AMD's Ryzen CPUs (Ryzen/TR/Epyc) & Vega/Polaris/Navi GPUs

@Vasudev @hmscott

maybe you guys donno this but the gen 4 SSD will all need active cooling. the foot print is to small currently for anything of that speed and it will overheat.

also even with passive cooling may not be enough. i have the optane m.2 ssd and currently using it so i know the heatsink only expands the capacity and offers barely any passive cooling. they need to come up with better controllers and lower those power usage before it'll be any useful in a laptop.

 

AMD "Raven Ridge" and "Summit Ridge" CPUs Won't Work on X570 Chipset Techpowerup.com | May 27, 2019

If you own an AMD 300-series motherboard and are looking to upgrade to Zen 2-based Ryzen 3000 series processors, you have nothing to worry about, as long as your motherboard vendor puts out a BIOS update that adds compatibility (most vendors already have). If you belong to the demographic which has a 1st generation Ryzen "Summit Ridge" processor or Ryzen 2000 "Raven Ridge" APU laying around and looking to buy an AMD X570 motherboard, we have some bad news for you.

 

5GB/s w/o active cooling is asking to sign death warrant. BTW, how did you fit M.2 22110 into your Clevo?
I don't think current M.2 design on jokebooks will suffice to cool these PCIe 4 NVMe SSD w/o boiling. I think MSI and GB might opt for beefier single huge NVMe drive with active cooling in their laptop using air intakes or fans.

 

Intel Replies to AMD’s Demo: Platinum 9242 Based 48 Core 2S Beats AMD’s 64 Core 2S

Intel does its own run of the AMD Computex 2019 demo with the Intel 9242
The company also tells me that AMD was not using the correct NAMD optimizations during the Computex 2019 demo, which is to be expected considering it is a first party benchmark designed to showcase something in the best light possible and you should always take first party benches with a grain of salt. To this end, Intel gave me a run with the optimizations in place and an SKU that consider to be positioned against AMD’s Rome die: ie the Intel Platinum 9242 – a 48 core part.

Not only Intel prefer dirty tricks

-----------------------------------------

AMD Ryzen 9 CPU With 16 Zen 2 Core Overclocked & Tested in Cinebench, Faster Than i9-9980XE – 7nm Ryzen Threadripper HEDT CPUs Still Happening
https://wccftech.com/amd-ryzen-9-16-core-zen-2-cpu-leak-overclock-benchmarks/


-----------------------------------------------------

AMD Unveils Ryzen 9 3900x 12 Core, Shows Off Radeon RX 5700, And So Many Leaks Were Wrong

 

I know the games have begun, but that was fast even for me.

From the link in the quoted post below:

The camera slowly zooms out, shows a lonely figure standing at the edge of a canyon, yells out 'Hey AMD, the ball is in your court again!'. The figure listens to the booming echos fade and then slowly walks away as the sun goes down and engineers all around the world keep working towards the next great launch...

Lol...

Why do I find this so much fun? Because Intel will be forced to lower their prices or else really give us the performance I know they're capable of offering soon(er).

 

I did some math that should be relevant to the discussions today:

Edit: Also, on the Intel comparison, what they also don't mention is Intel having a TDP of 350W+, that those 48-56 core parts REQUIRE, not suggest, WATER COOLING, and that they have a significant advantage on memory bandwidth (which we have to be honest about, because they have 12-channels per socket rather than 8-channels, which is impressive if you need memory bandwidth).

Just some stuff that WCCFtech left out of that article.

 

My opinion:

For those with a high-end X470, unless you are looking to upgrade to a 16 core, for god sakes, don't get the new X570 boards.
- It's clear to me, Gigabyte, ASRock and Asus x470 top boards were not made for 2700x. It's absolutely overkill for 2700x 3 times over. When AMD announced Zen2 would be on AM4, I am absolutely convinced they made these x470 boards with Zen2 support in mind. Our boards were made for Zen2 up to the R9 12 core.
- If you are interested in building a AMD machine from Intel with Zen2, if you can get a top high end x470 on sale, I probably would get that over the overpriced x570 monsters that are coming.

