What will the difference be between Zen 2 and Zen 3?

better than Intel, who'll still be on 14nm

im waiting for ryzen 4

Either will last you a long time. But after zen3, there's very little reason to hold out for more. Pull the trigger in 2019 or 2020.
Zen3 will probably be to zen2 what ryzen was to zen+.

I'm totally out of the loop for AMD processors, can someone explain the difference between Zen and Ryzen? Is Zen like for servers or something?

zen is ryzen.
No.

Zen is the architecture, Ryzen is the marketing name for the consumer parts.

Oh okay, thanks guys

No it wont.

Yeah Threadripper is worth it especially if you're planning on having the same processor for 25 years. The 64-thread model with a TDP of 250W will last you a very, very, very long time.

Just wait for Zen 5 bro

isnt like ryzen 7 out

but isn't threadripper insanely expensive to maintain? I mean sucks massive amounts of electricity compared to ryzen

Attached: Power.png (1294x1257, 46K)

Meh, upgraded from a q6600 at 3.6 to a 2600, overated as fuck. Keep your q6600 until 3nm or something.

>threadripper is straight up double ryzen at same clocks
>shocked it uses more power

You astound me user. Also: zen is extremely power effecient - particularly when clocked around 3-3.3ghz.

Attached: zen fmax.png (1509x905, 55K)

it'll be a refinement over zen 2
just like zen+ was a refinement over zen 1

Well yeah that's the point of having a 250W TDP.
Protip user. Computers run off of electricity. The more electricity a circuit uses, the more capable that circuit can be. Threadripper chugging 250W means that Threadripper can do a shit ton more work than a crappy intel chip running at 40W.

>cheaper on the electricity bills which is a good reason to wait 2 more years desu.
Is this nigger for real

maybe he lives somewhere where power costs 25 cents/KWH

Zen 2 will be new 7nm node shrink. Zen 3 will be the optimization phase for that.

sounds dissapointing

i was told its like day or night

i just want many vst instances at once on flstudio, faster sony vegas etc

>4 times higher mt performance
>disappointing

Get a fully equipped i7 3770 dell optiplex from ebay for 200 bucks. You can get one with at least 8 gigs o ram and an ssd for that price. Add your soundcard and a 1050 ti and you will be fine until something actually worthwhile comes out. Just get a VERY nice productivity boost now on the cheap then you can wait comfortably for far longer.

>16-core CPU with SMT2
>only 257W
Holy shit that's the same power draw of the FX-8350 (and my shitty overclocked Steamroller APU)

Zen 3 will be a refined uarch. Each numbered iteration will be a new refinement. Zen+ was an unplanned product. When zen was launching Papermaster gave an interview and said Zen's development map would be "tock, tock, tock" referencing Intel's tick-tock, with tocks being new architectures. At the time we vaguely knew about zen 2 and 3, I dont think it was officially confirmed but mentioned to be in the works at the time of the interview

>What will the difference be between Zen 2 and Zen 3?

Nobody knows yet, but will probably be a minor bump more akin to Zen -> Zen+ than Zen+ -> Zen 2.

Assuming the worst case of Zen 2 being just 12c chips shrunk to 7nm with zero other novel features, potential Zen 3 improvements in could be any of:
- 16c chips
- Wider AVX
- PCIe 4.0
- 4 MB L3/core
- Larger L1 and/or L2
- USB 3.2+
- random enterprise RAS/security shit

>1800x 65w in idle
I don' t think so

Attached: nigger.png (1721x835, 426K)

AMD literally said that zen 2 improves upon zen on multiple dimensions, it's not just a die shrink with small tweaks

>improves upon zen on multiple dimensions
yeah X and Y
it's just a die shrink dude

Total System Power != CPU power

...and still have better IPC.

...for muh gayms

It's not. Zen2 will be wider and have more cache at the very least.

>AMD literally said that zen 2 improves upon zen on multiple dimensions, it's not just a die shrink with small tweaks

Are you illiterate? OP question was Zen 2->3, not Zen 1+ -> 2, or Ryzen 2000 -> 3000.

