Speculation: Ryzen 3000 series

[BigDaveX]

I dont get the excitement honestly, Amd still wont beat Intel on core per core performance, not on gaming performance either so...

Whats so exciting about this release again?

According to AMD's slides, Zen 2 does beat Coffee Lake on core per core performance, albeit just barely. And it's cheaper and uses less power. If AMD's claims hold water, there will be literally zero reason to consider purchasing any Intel desktop CPU until Icelake shows up...

But, admittedly, we should wait for proper, independently-conducted benchmarks before making a final judgement. It's actually kind of funny watching the same people who are calling Intel's slides about Icelake's GPU performance a pack of lies in the comments section of that front page article then turn around and berate people for not taking AMD's Zen 3000 slides as complete, gospel truth in that article's comments.

 

[itsmydamnation]

I dont get the excitement honestly, Amd still wont beat Intel on core per core performance, not on gaming performance either so...

Whats so exciting about this release again?

No the honestly(lack) bit is its obvious bais ( just look at your posting history).

We still also dont know what XFR or what the OC head room looks like, but we do know AMD now have an IPC lead and a significant perf per watt lead and a core count lead, and a NVME performance lead..........

 

[inf64]

I dont get the excitement honestly, Amd still wont beat Intel on core per core performance, not on gaming performance either so...

Whats so exciting about this release again?

AMD just achieved ~15% IPC jump in 2 years, doubled the amount of cores (there will be 16C part on AM4) - meaning they are on par or ahead of Skylake derivative cores in ST and VASTLY ahead in MT workloads (except in AVX512 niche of niches). They did that while burning almost 2x less power per core and having a scalable design that goes to 64 cores with superior SMT implementation (not vulnerable to exploits). Oh and they almost got to clock parity on first round of products. Not impressive at all.

 

[tamz_msc]

Apparently there's a 16C eng. sample doing the rounds -

https://imgur.com/a/aHsWi4u

 

[Head1985]

1.42v on 7nm?

 

[tamz_msc]

1.42v on 7nm?

Could be a misreading. OR perhaps that's the voltage wall. Time will tell.

 

[davidbepo]

I dont get the excitement honestly, Amd still wont beat Intel on core per core performance, not on gaming performance either so...

Whats so exciting about this release again?

better IPC
better perf/watt
better value
almost the same performance

thats exciting, but it seems youre a troll, why do i even bother?

 

[TheGiant]

AMD just achieved ~15% IPC jump in 2 years, doubled the amount of cores (there will be 16C part on AM4) - meaning they are on par or ahead of Skylake derivative cores in ST and VASTLY ahead in MT workloads (except in AVX512 niche of niches). They did that while burning almost 2x less power per core and having a scalable design that goes to 64 cores with superior SMT implementation (not vulnerable to exploits). Oh and they almost got to clock parity on first round of products. Not impressive at all.

we got more coarz on desktop, we got more trolls on forums....you know, moar coarz, less waiting, more time for trolling- that is a very specific use case that can eat all your computing resources
never forget that




Trolling in the tech forum is not allowed.


esquared
Anandtech Forum Director

 

[french toast]

I think that Icelake is still one year away for maisntream desktop parts. The question is wait for it or go for Zen2? It depends on whether you need more cores now or in the near future.

If waiting for Icelake is an option, presumably so is waiting for zen 3..

 

[B-Riz]

Now we now about X570 and the first round of Zen2 CPUs.

I think B550 will bring low power chips, and MAYBE an AM4 16/32 AFTER Intel releases their next thing.

BUT, I don't see why AM4 would get 16/32 because of TR3.

I can see AM5 starting at 16/32 65w $329.

 

[DrMrLordX]

Really? This is a "tock" last Ryzen was a "tick".

Intel would kill for ticks or tocks like that. Seriously, +15% IPC and we're debating whether this is a sea change in performance?