Here are the reasons why:
At minimum, the top x470 boards from Gigabyte, ASRock and Asus can handle 400 amps. Nearly 2x what a 12 core R9 will pull. (Sorry, MSI x470 boards aren't even worth mentioning. Their B450 were great value, but that's it).
PCIE 4.0 will make no difference. Currently even for a 2080Ti, no difference between PCIE 3.0 x8 and x16 @ 1440p or higher. You may get insane read/write speeds on your NVME, but I already got 3000 write speed NVME, so couldn't care less.
With Zen2 on X470 I expect we will be able to overclock our ram to 3866 stable. The difference between 3866 and say 4.2 ghz DDR4 in games will be nill. I'll be shocked if you get more than 1 FPS difference.

x570 chipset is definitely impressive. It really is. But I'm just saying, with current hardware and current state of software, there are zero reasons to upgrade to from x470 to x570 unless you need a 14 phase monster for 16 core overclocked to 4.4 ghz all cores...
- Gigabyte x470 Gaming 7 - 10 + 2 phase w/ 40A chokes with overkill heatsinks
- Asus Crosshair VII - 10 + 2 phase w/ 60A chokes
- ASRock Taichi - 12 + 2 60A chokes

Among these 3 boards, with the 3800x I don't expect VRM temps to reach even 70C. Seriously, I truly believe for those of us with those boards I mentioned, 3800x with PBO won't even consume half of the available power.

So please, IMO, save your money. There are no benefits to upgrading to X570 if you got a top end X470. My suggestion, upgrade your x370 or x470 board when the next generation of boards come out that support DDR5.

 

First and last please stop this talk about fanboyism. You as others know I only look on performance + overclocking.
I can't see anything in my posts that should offend you. I post info found from the internet. Nothing more or last. So please stop with it. Thanks

I haven't shot down AMD's new Ryzen line of processors. So this can be said.

But look at these numbers below for 3800X. Then compare it with todays 8 cores Intel 9900K. 2% lead in CBR-20, Yeah not much in front. And will the same numbers for Multi-Thread show up when we see 9900KS out? 3rd gen 8 core Ryzen is brand new arch whitch help AMD perform better. Compare it with old Intels old 14nm++ is somewhat lopsided/crooked/skewed. But still... Can still not crush Intel's old 8 cores in performance. They need 12 cores for that. And none know how they really will overclock. Still to early for that type info.

 

On Epyc that still is a bit scarce on info, are these both air cooled and a host or other questions too.

 

That was a repost to someone else on another forum. Sorry for the confusion, it was NOT directed at you. I just did not want to retype my analysis (but I should have edited out those parts). That is completely my fault and I do apologize. That was not directed at you.

Now that that is cleared up, we can get to the substance. Intel's single core goes up about 2% per 100MHz. As such, with the 9900KS having a 5GHz all core boost (300MHz over a stock 9900K), we can assume the single core boost has been raised an equivalent 300MHz at stock (not talking overclock yet, although I don't believe there is much in the tank left for overclocking considering LUUMI showing his R0 revised stepping hitting a wall at 5.4GHz due to heat in his location, otherwise 5.5 for benching, meaning 5.4 may be a more common daily driver). That would give Intel approximately a 5% faster single threaded performance at around 5.3GHz single core boost, in theory.

Now, Intel's all core boost going from 4.7GHz to 5GHz is about 6.4% improvement, which would put it closer to a 41% improvement from their baseline 9700K. That means AMD would need a little more than 4.5GHz all core OC to catch it at stock.

We will have to watch, but if the delta is 6% in performance but 21% extra cost (if priced the same as a 9900K), then there is a question.

 

Intel already disclosed their 48-Core and 56-Core AP chips require water cooling in servers! CLCs have already been designed for them in 1U and 2U parts.

Moreover, Rome chips do not require water cooling that I know of, but water blocks ARE being used with them in both the Shasta and Frontier supercomputers.Theoretically, if rome was not watercooled there, it could boost higher for longer and could change the equation.

Also there is a 100W difference in TDP, IIRC.