Zen 3 will have DDR5 and PCIe 4.0/5.0 support.

Nah, Zen 3 has been stated to be the end of the line for AM4/TR4/SP3 platform and will still be DDR4 even if it gets a bump to DDR5.

Zen 5 (assuming 4 gets skipped due to superstitious Chinks) will be the DDR5 platform and -maybe- get PCIe 5.0 support. 33+ Gbps IO speeds are still a pain in the ass to get working over more than a few inches of quasi-shielded copper stripline in a PCB, so take nothing for granted there.

Yes wait. Wait till you're dead.

>even if it gets a bump to DDR5.

bump to PCIe 4.0 I mean, fuck.

>PCIe 4.0
Do any cards even use the full capacity of PCIe 3.0 yet? I was under the impression that PCIe 3 as a whole was still more or less snake oil.

AMD's mobile APUs are dead-on-arrival next year. The next-gen mobile Nvidia GPUs and the eight-core Coffee Lake refreshes are going to wipe out any hope of AMD competing in their target enthusiast laptop/high-end workbook market. AMD is going to get stuck in the budget-tier laptops for at least the next two years until they can up the core count and change GPU architectures away from Vega.

Here's how the playing field will look like in 2019
>AMD Ryzen 3000 APUs
>4 core/11CU Vega on 12nm+
versus
>Intel Coffee Lake Refresh U
>up to 8 core on 14nm++
and
>Nvidia low-end mobile GPU (GTX 2050 Ti)
>better than GTX 1060 mobile performance, but with 80% of the power consumption
>power consumption puts it UNDER Vega11 on 14nm

The only market AMD can compete in from late 2018 to 2020+ is the mid- and low-end Core i5 CPUs and sub GTX 2050 Nvidia mobile GPUs. They have no hope of ever catching up to Intel or removing Nvidia as a threat to the budget market.

Video cards dont, but PCI-E SSDs sure as hell do.

>AMD has 90% of the IPC of Intel
>somehow thinking AMD won't be improving their IPC in the next two years
You're a real Brane Genyis arent you?

The early adopters for new PCIe speed ratings are always enterprise NIC vendors, followed by enterprise SSD vendors. It's the only way to get multiple 100 GbE/200 GbE/... ports on a single card, since PCIe 3.0 x16 = 16 GB/s = 128 Gpbs, before overhead.

Mellanox started selling PCIe 4.0 cards before the spec was even finally approved.

Coffee lake refresh U will be 400+ for a chip, meanwhile the actual competition for AMD from Intel will be Dual Core shitheaps with the same GPU from 3 years ago.

Low-end and embedded SoC market doesn't look too great either. The upcoming Intel Atom-based SoCs will be as fast as the lowest end dual-core non-SMT Ryzen embedded, but with a significantly lower power consumption. Ryzen embedded is still too expensive, hot, and power-hungry for the tablet and low-end laptop market (which is why we haven't seen a below $300 Ryzen laptop based off of either the dual-core Ryzen APUs or Ryzen embedded).

For IoT devices, AMD has almost zero market share and it will stay that way for the time being, because the x86 instruction set is not optimized or nowhere near as power efficient as ARM64 for this part of the market. For x86 devices, Intel is literally the only option because of it's power efficiency, lower heat dissipation, and better range of very low TDP SoCs. AMD stands no chance of regaining this critical and exploding part of the market.

Talk about being retarded.

First off all. Amd laptop cpus are better than Intel. And with more cores.

And Vega low power is just as good as similarly priced nvidia.

But the benefits are still there because the cpu and GPU or ok the same ccx.

Even the APU is considered better.

>Coffee lake refresh U will be 400+ for a chip
No, the price will go down to $300 and the $380+ CPUs will be the domain of the eight core SKUs.
>meanwhile the actual competition for AMD from Intel will be Dual Core shitheaps with the same GPU from 3 years ago.
That is changing. Core i3 Us will be dual AND quad cores using (ironically enough) Radeon "Vega" GPUs and low-end Nvidia mobile GPUs. It's the Nvidia GPUs that will close AMD's marketshare in this part of the market, because the GTX 2050 replacement will soundly beat the Vega 11 AND use the same amount of power with a quad-core Intel U CPU as an entire Ryzen 7 2700U package.