Sure, it isn't like Piledriver -> Summit Ridge, which was over 40% uplift in . . . pretty much everything. But Pinnacle Ridge brought less than 15% to the table per clock when it shipped, and if you wanted higher clocks, you were going to pay for it in power. A 2700x at a fixed 4.3 GHz OC can pull it some serious juice. Matisse is delivering bigger gains than Pinnacle Ridge, and doing it at much lower power.

I was hoping for more. I thought they would crush a 9900k.

The 3900X does, technically. And wait until somebody overclocks one of these 3700X chips. If they hit 4.6-4.7 GHz or higher, look out. All that for $329?

So i am only one that think 8cores are little bit overpriced vs ryzen 2000?Why they comparing 329USD 2700x vs 399USD 3800x?

I think I've posted it this thread already, but I'll say it again: Matisse prices match Summit Ridge perfectly. $499, $399, and $329 were the launch prices for 1800x, 1700x, and 1700 respectively. If you were hoping for AMD to maintain the prices from Pinnacle Ridge, sorry, it didn't happen. I do think they have room for a cheaper 8c product in there somewhere.

s gaming / memory latency performance. If 1/2 if clock option means high ddr4 with g

Games may make it painful to use the 1/2 clock option. Depends on threadhopping between CCXs.

also what interests me is ability to run 32GB of ram at high speeds

Ditto. I'd love to see the latest IMC handle dual-rank DIMMs better than Pinnacle Ridge/Summit Ridge. Interestingly enough, there ARE some performance gains to be had from using dual-rank DIMMs on existing AM4 CPUs, but they are overshadowed by necessary reductions in clockspeeds. This phenomenon has been observed in some benches. If AMD could figure out how to get dual-rank DIMMs (notably 16GB DIMMS) running at the same speed as single-rank DIMMs then we could be in for some interesting performance surprises.

IMO current winner- the new 3600 non X, oced to 4,4GHz+ for 199 = WIN

For the "budget" overclocker, that's a chip to watch. Not sure if it's gonna knock off the 2700x, but it will give hell to some older AM4 chips, that's for sure.

I think that Icelake is still one year away for maisntream desktop parts. The question is wait for it or go for Zen2? It depends on whether you need more cores now or in the near future.

What makes you say that? None of Intel's roadmaps have shown IceLake-S anywhere. Not leaked ones, not official ones . . . IceLake-S is completely MiA.

I dont get the excitement honestly, Amd still wont beat Intel on core per core performance, not on gaming performance either so...

Whats so exciting about this release again?

You'll find out soon enough. Or not, if you keep wearing blinders.

1.42v on 7nm?

Probably a flaky ES.

 

[itsmydamnation]

Games may make it painful to use the 1/2 clock option. Depends on threadhopping between CCXs.

1/2 clock is only 1 extra clock cycle to wait in terms of latency, its nothing compared to the latency saved if it allows significantly higher mem clocks.
obvoiusly then you also have to consider fabric throughtput

 

[inf64]

According to some leaks ~4400Mhz for memory is achievable with good DDR4 sticks. That *should* be enough to feed 16C OCed in MT workloads.

 

[DrMrLordX]

1/2 clock is only 1 extra clock cycle to wait in terms of latency, its nothing compared to the latency saved if it allows significantly higher mem clocks.
obvoiusly then you also have to consider fabric throughtput

Then I say, bring on the benchmarks!

According to some leaks ~4400Mhz for memory is achievable with good DDR4 sticks. That *should* be enough to feed 16C OCed in MT workloads.

Yes, but the question is, is that with full 2200 MHz IF, or reduced 1100 MHz IF?

 

[inf64]

Then I say, bring on the benchmarks!



Yes, but the question is, is that with full 2200 MHz IF, or reduced 1100 MHz IF?

If I'd had to guess I'd say full IF speed.

 

[DrMrLordX]

If I'd had to guess I'd say full IF speed.

If you're right, then I'd be thrilled to push my DDR4-4400 to its limits. I've been underclocking RAM on AMD platforms since . . .2009? It's about time they had a competitive IMC.