 

get an adapter. and re route to 2ndary gpu slot as I only have single GPU so there are space. look up era-adapter there are some cable. also grab m.2 extender from bplus as well.

but yea the optane run damn hot, about 2w more than my crappy pm961, the performance is worth it though.

thats way too nice for intel. for 48 cores+, that 100w will easily turn into 200w as soon as you overclock, then 300w, then 400..

 

just found my cooler with clearance of 132mm of s24t.

Thermalalright Silver arrow 130
Noctua U9S 2 fans

both perform very similar to Noctua D-15 which will not fit as it is at 165mm. +/- 2% in cooling capability.

if I can get two fans on silver arrow 130 then it'll likely be the best air cooler.

 

awwww sheeeeet

https://twitter.com/BitsAndChipsEng/status/1133044083772088322

 

Interesting


AMD will need 5.0GHz chips in 2020.

Thanks. +rep

 

Well, just going off official specs of AMD @240W and Intel's official 350W. We all know both companies can blast past that, just that, in the end and overall, that Intel machine will be sucking down at minimum 100W per server.

 

I would hope AMD with the Epyc comparison is trying for comparing a server that is just heat sink cooled, for theirs and the Intel system. If you are two throw everything out the window Intel would probably dominate but you would have just one system per rack and the rest being power supply and cooling solution. So I think it is safe to say for the physical environments Epyc is designed to run in it is the performance leader.

 

Don't even know how to do it on my BGA because it has only single CPU and GPU.

 

I should also note, even if a Zen 2 system is 10-15% faster than your current new Intel system you probably should not grade. I mean if your current system is working for you, do not fix what is not broken. Now there are the exceptions where small increases warrant the upgrade anyway, but they are few and far between.

 

how to make the difference bigger to justify an upgrade:

- patch mitigation 1
- patch mitigation 2
- patch mitigation 3
- benchmark and look at the poor performance compare to ryzen 3000
- upgrade

@tilleroftheearth intel 10nm 18% IPC WOOHOOO so where are the benchmarks..

 

If I get rid off Intel compiler optimisations used by all SW vendors then my Intel CPUs is barely any better than AMD Bulldozer by using generic code path compatible with all architectures. I ran CB R15 and scores were disappointing so I won't post a screenie. A high end Skylake H chip is performing close to Intel Pentium chips even with more cores, so you can come to a inference the scores are pretty bad.

 

damn. dont patch those security issue may never affect you.

aside from intel claim their new 10nm has 18% IPC rather than performance, i'll wait till I see the product before i believe it. on the other hand i saw something interesting, Intel Performance Maximizer.

it looks like a nice tool to check things for the user, kinda sad that it comes this late but it can be useful for their future gen cpus for sure. I believe AMD already has PBO2 which kinda helps you boost all core semi efficiently to an extent for the user but less control. maybe @ajc9988 can share more light on this POB2 vs Intel Performance Maximizer.

 

You have to remember too, IPC on the 10nm is probably with all the software mitigations in place and now there are hardware mitigations on the 10nm. makes that 18% look not all that good IMHO.

 

There is that nagging little issue with Intel CPU's losing performance due to security mitigations in the firmware and the OS, and applications, and VM's, and adding the need to disable HT completely - with total performance dropping 5%-40%. Sometimes even worse with high interrupt activities like IO slowing down even more.

The AMD CPU at 10%-15% + 5%-40% becomes 15%-55% faster + PCIE 4.0 storage speed increases, so it does make a lot of sense to upgrade from Intel - depending on the application and desire for secure computing to migrate off of Intel to AMD Ryzen 3 / ThreadRipper 3 / Epyc 3.

Some of the "in hardware" mitigations are merely integrating "burning in" the onboard firmware the patches that would normally be loading via the BIOS or OS.

Intel says there are some vulnerabilities that have been eliminated through hardware in the newest 8th and 9th gen - check the stepping of your CPU to be sure - as testing on available steppings has been proving that not to be the case - and some vulnerabilties still need OS mitigations still reducing performance.

