New Zen microarchitecture details

[PhonakV30]

The SFF I/O options are really a disappointment. Just 4 USB 3.0, no USB 3.1? Just 2x SATA600 + 1xNVMe x2 ? I need at least 6x USB, 3x SATA600 and would prefer a NVMe PCIe 3.0 x4...here's hoping someone will put an A320 into a mini-ITX board.

AMD A320 Chipset (SSF/Budget) - Then there is the A320 and A300 series. These are intended for value, budget and small form factor products. This chipset will offer 4 PCI-e Gen 2 lanes with 1+2+6 (USB 3.1 Gen2, USB 3.1 Gen1, USB 2.0) support. A300 should be the chipset for SSF computers (like tiny home theater builds etc).

So Can you get 6X USB 3.0 from AMD A320 Chipset ?

 

[CentroX]

Wait a sec. Is A12 9800 (excavator) the cpu that ryzen is suppose to be 55% improved IPC? Because i cant find a single gaming benchmark of bristol ridge. I thought it came out last year what happened?

 

[inf64]

Wait a sec. Is A12 9800 (excavator) the cpu that ryzen is suppose to be 55% improved IPC? Because i cant find a single gaming benchmark of bristol ridge. I thought it came out last year what happened?

Yes core Vs core. But Bristol Ridge is a 2M/4C chip with cut down L2 (Vs Steamroller) and no L3, while Zen has order of magnitude faster (and bigger) cache subsystem. Since Carrizo is around 15% faster than Piledriver core, just find any latest CPU review that has FX8350 and you can get a ballpark single thread performance from that. You have 1.55x1.15, a crude estimation of IPC vs Piledriver core and then you have another ~20-25% SMT gain in threaded workloads over that.

That is exactly what CanardPC preview showed us- around 75-80% higher ST IPC than Piledriver core and 1C/2T Zen having around 35% higher throughput than 1M/2T Piledriver at the same clock.

 

[CentroX]

Yes core Vs core. But Bristol Ridge is a 2M/4C chip with cut down L2 (Vs Steamroller) and no L3, while Zen has order of magnitude faster (and bigger) cache subsystem. Since Carrizo is around 15% faster than Piledriver core, just find any latest CPU review that has FX8350 and you can get a ballpark single thread performance from that. You have 1.55x1.15, a crude estimation of IPC vs Piledriver core and then you have another ~20-25% SMT gain in threaded workloads over that.

That is exactly what CanardPC preview showed us- around 75-80% higher ST IPC than Piledriver core and 1C/2T Zen having around 35% higher throughput than 1M/2T Piledriver at the same clock.

So basically 1.80 x fx 8350 which ends up around i7 6700/7700

Much better then haswell as first suggested in this thread.

 

[inf64]

So basically 1.80 x fx 8350 which ends up around i7 6700/7700

Much better then haswell as first suggested in this thread.

Well the thing is we don't know the exact figure until AT and other review sites do the analysis. CPC Hardware on purpose skipped the ST benchmarks because that would easily give away the source who provided them the ES and platform for their test. IPC is not uniform and can vary a lot depending on the workload. I guess AMD cited 40% which now seems to be the lower end of the spectrum and upper being 55%. IPC could be somewhere in between for all we know. Important thing is that with this kind of a jump they are right in the Haswell/BDW territory and just a bit short of Skylake, which is amazing achievement. Now we know that clock speeds are up there as well so they covered all the bases it seems. Now we have to just wait a bit and see what we get in the end.

 

[AtenRa]

EXCLUSIVE INTERVIEW | AMD’s Robert Hallock Spills The Beans on Ryzen - RedGamingTech

https://www.youtube.com/watch?v=mKttIZ3fZjM

Nice thanks,

As i have said before Im aiming for a x300 mini-iTX + 6C 12T 65W TDP ZEN for a SFF system to replace my Core i7 3770K Micro-ATX system.

 

[Sonikku]

Anyone know how well this thing would stack overclocked in singled threaded against my i5 2500? Would it be worth the upgrade?

 

[bjt2]

So basically 1.80 x fx 8350 which ends up around i7 6700/7700

Much better then haswell as first suggested in this thread.

AFAIK Haswell is not so far from BW and SKL... Less than 10% includes all three architectures... So it's simple for Zen forecasts to go from HSW to BW and SKL given the incertainity that we have on the actual gain over CON cores... We can be sure only when we will have the actual benchmarks...