>Amd laptop cpus are better than Intel. And with more cores.
That hasn't been the case since this year. Core i5-8250Us are better than the high-end Ryzen 7 2700Us while costing ~$30 less, but it's main problem was the Intel iGPU. With the cheaper Core i3 quad-core CPUs, you will now see Ryzen 7 2700U and the Ryzen 7 3700U competing against budget Core i3-9300Us with four cores and GTX 2050M that will use about the same amount of power as an entire Ryzen 7 2700U. AMD has no response for this shift in the mobile market until 2020, when the Zen2-based APUs with non-Vega GPUs hit the market.
>And Vega low power is just as good as similarly priced nvidia.
That changes next year with the next generation Nvidia mobile GPUs.
>But the benefits are still there because the cpu and GPU or ok the same ccx.
You have no idea what you're talking about
>Even the APU is considered better.
You know what, your entire comment belongs in the trash

>Mellanox started selling PCIe 4.0 cards before the spec was even finally approved.

check out pic related nig-rigging. You can get a full 200 Gb on either of the two ports on an existing server if you're willing to burn two x16 slots. The internal twinax dongle (or secondary PCIe port) apparently doesn't do PCIe 4.0, so you still won't get full bandwidth on both ports even with a future PCIe 4.0 host.

The primary purpose of PCIe 5.0 in practice is to allow single-port 400 GbE cards to exist in a few years.
800 GbE/1 TbE is still in "who the fuck knows how we'll ever make this work" territory on both the network and host connection sides.

Attached: gfx_03820.png (900x645, 436K)

>Zen 3 ill have 6 core CCX
Who said this?

besides MSI marketing their current boards are compatible with '8c+' in their latest PR video there is no confirmation for more cores on Zen2, not to mention Zene3

You forgot to mention that maybe ASF will finally become a thing, or something like it will pop up not very thrilled to have one specific usecase where intel is massively better because TSX.

Or they will continue to be the most popular because of intel's stanglehold on the market. I know currently the zen apus aren't that much better than the competition but the big thing for them is that they're SOCs with cpu, graphics and io all on one die.In theory it should be cheap shit for laptop manufacturers, it's intel that gives the extra incentives by bundling cheap displays, ssds and whatever else.

Don't know how intel plus nvidia discrete is going to go below 15w either.

>i dont know how any of this works
these threads are just full of retards fuck

lol

Attached: untitled-14.png (674x800, 53K)

Went from a Q6600 to a i5 2500. Same shit.

are you retarded

AMD is edging close to Intel's Coffee Lake IPC with Ryzen+. Ryzen 2 will for sure be matching, if not, slightly higher than their current performance with the node size shrink from 14nm to 7nm.

What? The only difference between Zen and Zen+ that can improve IPC is L2 cache latency going from 17 to 12 cycles.
Intel is still superior when it comes to IPC.

wait, I heard it's gonna be better than bulldozer

A whole 2% faster.

Except that's not true. Not true at all. Gemini Lake it's a non-existent increase over the last general in actual performance. That shit family is cucked as fuck.

Only a 15% difference in IPC from my 4770K. Intel WTF are you doing?

The super low end is kind of pointless because the margins are so thin. Intel already retreated from the smartphone market and I predict they'll do the same in IoT.

x86 is simply carrying too much decades old baggage (need to spend too much silicon to maintain backwards compatibility which is completely unnecessary for IoT) to be efficient in IoT.

>not even a 1% gain in IPC between the 1800X and the 2700X
wow AMD, way to Jew your customer base

Don't forget to post 240p benchmarks to prove your superiority as your house burns down.

get a power meter for total system power consumption, no system monitoring software will be able to tell you. and you won't be able to compare with any chart posted here or elsewhere, btw. 65W idle is freaking low for a system with a ryzen 1800x and a GPU and a SSD and some HDDs.
this is probably a stripped down bare-bones system for the sake of comparison.