Mr.Radar
"Originally posted by aphysically
" One of the Intel writeups said that their 8th and 9th gen processors have hardware mitigations, but my 8th gen processor defaulted to the same "Clear CPU buffers; SMT vulnerable". Is it safe to disable the MDS mitigations on 8th and 9th gen processors? Will less aggressive mitigations be possible on those in the future?"

Only certain (very recent) steppings of the 8th and 9th gen CPUs have hardware mitigations for the MDS vulnerabilities. My brand new (under 1 month old) work laptop with an i7-8750H has an 'A' stepping CPU which is vulnerable, only the 'C' and 'D' steppings aren't vulnerable. Cascade Lake has one stepping vulnerable to two variants of MDS and two newer steppings that aren't vulnerable to any variants."

And, here's an example of an application that needs application changes along with the firmware (hardware) and OS mitigations for protection:

"Some operating system mitigations will also require changes in Chrome which we shall include in subsequent Chrome releases. Users should ensure their version of Chrome is always up to date."
https://www.chromium.org/Home/chromium-security/mds

Hopefully all other browsers and development branches stay patched, and other vulnerable applications - that need mitigation in addition to the CPU hardware / firmware + BIOS + OS - will be fixed and updated versions rolled out. Likely as not some applications will remain vulnerable.

I can't find the other info I was looking for, if I find it I'll update here later...found it:
http://forum.notebookreview.com/thr...atches-and-more.812424/page-127#post-10916142

 

These ASRock AMD Navi Radeon RX 5000's look pretty cool, still too early to declare performance levels or pricing, but they are doing their best to share what they have and can show off, and there are a bunch of photos in a gallery in the article:

ASRock Shows Off Concept AMD Radeon RX 5000 Navi Graphics Card Designs – Two Variants In The Works, a 180W & 150W SKU
By Hassan Mujtaba, 2 hours ago
https://wccftech.com/amd-radeon-rx-5000-navi-gpu-7nm-asrock-two-variants-report/

 

First, wanted to point you to this post I just wrote:
http://forum.notebookreview.com/threads/intels-upcoming-10nm-and-beyond.828806/page-14#post-10916234

Next, the tool IPM is closer to Nvidia's overclocking utility than it is to PBO2 or XFRx. To start, it is done in software, similar to how DIP5 and other auto overclocking software is done. It is able to use the per core overclocking of Intel's CPUs, it is able to test each core similar to AMD's pre-binning and marking the best core per CCX and overall (except their implementation is done on a hardware level, not software identification), and then it can overclock the single core boost and all core boost while monitoring the heat output and thermal headroom, along with checking voltage fluctuation and delivery.

Now, there is a question on implementation. For Nvidia's software overclocking, manual tuning still gives better results. For AMD's solution, it is so good that generally even the best OCers mention using it instead of a manual OC because it is so close it doesn't matter. For Intel's software implementation, we will have to wait and see how good it performs.

That isn't a perfect explanation nor a deep dive, but hope that clears some things up.

 

Moving this back to the AMD thread because I have nothing to add on Intel. If they do this, considering the Zen 3 Epyc I/O die is rumored to also be on 7nm+, then it means that the TR I/O chiplet would use the Epyc implementation, not the Ryzen I/O chiplet, meaning it will have a divergence between the two. It also means that if the rumors of HBM on I/O chiplet are true, the HEDT TR would also be able to get that as another memory layer, which is exciting and expensive. Still worth it IMO!

But, because the mainstream will only have PCIe 4.0 and DDR5, I would expect that there be at least two I/O designs next round. But we likely won't find out more on that until Hot Chips in August.

 

Head's up for sign up to attend the E3 2019 AMD Event / Livestream:
https://www.amd.com/en/events/e3

AMD “Next Horizon Gaming” Streamed Event from E3 2019 to Showcase Next-Generation Gaming Products and Experiences
AMD fan event to unveil next-generation products that will power new PC, console and cloud gaming experiences for years to come; leading developers to demonstrate never-before-seen game content
SANTA CLARA, Calif. 05/14/2019
https://www.amd.com/en/press-releas...aming-streamed-event-e3-2019-to-showcase-next

" AMD (NASDAQ: AMD) today announced it will hold an event and livestream during E3 2019 in Los Angeles, California to unveil the next generation of AMD gaming products. Hosted by The Game Awards creator Geoff Keighley, AMD “Next Horizon Gaming” will be held at The Novo at L.A. Live on Monday, June 10, 2019, at 3 p.m. PT.

