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

So, the contact point of the sensor is right between the die and the IHS. It is set to read 20 degrees higher, but, it is a matter of informing and having the monitoring software accurately convert it for reporting. Now, it is a question of when the common software for monitoring was made aware of the reporting and when the fix comes out.

Meanwhile, with the mention of the X399 chipset (approximately 4 months away if rumors are correct, meaning that it will have closer to the server chip for an HEDT platform (also means better motherboards) which will be released either the month before or the month of Intel's X299 platform for a direct head to head), I am glad I waited, as there is a chance for the better comparison for my needs. So, patience does get rewarded at times (also, potentially having a higher TDP 8-core for the other platform). Time will tell, but it is exciting.

 

Seems like an error or lack of communication between AMD and everyone else, I think I even heard that CPU temperatures read in motherboard BIOS screens were even wrong. Still seems simple to me, can't understand why they didn't have it sorted for launch.

EDIT: yeah, a quick google shows that temperatures reported in BIOS screens were even wrong. Simple communication error between AMD and everyone.
https://www.kotaku.com.au/2017/03/amds-ryzen-cpus-arent-running-as-hot-as-they-say-they-are/

 

You would think. The reason I believe HWMonitor64 is running correct atleast on my system is the built in MSI command center temperatures read exactly the same and that is enough for me to believe they correct.

They seem correct as well. Running low 60's max with 1.25 Vcore and a Noctua NH-D15. Sounds about right to me compared to other platforms.

I hear ya, I did not want to wait as I wanted a new system but this one will last me a while and heck if I don't like it I can toss the MB, sell the processor and buy a new one not a huge loss.

 

Does yours read the same temperatures in the BIOS screens? NH-D15 is a beast though, maybe you're in the 40's! ;-) (and 1.25 Vcore is low, so I'm conjecturing it as a possibility).

 

The bios reads the same as well when idle.

 

That is what I'll do with my recent skylake build. But it being relatively new is exactly what gave me the patience. But I intend on building something much more powerful, meant to last a long while, with upgrade path and then an ability to skip for awhile a round or two after those upgrades.

Sent from my SM-G900P using Tapatalk

 

Makes sense. I have another desktop with a 5820K setup but my wife uses it a lot more recently so I really wanted a additional desktop.

 

Well maybe you are in the 40's rather than the 60's, so unless you've got an updated BIOS that reads the sensor correctly then I reckon that's the case.

 

Well I guess that would be better. I do have the latest bios they put out but did not really read through the notes to see what all they changed.

 

Would be interesting to know what those release notes say, because if you're in the 40's that would be great! I've got the NH-D14, it's a great cooler, and you have the NH-D15, they're both very similar in performance - in the world of air coolers, they're a beast!

 

Yes, there is really no reason to get a water cooler with these sealed radiators unless you are going to design your own water cooling system as they put out around the same or higher temperatures than the NH-D14 and NH-D15.

 

That's right, they both have the same performance as the Corsair H110 for instance:
http://www.guru3d.com/articles_pages/corsair_h110_review,12.html
http://www.guru3d.com/articles_pages/noctua_nh_d15_review,10.html

 

Yup, and many other coolers. That is why I have never bought one of those. I did have my own water loop with D4 pump and 2 - 240mm radiators a few computers back but would never personally chance a leak for the same performance as a air cooler.

Still have the 3rd best overclock on overclock.net with this CPU. Shows you really don't need water to max out this CPU. Kinda sad, wish you could get a little better overclock out of it.

https://docs.google.com/spreadsheet...IKBE3ATCDRqVRpAHFsKnieEncv0/edit#gid=87938175

 

Don't presume that this is AMD's fault though... it might not be.
I had Intel Dual Cores in my laptop which were incorrectly reporting temperatures through third party software until the software was patched.

 

I don't believe so that was just max in CPUZ but I have been running 4.0 at 1.3625 on air for a while now, not sure if the temps are accurate or not but have had not has a crash yet and everything has been very stable.

I have some newer memory now since that last run. 2 sticks instead of 4. Just ran it again but was unable to get any higher. The next clock in the bios is at 4225 but just crashes, so this is as high as I can get for now.

http://valid.x86.fr/a6q1j0

 

16-core Ryzen part would reportedly run at 3.1 GHz base and 3.6 GHz Turbo - Guru3d.com Sorry if this is posted before

"Last week we reported on a new rumor that AMD is working on a high-end desktop 16 core 32 threaded Ryzen processor series. These processors initially have been reported with a low base-clock in the 2.4 to 2.8 GHz range. It's reported that this 16-core and 32-treaded beast would do a 3.1 GHz base and 3.6 GHz Turbo frequency, yep that is like two R7 1700 CPUs"

 

And think about it, R7 1700 has 65W TDP, 2 x R7 1700 means 130W TDP? That's lower than most CPUs for the X99 platform

 

Re-read the link

"This processor would be quad-channel compatible. The TDP for this product would sit in the 150 to 180 Watt range. So basically the package holds 4 CCXs"

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

Edit. More news.
AMD X390 and X399 chipsets diagrams reveal HEDT Information - Guru3d.com

AMD X399/X390

"Then there are two more diagrams showing the X399 chipset for dual-CPU motherboards and X390 for single-CPU motherboard. The first diagram might be the workstation/server platform for Naples CPUs as it has ECC memory support. X390 corresponds to earlier rumors about Ryzen HEDT, which is meant for 16-core 32-thread Ryzen processors. So if you look at the info and those diagrams are a real thing then X390 chipset would support up to 44 PCI-Express lanes. That would be even more than Intel’s flagship Core i7 6950X on X99 platform (40 lanes). Interesting to see is that the dual-socket board is listing the processors as RZ4700, and the single socket version as RZ 2700."