 

[PotatoWithEarsOnSide]

That 16c ES at 4.25GHz is pulling 5.5w per core. That's 88w for the cores.
The IF will be drawing a fair amount too, but 16c within 135w TDP looks eminently doable.

 

[Zucker2k]

Yes i9 9900ks is closer in performance than bulldozer was back in the day, but i9 probably consumes more power, has security vulnerabilities, doesn't come with a cooler..fx9590 led in some Mt scenarios. Both desperate attempts to save face, equal in my opinion.

The 9900KS is RO stepping so most, if not all security vulnerabilities have been fixed in hardware. Plus, the FX 9590 has nothing on the i9 9900k. The i9 9900k is still the best desktop processor even at 95watts until anybody can go out and actually purchase a 3rd gen Ryzen 3900x (and possibly the 3800x). We shall soon see.

thoughts #3

pricing is too low by AMD- it shows they know they are now competitive performance wise

from my POV AMD is cannibalizing between r7 2700X and r5 3600X and r7 3700X

since the new chips are banckwards compatible (looking at the reviews really) you can put a brand new r5 3600 (non X), OC it to 4.4GHz+ with b450 (550 maybe) board and give a smile to the 2700X....which is 300 EUR atm
2600,2600X,2700 and 2700X are made completely obsolete by AMD

IMO current winner- the new 3600 non X, oced to 4,4GHz+ for 199 = WIN

Based on both 1st gen and second gen Zen releases, I'd be surprised if there's any more room for overclocking these chips. Remember that chips can be thermally binned, so that headroom is not only about power consumption, but thermal dissipation characteristics of each core/chiplet as well.

It's actually kind of funny watching the same people who are calling Intel's slides about Icelake's GPU performance a pack of lies in the comments section of that front page article then turn around and berate people for not taking AMD's Zen 3000 slides as complete, gospel truth in that article's comments.

This. It has always baffled my mind why this obvious bias, but Intel has failed to hit their targets in recent memory so some skepticism is warranted.

AMD just achieved ~15% IPC jump in 2 years, doubled the amount of cores (there will be 16C part on AM4) - meaning they are on par or ahead of Skylake derivative cores in ST and VASTLY ahead in MT workloads (except in AVX512 niche of niches). They did that while burning almost 2x less power per core and having a scalable design that goes to 64 cores with superior SMT implementation (not vulnerable to exploits). Oh and they almost got to clock parity on first round of products. Not impressive at all.

A full node reduction is always going to provide impressive numbers, power -wise but 8 cores @ 3.3 /4.4 @ 45watts is new territory of efficiency, as it should. The power numbers, if they hold, are going to be Zen 3s strongest suits against Intel's offerings.

 

[mohit9206]

Lisa is raising those ASPs and holding back that 16 core the moment AMD is in the dominant position, heh. Still excellent CPUs, though.

Overall, performance wise, it looks like my expectations are met. No big surprises there. I was hoping the prices would be a bit lower, but there's nothing surprising there, either. I hope all the YTers that were so sure about 5GHz haven't ended up making people disappointed.

They used CB R20, so I'm not complaining. Looks like they got a solid 10-15% increase per clock compared to Zen+ in that. And yeah, it'll be behind in gaming, but I don't care.

Side grade is an understatement. That 3600 will be a little beast.

I am a bit disappointed considering i was excited at the possibility of leaks pointing towards 8C/16T Ryzen 5 and 6C/12T Ryzen 3. Now i don't know how to feel.

 

[french toast]

The 9900KS is RO stepping so most, if not all security vulnerabilities have been fixed in hardware. Plus, the FX 9590 has nothing on the i9 9900k. The i9 9900k is still the best desktop processor even at 95watts until anybody can go out and actually purchase a 3rd gen Ryzen 3900x (and possibly the 3800x). We shall soon see.

Based on both 1st gen and second gen Zen releases, I'd be surprised if there's any more room for overclocking these chips. Remember that chips can be thermally binned, so that headroom is not only about power consumption, but thermal dissipation characteristics of each core/chiplet as well.


This. It has always baffled my mind why this obvious bias, but Intel has failed to hit their targets in recent memory so some skepticism is warranted.