At Next Horizon Gaming, AMD president and CEO Dr. Lisa Su will present to a live and global streaming audience details about upcoming products and technologies that will power gaming from PC to console to cloud for years to come. The event will also feature appearances by leading game developers who will provide exclusive looks at some of the most anticipated new titles of the year.

AMD fans, community members and press are invited to attend Next Horizon Gaming at E3. Attendees can register for the event here starting today.

The live webcast of the event can be viewed on the AMD YouTube channel and Facebook page. A replay of the webcast can be accessed a few hours after the conclusion of the live event.

• Learn more about AMD activities at E3 here
• Follow AMD on Twitter at @AMD
• Become a fan of AMD on Facebook"

 

You know, AMD could have used one chiplet right now for Navi iGP and have it function for up to 8 core/16th Zen 2 line (since each chiplet has 8c).
But, if the 7nm+ gets I/O shrunked as well, it might be easier to implement a dedicated on-chip memory with large bandwidth that the CPU and iGP could use... although, I'm not sure if that would work.
There will be space gained on 7nm+, so they could use that to put Navi iGP into a chiplet with a certain amount of video memory... but don't think that HBM would fit on there... would it?

 

So, first is addressing why they could not just put an 8-core and a Navi on the existing I/O die. Then I will address HBM on package.

For the existing I/O die, it does NOT contain the necessary components for an APU. Specifically, you have no HDMI or DP on the I/O die. Because of that, even if you slapped down a Navi chiplet on the package, it wouldn't matter because you don't have display out. Now, you could use it as a supporting role for a dGPU, but it defeats the purpose of buying an AMD APU at that point. But, once they have a new I/O die for APUs, then they could easily slap down a Navi iGP and do it. Remember, the 3000 series APUs are based on Zen+ and Vega.

For the mainstream chips, there is a question whether they will use 7nm processes for the I/O. If they don't need to, they may use 14nm. Because server market has more margin, and because it is costly to design on 7nm, with or without EUV, there is still an open question.

Now, for HBM, you can actually put the stack on top of the I/O die or on the die to use it, like GPU die chiplet. The first Navi iGP will likely be the size of the Zen 2 chiplet, making it quite small compared to the size of HBM2. Moreover, the I/O chiplet COULD have the HBM2 stack placed right on top, and the I/O die runs COOLER than the Core die or the iGP die, meaning less worry about thermal issues related to the memory.

Once you shrink the I/O die onto 7nm, that may not be the case for mainstream. But, the I/O die then could be the size of the Zen 2 core die, meaning you can fit a LARGER but lower clocked iGP where the I/O chiplet used to be placed, then place the HBM2 on there. Or, once HBM3 and loco HBM (low cost HBM) hit the market (expected in 2020), then that could lower the cost of implementing HBM on package, allow for higher bandwidth, and you can use system memory to back it up and keep it fed.

Just toying with the possibilities. What they decide to do, no one knows yet!

 

coretek made it clearly amd wont be competing high end GPU market and it is byproduct of their enterprise computing gpus. if it comes out good and can trade blows with nvidia, great, it just happens to be that way. all of this is somewhat still fine til nvidia moves to 7nm, which they mentioned they have no need to, milk continues with their overloaded inventory.

 

I thought tomshardware was supposed to be anti-AMD and an Intel shill? They just gave AMD an award for best CPU in Computex 2019, over the Intel 10nm 2in1 laptop CPU's and the Desktop 9900KS CPU, very nice.