 

[bjt2]

Well the thing is we don't know the exact figure until AT and other review sites do the analysis. CPC Hardware on purpose skipped the ST benchmarks because that would easily give away the source who provided them the ES and platform for their test. IPC is not uniform and can vary a lot depending on the workload. I guess AMD cited 40% which now seems to be the lower end of the spectrum and upper being 55%. IPC could be somewhere in between for all we know. Important thing is that with this kind of a jump they are right in the Haswell/BDW territory and just a bit short of Skylake, which is amazing achievement. Now we know that clock speeds are up there as well so they covered all the bases it seems. Now we have to just wait a bit and see what we get in the end.

The difference between BW and SKL are under 4%. So if Zen has same IPC than BW (as Handbrake, +8%, and blender, +2%, tests would indicate), Zen has to have just few (200) MHz of more clock. And Base clock for the flagship is going to be at least 3.6GHz, +400MHz than BWE base... So the only chips that can best Zen currently are Kabylake with its higher clocks, but only on 4c/8t chips, since we don't currently know if AMD will manage to produce a 4c/8t clocked enough... To best even KBL AMD nees to produce 4.5/4.8 4c/8t chips. And if they are harvested 8c chips I doubt they will manage to do so...

 

[Spartak]

AMD A320 Chipset (SSF/Budget) - Then there is the A320 and A300 series. These are intended for value, budget and small form factor products. This chipset will offer 4 PCI-e Gen 2 lanes with 1+2+6 (USB 3.1 Gen2, USB 3.1 Gen1, USB 2.0) support. A300 should be the chipset for SSF computers (like tiny home theater builds etc).

So Can you get 6X USB 3.0 from AMD A320 Chipset ?

AMD A320 isn't intended for SFF according to the diagram, that's why I ask. A320 has enough lanes to provide anything I want and then some. It's just weird there isnt a middle ground for normal users, it's either bare bones or full featured.

 

[imported_jjj]

Anyone know how well this thing would stack overclocked in singled threaded against my i5 2500? Would it be worth the upgrade?

We don't quite know where it lands vs Intel's current IPC ,for the shipping product but it's gonna be close either way.
Single core clocks could be very interesting. If AMD's turbo is as flexible as it appears and it scales with better cooling.
How it will OC is even trickier to guess, even without OC, the single core clocks could be very high.
If it's worth the upgrade for you, that depends on your needs.
How many cores you can get at 200$+,remains to be seen,might be 6, could be 8 if you go above 250$.

 

[cytg111]

So basically 1.80 x fx 8350 which ends up around i7 6700/7700

Much better then haswell as first suggested in this thread.

Canard was only on record stating, just after, that ST was below Skylakes ..

https://twitter.com/CPCHardware/status/817133286224957443

 

[imported_jjj]

The difference between BW and SKL are under 4%. So if Zen has same IPC than BW (as Handbrake, +8%, and blender, +2%, tests would indicate), Zen has to have just few (200) MHz of more clock. And Base clock for the flagship is going to be at least 3.6GHz, +400MHz than BWE base... So the only chips that can best Zen currently are Kabylake with its higher clocks, but only on 4c/8t chips, since we don't currently know if AMD will manage to produce a 4c/8t clocked enough... To best even KBL AMD nees to produce 4.5/4.8 4c/8t chips. And if they are harvested 8c chips I doubt they will manage to do so...

Why have a 4 cores module and not make a native quad die? Having that modularity and not using it wouldn't make much sense.
Besides, single core clocks could be way high even for the octa SKUs, with the smart turbo.

 

[The Stilt]

Why have a 4 cores module and not make a native quad die? Having that modularity and not using it wouldn't make much sense.
Besides, single core clocks could be way high even for the octa SKUs, with the smart turbo.

$$$

Basically that's why.
Zeppelin is cheap to make, so it would take long to amortize the cost required by additional die designs, mask and packages.

 

[Spartak]

Basically "why".

I understand it means why, but I don't understand why that is such a weird thing to expect, and is coming off so aggressive.

Every 'normal' mini-ITX board for desktop chips has come with 6x USB and 4x SATA for as long as I remember, and that's from when mini-ITX first came to market. I built my first mini-ITX based system already in 2006 and now on my third system.

