP. The reason being that largest parts were designed before PCIE 4 requirement became a thing
- All Artix US+ parts will be included in free WebPack license (yay!)
- Largest two parts are going into production in Q3 2021, smaller devices - Q4 2021 or Q1 2022
- They are working with third parties to design devboards
- Tools support will came in 2021.1.1, because 2021.1 was already fully booked for other stuff. 2021.1 will come out in June, 2021.1.1 - July
- Smallest Zynq US+ (ZU1) is in pre-prod right now, full production is late Q2/early Q3

[NorthGuy]

- The motivation for Artix US+ is that while compute abilities are good, transceivers are out of date. Hence 12.5/16G transceivers.

DDR3 is getting obsolete too. Many DDR3 parts are already EOLed. They may be difficult to source few years from now. DDR4 support is certainly needed. Although DDR5 is already around the corner.

- All Artix US+ parts will be included in free WebPack license (yay!)

That's a good news. Did they say anything about chip pricing. I guess not

[asmi]

DDR3 is getting obsolete too. Many DDR3 parts are already EOLed. They may be difficult to source few years from now. DDR4 support is certainly needed. Although DDR5 is already around the corner.

Well if you look at the Digikey, it seems that DDR4 is harder to source, while DDR3 is still plentiful. But I agree, DDR4 is a more future-proof solution, and they also can provide higher capacity (once chips larger than 4G will become widely available) at lower price, all while providing higher bandwidth as well. That said, technically nothing stops you from implementing a DIMM/SODIMM slot and buying a module at any computer store and use that instead of soldered-on chips, but that solution has it's own drawbacks (like higher power consumption, or larger physical board size).

They also highlighted that HP pins natively support MIPI up to 2.5 Gpbs per pin pair, infact the case study they presented was a hi-res camera, which also uses their new chip-scale InFO package with 0.5 mm ball pitch. I wonder if these can be utilized as general-purpose differential signals, as latter would open up some interesting options like implementing some serial protocols which were only possible using MGTs - like PCIE 1.0, or DisplayPort RBR, HDMI 1440p, and 2.5G Ethernet.

Also US+ DSPs with 27x18 multipliers running at up to 775 MHz is a serious compute power too, and even the smallest AUS+ will contain 400 of them.

That's a good news. Did they say anything about chip pricing. I guess not

They said preliminary pricing as available under NDA if you talk to your sales rep, but since the parts we are most likely going to be interested in (smaller AUS+) are not even in pre-prod yet, it's too early to say anything for sure. Heck, they don't even have confirmed final pinouts, nor have they done any characterization to see what the real performance is going to be like exactly - hence the lack of official performance datasheets, only marketing mumbo-jumbo about what they are aiming for.

[asmi]

Vivado/Vitis 2021.1.1 is now out with support for Artix Ultrascale+ XCAU25P.

[AaronLee]

Vivado/Vitis 2021.1.1 is now out with support for Artix Ultrascale+ XCAU25P.

Have you tried it yet? I'm still using 2019.2. I found 2020.2 to be too buggy which that made it unusable for me.

[asmi]

Have you tried it yet? I'm still using 2019.2. I found 2020.2 to be too buggy which that made it unusable for me.

I've been using 2021.1 since release, and had no issues, aside from that pesky floating MIG issue (a.k.a. "unable to rename blah-blah-blah: permission denied") which have been happening to me from time to time since time immemorial.

[asmi]

For those who are interested - Xilinx is going to have a free webinar tomorrow, August 5, 2021 at 7:00AM PDT, about Artix Ultrascale+. They will also showcase AU25P Evaluation kit from Opal Kelly. You can register here: https://pages.xilinx.com/EN-WB-2021-08-05-ArtixUltraScale_LP-Registration.html?utm_campaign=artixu&utm_source=xweb&utm_medium=webinar_page

[SMB784]

For those who are interested - Xilinx is going to have a free webinar tomorrow, August 5, 2021 at 7:00AM PDT, about Artix Ultrascale+. They will also showcase AU25P Evaluation kit from Opal Kelly. You can register here: https://pages.xilinx.com/EN-WB-2021-08-05-ArtixUltraScale_LP-Registration.html?utm_campaign=artixu&utm_source=xweb&utm_medium=webinar_page

Jesus, 7AM PDT? I guess they aren't interested in having people from the east coast join that webinar

***EDIT*** Nevermind, I did the conversion in the wrong direction. 10AM is much more reasonable than 4AM

[asmi]

Jesus, 7AM PDT? I guess they aren't interested in having people from the east coast join that webinar

Why? That's 10 AM EDT. I think it's a reasonable time.

[asmi]

Looks like the fabric in these new Artix'es is seriously fast, just for fun I've done a place&route of my rv64i core on au25p speed grade 1, and it was able to close timing on 377 MHz! The same core only closes at 173.9 MHz on Artix speed grade 2. That's over 2x frequency!

[ali_asadzadeh]

Try it on the speed grade 3 to see what we can do

[SMB784]

From the webinar today, see the attached image for the performance comparison between Artix UltraScale+ to Artix7.

The price point, they mentioned, will be similar to Artix7 for equivalently specced parts, so you can expect to pay for these new Artix UltraScale+ parts what you used to pay for the Artix7 parts.

I have also attached the slides presented in the webinar.

[asmi]

Try it on the speed grade 3 to see what we can do

There is no speed grade 3 for AUP devices.

[asmi]

The price point, they mentioned, will be similar to Artix7 for equivalently specced parts, so you can expect to pay for these new Artix UltraScale+ parts what you used to pay for the Artix7 parts.

Yea, that's the great news. I was kind of afraid that they price them way too high. My guess is they are going to be about US$100-120 for AU10P, about US$170 for AU15P, and somewhere around US$220 for AU20P, which is certainly not cheap, but I've built and hand-assembled a number of boards with Artix A75T and A100T for my clients and I'm quite comfortable working with US$150-200 chips, so I'm certainly looking forward getting my hands on these devices and making some cool projects.

They've also announced an Opal Kelly's devboard with AU25P at US$1300 shipping in October. That's a bit overpriced in my book for a device with a price somewhere around US$300, but then it's just about 30% more than Genesys 2 board with Kintex-325T device with roughly the same resources as AU25P (aside from faster transceivers, IO and fabric). But I love that they've stuffed a whole bunch of SYZYGY connectors on this board, so maybe I will buy one - will see. One thing I don't like is that they've just placed a single DDR4 x16 memory device. I would really like to see at least x32 memory interface for the kind of IO that this board has.

Finally, Xilinx typically makes a recording of a webinar available on-demand in a few days after it happened, so if I remember, I will post a link to it once it appears. Unfortunately they didn't answer my question regarding USB 3 5/10G physical signaling support in their transceivers (or I missed it - was in a virtual meeting at work at the same time), so I guess we will see about that too.

[EEexpert]

From the webinar today, see the attached image for the performance comparison between Artix UltraScale+ to Artix7.

Thanks for the news.

For me it's clear now that Artix US is not the replacement for Artix, for some applications US suits much better, but for others Artix is fine.