"The intel CPU gets 20 extra frames per second (5250 vs 5230) on zsnes running f-zero!"

dying.

>threadripper is just a chink conspiracy, it isn't real, Intel still wins on workstations and servers, it isn't real

>6core ccx

What an absolute retard. *maybe* AMD will design 8core ccx once they hit insane core counts and are hitting bandwidth limits due to the amount of nodes. but thats so far in the future that even AMD doesn't know the details.

>1950X
>16 cores
>~250W

>7820X
>8 cores
>~250W
lmao

>What? The only difference between Zen and Zen+ that can improve IPC is L2 cache latency going from 17 to 12 cycles.
>Intel is still superior when it comes to IPC.

Zen arguably has stronger internal integer IPC potential than newer *-Lake chips, by way of a more robust scheduler/dispatcher. Where things fall apart (besides AVX) is everything related to memory latencies, in not just CCX-to-CCX and from DDR4 but also in read bandwidth and alignment/bank aliasing constraints on L1D loads to register. Haswell and Broadwell actually but a lot of effort and silicon into removing the various cycle dings for loads on all kinds of corner cases, which Zen 1/1+ has not yet begun to try to attack. For some benchmarks with conveniently patterned read accesses, it is very much already possible for Ryzen to beat out Skylake in IPC already.

t. HFTanon

clock improvements
cache improvements
latency improvements
ipc improvements
memory controller improvements
xfr improvements

>get a power meter for total system power consumption, no system monitoring software will be able to tell you. and you won't be able to compare with any chart posted here or elsewhere, btw. 65W idle is freaking low for a system with a ryzen 1800x and a GPU and a SSD and some HDDs.

>Review CPU's
>measure total system power
>user shows exactly how pointless it is by showing that the cpu is around 1/3 of system power.
>(you) get upset

Attached: 1527690603812.jpg (494x345, 29K)

>What an absolute retard. *maybe* AMD will design 8core ccx once they hit insane core counts and are hitting bandwidth limits due to the amount of nodes. but thats so far in the future that even AMD doesn't know the details.

Do you even understand the range of reasons why it would be advantageous for AMD to move towards a 6c CCX now rather than later? Zen has very fast intra-CCX latencies and poor inter-CCX/Infinity Fabric latency, so it behooves them to keep as much traffic between cores and cache within a CCX as they can manage, which means bumping CCX size for their 12c parts if they can manage it instead of adding more CCXs. And considering that 6c is finally the biggest consumer market, they probably want to be able to sell 6c APUs without 25% of the cores fused off of 2*4c parts.

nah

You are genuinly retarded. Ccx latencies are low DUE to only having 4 cores. It is designed to be copy pasted for mobile to desktop and server market. AMD only had money for like one mask. Redesigning the entire thing because you want AMD to mimic intels bingbus is retarded.

When was Moore's "law" ever a law?

That will be Zen 2

Has to be a gaymer fag

Just wait long enough till Chinese dump crash the RAM market.

Intel 8080 -> Intel 8086

that's literally it

You're out of your intellectual and professional depth here.
Full Zeppelin L3 read latency is ~40 clocks, and the L3 PLL matches the clock of the fastest active core in a CCX. Going to a 6c design will bring them nowhere near the 70-80 clocks of Skylake-E/-X. Skylake gets its huge latency bump for any request being routed from the core to the mesh (2 transitions even for a "local" L3 slice hit), not for mesh hops, and Zen gets whacked from the high-latency global crossbar traversal and HA processing.

Could you tell me, specifically, where in the internal crossbar design you would argue even a 10 cycle penalty would accrue from moving to the wider design?

Ryzen 4 is zen3

>When was Moore's "law" ever a law?
Until about 2003, in every meaningful way, taking Pollack's Rule (linear IPC improvements cost quadratically more transistors) into account.
Technically Moore's Law is about transistor count not just speed, so multi-core kept that going for a while longer, but real end of the line was stumbling with new lithography tech in the last 6-8 years.

Attached: integer-perf.png (556x454, 72K)

It means your can use fewer lanes and still get the same performance.

So it's not a "law" but an observation

this