Best of Computex 2019: Overclocked with Innovations
by Tom's Hardware News Team May 29, 2019 at 3:30 PM
https://www.tomshardware.com/picturestory/873-best-of-computex-2019.html#s5

" Best CPU: AMD Ryzen 9 3900 X
AMD’s Ryzen 9 3900X sets a new high watermark of 12 cores and 24 threads for the mainstream desktop, easily beating the Intel flagship’s eight cores and sixteen threads. The 3900X’s performance encroaches on the high end desktop space, but you can access it for a surprisingly low $499 and pair it with a new X570 motherboard, or even drop it in as an upgrade on some existing chipsets.

The chip comes packing the 7nm process, new Zen 2 microarchitecture, and plenty of performance, but next-gen connectivity is also a big attraction. Third-gen Ryzen’s support for the PCIe 4.0 interface unlocks faster SSDs and paves the way for PCIe 4.0 graphics cards, like AMD’s upcoming Navi, making it a great forward-looking upgrade. -- Paul Alcorn

Read more: AMD Unveils 5 Third-Gen Ryzen CPUs, Including 12-Core Flagship "

Wow, 2 awards from Tomshardware for AMD related announcements at Computex

Best Motherboard: X570 Aorus Xtreme
https://www.tomshardware.com/picturestory/873-best-of-computex-2019.html#s8

"We saw well over 30 motherboards based on AMD’s new X570 chipset at the show, but Gigabyte’s top-end X570 Aorus Xtreme easily wins best of the show, for a few key reasons. It’s the only board we’ve seen that manages to passively cool the power-hungry chipset, thanks to a design that turns the metal on the front and back of the board into a massive heatsink.

The X570 Aorus Xtreme’s 16-phase Infineon digital VRM and eight-layer PCB should lead to excellent overlocking. And at $599, while it’s certainly expensive, it’s downright affordable compared to some X570 flagships we’ve seen. Yes, we’re looking at you, ASRock X570 Aqua. -- Matt Safford

Read more: Gigabyte Shows Off Six Aorus X570 Boards for Ryzen 3000 CPUs "

This 3rd award I'd count as shared with AMD, as it's their Ryzen 3.0 CPU PCIE 4.0 interface + Navi GPU PCIE 4.0 + x570 PCIE 4.0 that drove the need for new Gen4 PCIE 4.0 SSD's!

Best SSD: Gigabyte Aorus NVMe Gen4 SSD
https://www.tomshardware.com/picturestory/873-best-of-computex-2019.html#s12

"The PCIe 4.0 era is upon us, courtesy of AMD’s third-gen Ryzen processors, and new, faster SSDs are paving the path to blazing storage performance. Gigabyte’s Aorus PCIe 4.0 SSD comes with Phison’s latest E16 SSD controller and pushes out 5 GB/s of read performance and up to 4.4 GB/s of write performance. That type of speed is sure to kick up some heat, but the Aorus comes with a hefty solid copper heatsink that won’t break a sweat. The drive comes in a 2TB capacity for $299, while the 1TB and 500GB models retail for $269 and $269, respectively. That’s a fairly tidy sum for the fastest flash-based SSD on the market. -- Paul Alcorn

Read more: Gigabyte's Aorus PCIe 4.0 SSD, Under the Hood at 5GB/s "

 

The only thing I worry about with the M.2 Gen4 socket placements on the new x570 motherboards, is that M.2 Gen4 SSD's like the new Corsair with the huge heatsink - may interfere with plugging in PCIE 4.0 cards next to the M.2 Gen4 socket... it may be that new ideas for where to slide in those M.2 Gen4 sockets will need to be implemented.

The Corsair M.2 Gen4 package looks too tall to fit in the M.2 Gen4 sockets between the PCIE 4.0 sockets:

MSI Unveils the MEG X570 Ace: Black and Gold For AMD 50
by Gavin Bonshor on May 29, 2019 11:00 AM EST
https://www.anandtech.com/show/1439...570-ace-motherboard-black-and-gold-for-amd-50

 

Gigabyte Shows Off 15 GB/s PCIe 4.0 SSD Techpowerup.com | May 29,2019

With AMD chipset based motherboards like X570 and next-generation Ryzen 3000 series CPUs delivering the 4.0 version of the PCIe protocol for consumers, products based on the faster protocol are bound to take advantage of its improvements - especially in terms of better bandwidth.