A full node reduction is always going to provide impressive numbers, power -wise but 8 cores @ 3.3 /4.4 @ 45watts is new territory of efficiency, as it should. The power numbers, if they hold, are going to be Zen 3s strongest suits against Intel's offerings.

Do we know that R0 stepping fixes these vulnerabilities? My understanding is the hardware fix would incur it's own performance penalties, probably alot less than software.. But still be Interesting to see.
9900ks is a horrible cpu, that thing is going to push past 200watts easily, probably will edge out a R9 3900x in ST and gaming slightly, but get wiped out in every other area.

 

[jpiniero]

Do we know that R0 stepping fixes these vulnerabilities?

It's the same as Cascade Lake. So not everything but MDS is covered with no hit. Now I haven't seen any Windows gaming benchmarks taking MDS into account, but you can be sure that Intel will want reviewers using R0 as a comparison because AMD is now so close. A 9600KS and 9700KS would make sense, maybe tweak the turbo clocks a bit.

 

[DisEnchantment]

According to some leaks ~4400Mhz for memory is achievable with good DDR4 sticks. That *should* be enough to feed 16C OCed in MT workloads.

However multiple cores have to contend with the memory access arbitration and not just bandwidth is at play.
There are multiple patents from AMD on this topic to mitigate the observed latency due to concurrent access, I am not sure if they made it to Zen2, but they are from late 2017.

 

[french toast]

It's the same as Cascade Lake. So not everything but MDS is covered with no hit. Now I haven't seen any Windows gaming benchmarks taking MDS into account, but you can be sure that Intel will want reviewers using R0 as a comparison because AMD is now so close. A 9600KS and 9700KS would make sense, maybe tweak the turbo clocks a bit.

Mmm interesting, a clever play by Intel if the fixes circumvent the bloated patches and don't incur a performance deficit themselves, Yea they would want the reviewers using those, as that would show coffeelake still edging out the Matisse on St and gaming.
Efficiency will be hilarious though

 

[inf64]

Threadripper 3000 lives!

https://videocardz.com/newz/amd-you-will-definitely-see-more-threadripper-from-us

PCWorld took part in the post-conference meeting with Dr. Lisa Su. AMD CEO confirmed that Threadripper is not canceled.
AMD CEO Dr. Lisa Su:

“You know. it’s very interesting, some of the things that circulate on the Internet—I don’t think we ever said that Threadripper was not going to continue—it somehow took on a life of its own on the Internet,” Su said, speaking to a small group of reporters following her keynote. “You will see more [Threadripper] from us; you will definitely see more.”​

When asked about the core count she replied that Threadripper 3000 should definitely receive more cores, as the mainstream is going up.

“If mainstream is moving up, then Threadripper will have to move up, up—and that’s what we’re working on,”​

My guess is (late) Q3 with a new updated chipset and up to 48 cores (staying on the safe side), keeping the base clocks similar or slightly higher while driving boost close to 4.7Ghz. That thing will be a monster. 50% more cores with ~2x better AVX performance (~3x combined uplift versus 2990, YIKES), better memory management and no NUMA quirks.

 

[SPBHM]

a 12 core that performs like this for MT and is not slower for gaming/ST (and can work on more mainstream motherboards) sounds really good, this also probably means they can make a 32 core TR that is not badly limited by the ram configuration. (maybe even one with more cores, given the chiplet+I/O design)

I think we've been spoiled by the cheap 1700s and whatnot so a new Ryzen 8 core for $399 doesn't sound that exciting, but the IPC and clock gains are certainly nice, and their pricing looks decent compared to Intel,

for now the cheaper 6 core with OC sounds like the better deal for gaming/mainstream use.

it's not very clear for me, but will the 12 core CPU work with some let's say, B350 board with bios update?
the prospect of an upgrade of this level, just with a new CPU if you built a system in 2017 with a 1600 or something to a 12 core Zen 2 is pretty nice, moar coars, moar clock, moar ipc, not really moar watts