I connect a mouse, keyboard, printer, scanner and external HDD and need one open USB to connect a smartphone or USB drive. That's six. I use two SSD's and an optical drive and want to continue using all of those on a new system. That's three.

 

[CentroX]

Anyone know how well this thing would stack overclocked in singled threaded against my i5 2500? Would it be worth the upgrade?

Zen is worth it for the new architecture.

 

[Insert_Nickname]

Every 'normal' mini-ITX board for desktop chips has come with 6x USB and 4x SATA for as long as I remember, and that's from when mini-ITX first came to market. I built my first mini-ITX based system already in 2006 and now on my third system.

Since there is no details yet on actual products, it is too early to say anything definite about I/O configurations on ITX platforms. There might be a product that provides what you need.

AM4 I/O is very flexible, and nothing is stopping board manufactures to add additional controllers like they usually do.

 

[cytg111]

Zen is worth it for the new architecture.

- Define these new architecture details that makes it worth it ?

 

[bjt2]

Why have a 4 cores module and not make a native quad die? Having that modularity and not using it wouldn't make much sense.
Besides, single core clocks could be way high even for the octa SKUs, with the smart turbo.

I hope that single core turbo is high, but I don't have much information to be sure of the actual maximum clocks. I can hope of >=4.5Ghz single core turbo (and i was mocked multiple times for even dreamed of this clock), but i am not sure that there are not some other limitation factors, such as power density or Vcore limits that forbid higher clocks...
For 8c dices, since the leakage is very low, there could be little differences between max turbo core with 4c even non native. Harvested dices have the 4c not powered so they are like native 4c, but with higher dissipating area and contact with the heatsink... They can be even better than native dices and have the advantage of use the bad 8c dices... The only disadvantage is that usually broken 8c dices are low quality and so low clocks. I don't expect harvested 4c dices to clock much higher than best 8c, because if they are harvested, probabily they are on a low quality part of the wafer...

 

[CentroX]

- Define these new architecture details that makes it worth it ?

How is it not? 2017 tech vs 2010 tech. 32nm vs 14nm. Ddr4, high speed sata, usb 3.1. Dual core vs quad, hexa or octa, Neural net, more energi efficient etc

 

[Spartak]

- Define these new architecture details that makes it worth it ?

That depends. IPC performance gains wont be that much, and clocks wont be that much higher either. But you could go with either slightly higher performance with a lower TDP (thus more silent system) or you could go with slightly higher ST performance but much higher MT performance if you can make good use of that. I think the biggest noticable gains for you would be in the ability to use a PCIe 3.0 x4 high performance SSD.

 

[cytg111]

How is it not? 2017 tech vs 2010 tech. 32nm vs 14nm. Ddr4, high speed sata, usb 3.1. Dual core vs quad, hexa or octa, Neural net, more energi efficient etc

I dont give a *bleep* whats under the hood as long as im winning the race.
- and if I come in a close second I must really evaluate if the new and shiny is really worth it.

 

[imported_jjj]

$$$

Basically that's why.
Zeppelin is cheap to make, so it would take long to amortize the cost required by additional die designs, mask and packages.

Sure but how much do you need to ship and at what size difference(in mm2) to justify the native die?
If they also use Samsung, they would likely need a new set of masks anyway so they could fab 8 cores at one supplier and 4 cores at the other.

 

[cytg111]

That depends. IPC performance gains wont be that much, and clocks wont be that much higher either. But you could go with either slightly higher performance with a lower TDP (thus more silent system) or you could go with slightly higher ST performance but much higher MT performance if you can make good use of that. I think the biggest noticable gains for you would be in the ability to use a PCIe 3.0 x4 high performance SSD.

First it is all about what Sonikku uses his system for. Money may be better spend on a gpu upgrade instead.
As for SSD's, if you allready have a sata SSD, going for 200k rand iops vs 80k will yield you close to zero benefit in a desktop experience outside of benchmarks.
The only angle I can see is that he may want more than need those ~30% extra ST performance or need more than 4 *real* cores for whatever gama ala watchdogs2++

 

[CentroX]

I dont give a *bleep* whats under the hood as long as im winning the race.
- and if I come in a close second I must really evaluate if the new and shiny is really worth it.

2500k has bottleneck issues, ryzen likely not