At their Computex booth, Gigabyte showed off some pretty impressive results of their PCIe 4.0 based SSD card. The so-called "Auros AIC Gen4 SSD" is a performance monster. It is able to pack up to four PCIe 4.0 M.2 SSDs that can go up to 2 TB of density each, for a maximum 8 TB total storage. Put them in RAID 0 configuration, though, and you will get some amazing performance numbers. Gigabyte demonstrated speeds exceeding 15 GB/s in sequential reads and writes, providing much more bandwidth than what has ever seen before ina consumer-geared product. The SSD comes with an aluminum shroud with pre-applied thermal pads to facilitate heat dissipation. Additionally, there is a blower fan attached to the card to keep a constant flow of fresh air, which seems like a must if you're packing four M.2 drives inside a tiny aluminum case.

 

Wow, thanks, that's awesome. So maybe the chipset threshold maximum is 15.5GB/sec? Or, maybe the individual SSD's are less than 5GB/sec each?

I like how the 4x M.2 PCIE 4.0 enclosure looks like an AMD "Blower" Style GPU.

Discuss (19 Comments) - there are some insightful comments on the article.

kapone32
"I was reading that the chipset for X570 can support a maximum of 3 NVME drives but I also am thinking that they used an expansion card in the 1st PCI-E slot because unless I am mistaken it is the only place to put in a 16 lane expansion card. Unless the PCI-E lanes have grown from 16-8-4 to 16-16-4."

I wonder if that should read can support a maximum throughput from 3 NVME PCIE 4.0 drives? 3x 5GB/sec = 15GB/sec...adding a 4th drive shares the total bandwidth, increases the storage size but not the throughput.

Aorus PCIe Gen 4.0 M.2 SSD (x4) = 15GB / s! (for $ 1900) Computex 2019
bit-tech
Published on May 29, 2019
Aorus is making headlines by RAIDing up four PCIe Gen 4.0 SSDs on an add-in card

 

https://www.aorus.com/AORUS-NVMe-Gen4-SSD-2TB Up to 5000 MB/S. Probably reach a threshold.

 

I do not think wew will see the 16 core AM4 for5 a while yet. Thinking back if the new 7nm can pull 4,200 on CBR15 then it significantly out performs the TR2 at the same core level. This could cut well into those sales.

So we probably will see the 16 core AM4 around the same release date as the TR3. Maybe I am wrong here but this sounds like the most logical path for AMD to me.

 

thats small heatsink compare to what intel wanted the m.2 optane 905p have. front side back side, heatsink and thermal paste on memory chip LMAO.

probably a small reason to it. primary concern for AMD is that all 8 core chiplet must go towards server/enterprise, so no 16c until their biggest customers are satisified which means all the 8 cores ryzen 3000 CPU are all two chiplets at 4/8 cores working, no single 8 core die.


@ajc9988 @tilleroftheearth new gaming king, ryzen? https://www.chiphell.com/thread-2001351-1-1.html

4.2ghz 6c vs 8700k turbo to 4.6ghz and still beats intel in frame rate, pretty damn amazing. im hoping to see the review from Anandtech/Tomshardware where they compare Ryzen 3000 vs pre and post intel mitigation patch.

 

According too https://izodnews.com/2019/05/27/amd...-up-to-4-6-ghz-pcie-4-0-coming-7-7-anandtech/ the chips, other than Ryzen 9, are single chiplet. I think this is why she showed the dual chiplet right before announcing the 3900x.

 

single chiplet would most likely be not possible. only single chiplet of say 6/8 core for the 6 core variant make sense if all 8 cores chiplets are used towards eypc.

since i have my doubts about the article guess we have to wait until some overclocker ( der8auer cough) to delid and test bare die etc to find out. otherwise they'd have other reasons for using single chiplet 8 cores, maybe lower latency etc, this should also be interesting to see 8 core single chiplet cpu vs 8 cores (with 4 disabled from 12c) dual chiplet cpu performance and if there are any advantages.

 

ok, more of same, up too 8 core single chiplet;
https://www.techpowerup.com/news-tags/Ryzen 7 3700X
https://www.mytravellinguide.com/five-cpus-12-cores-for-499-up-to-4-6-ghz-pcie-4-0-coming-7-7/
https://www.tomshardware.com/news/amd-third-gen-ryzen-7nm-launch-intel-cpu,39449.html

Edit; what I find interesting is that I doubt there will be both dual and/or quad chiplet TR3, so maybe the 16 core will be avoided and only 24 to 32 core? if latency is not severe could there be a up to 64 core eight chiplet?

 

Am I the only one disappointed with Zen2? Zen+ already crushes Intel's consumer/gaming offerings for multi-threading, but it definitely showed with a good graphics card, 2700x was a bottleneck, not by much, but 9600K @ 5ghz consistently got between 5-20 FPS more depending on the game.

I was really hyped that 3700x would not just catch up but exceed Intel. But all we saw was the 3700x was equivalent of a stock 9700K. But we all know that 9700K can overclock like a monster. And we all know Zen architecture leaves VERY little room for overclock. And then it became known AMD's demonstration was fudged, disabling certain features that Intel excels at, I don't know the details, but Intel had legitimate reasons to be upset with AMD's demonstrations.

The hype for Zen2 was too good to be true. We expected 3700x to be a 12 core with 4.5 all core boost @ 329. Instead we got a 8 core LOW POWER offering with all core boost of approx 3.7 ghz, so pretty much the same as 2700x, granted with 15% IPC boost. It seems for AMD, what seems too good to be true is truly too good to be true. I'll have to wait for real benchmarks before making my buying decision. I don't trust AMD hype right now.

 

Overclocking depends mostly on the manufacturing process. UArch has some influence but not much.
Intel refined 14nm so many times that its possible their CPUs can overclock that high, whereas AMD was using a node designed for low clocks and mobile parts... But AMD actually surpassed intel now on 7nm

On IPC grounds alone, a zen 2 with 500mhz lower clocks than Intel achieves same or better performance... And lets not forget power consumption/efficiency in this equation as well.
Even an overclocked zen2 might be able to surpass an overclocked Intel while still consuming less power.

If you are that interested in overclocking, wait for independent reviews to see how high zen2 will overclock. If vega 7 is any indication, it may be able to overclock better than zen1 and +.

 

As far as features, I think this was with the Epyc comparison. Servers typically do not OC, and rarely do we want water cooled and/or refrigerated servers. so I take those complaints with a grain of salt or two.

I am not sure yet where Zen2 will be when overclocking. History does show AMD is not great at that but it is a new process and now TMC vs GF. This too is a wait and see.

As far s further clocks and cores, this may be later down the line. As I speculated they need to not sabotage their own TR2 sales with over doing performance out of the gate. Again a wait and see.

So in the end it all is a wait and see what we get and what we end up with after everything settles....……...

 

There could be several reasons due to AMD dropping TR from their roadmap. None of which would necessarily indicate they didn't want to usurp previous TR performance with similarly performing zen2 with much less cores... Mainly because as new generations of hw emerge, they are more or less bound to surpass previous generations.

Anyway, I don't think that it matters too much because AMD said they will release TR zen2.

 

I do not doubt their word I just like it in writing (roadmap). As far as next gen usurp the prior, this is a different case as their are clock increases, IPC and core count increases. A lot to stuff into one upgrade generationally.

 

I think AMD means, if developer use Ryzen specific compiler optimisations they might be able to exceed the advertised IPC. i think none uses them and most of them stick with Intel compilers because its very fast.

 

fixed for you!

 

No, Intel compilers gives very heavy analytics of how your program exploits all cores performance and also access regions. You can get real time hotspots and detailed bug check and fixes. Once optimised the code is 10x faster than MSFT on Intel only while on other platform it runs 20x slower. So, developers targeting multi platform use gcc or MS compilers to achieve a balance.

 

this is an area i dont go into, since i dont code.

intel had vast majority of market catering to their cpu for nearly a decade. i think any software out there right now that has been using intel compiler will continue to use it.