 

Oh what a bummer, I was expecting a real server CPU that runs cooler than Intel's offerings. But still 16C32T for enthusiasts? Intel gets pwned.

 

But AM4 / Ryzen supports ECC RAM already (if the mobo allows it), what's different about X399?

 

My concern is there is enough issues with the latency and just 2 CCXs so will having 4 CCXs compound the latency issues?

 

Likely yes. There is another layer bounding the two dies. (Then with the dual socket system there is another extra layer as well.)

Though you need to keep in mind that at 16 cores and beyond any system has the same problem. This is not limited to AMD or (Ry)Zen.

 

http://www.tweaktown.com/news/56845/amds-high-end-x390-x399-mobo-dual-ryzen-cpus-possible/index.html

Sent from my SM-G900P using Tapatalk

 

Imagine 4GHz with a water chiller!

It depends on how the layout is. There will be two CCXs per side. If they tied the cache together on the two CCXs on the same side, it would actually allow the 8 cores on that side not to take the latency hit. That is why I wanted more info on the chip. If they did not do that, then it is a latency hit per CCX. Time will tell.

The X399 seems to be the multi-CPU, whereas the X390 seems more like the HEDT/workstation boards. There are a lot of unknowns still. Also, the X390 & X399 support higher TDPs, quad channel and above on memory, and a different socket for server based chips. It also has more pcie lanes, allowing for more and higher end video cards (commercial) and NVMe drives.

Sent from my SM-G900P using Tapatalk

 

@ajc9988

Thanks.

I'm aware of the other differences. I was wondering what ECC support means in this context. Does it mean X390 will have ECC locked down? Or does it indicate formal ECC feature validation for server use which the Ryzen 7 doesn't have?

There is also the possibly that someone confused registered RAM for ECC, though it's unlikely a person responsible for such a diagram (if legitimate) would make such a mistake.

 

That is a very good question. I know unregistered ECC is currently supported, but would think if the X390 is being pushed for HEDT outside of server/workstation use, that they wouldn't lock it down (although the 2P X399 might find itself locked down). I believe the latter (ECC feature validation) is likely (or rather I'm hoping). ...



Sent from my SM-G900P using Tapatalk

 

had not seen this posted yet, apologize if it has been;

http://www.pcmag.com/news/352538/ryzen-7-chips-are-locking-up-pcs-amd-knows-why

 

Wow, the comments are fairly harsh at PC world, calling them intelbots, and clickbait. It is one benchmark using Intel Haswell specs, not really news, and somewhat of a misleading title to the article.

 

it is somewhat misleading to say the least.

 

I saw it, but because of content I did not post. Now, there are board bricks occurring in certain specific instances, so I thought I would read it to see if that was discussed. Nope. Literally just talking about a system hang for a single app that is not in wide circulation, meaning it is not relevant to anyone. Meanwhile, the MB bricking could put the MB manufacturers in a bad light, so you don't see articles about those instances anywhere. This is just meant to scare people to Intel (potentially suggesting purposely planted). The headline had nothing to do with the article and was meant to manipulate people that only read headlines!

Sent from my SM-G900P using Tapatalk

 

Covers some BIOS tuning settings I haven't seen covered elsewhere:

OVERCLOCKED AMD RYZEN 7 PERFORMANCE GUIDE


As always, read the youtube comments for more good info

Linus's WAN show has a few minutes of interesting Ryzen info:
00:14:04 - Ryzen's Infinity Fabric Clock Speed is Linked to Memory Clock Speed
00:16:53 - AMD found the root problem causing Ryzen to freeze desktops
00:18:31 - AMD rumored to have 180W 16-core Ryzen CPU to run at 3.1-3.6 GHz
00:21:25 - Level1Techs has 6th fastest 1800X build overall; fastest with single 1080Ti

 

More info AMD Ryzen Processor with 12 Cores and 24 Threads Spotted

"There's not a day that goes by with something Ryzen. We talked 32,16, 8, 6 and 4 core processors, we can now add a 12-core version as well. Hey, when it rains it poors."

 

All kinds of goodness rolling out from AMD. I am ready for one of these in a Clevo or Tornado

 

If this is true, then this suggests it is practically finished silicon (F4 silicon) and it is a Naples cutdown (8x8MB L3). I would have expected 32MB L3 (4x8MB as each CCX would get 8MB). Also, this shows the 24C will be coming. I will still wait, but depending if the cache is linked per side, that could be awesome. Also makes me wonder if higher than quad channel would be possible. But, this could just be an error with the software reading the chip.

Edit: it said ES2, so it just had QS left.

Sent from my SM-G900P using Tapatalk

 

Yeah, if you don't need it now. Wait and see. And Intel will have to run for their money

 

Definitely waiting. If the 16C is $1K and they keep roughly equal price spacing, Intel will need to cut margins drastically!

This makes me wonder if the 16C with quad channel will have 64MB L3 cache or if AMD will have two lines, an HEDT line, then the Opteron Naples line, with the Opteron chips all being Naples 32C with cores disabled, 64MB L3, the over 100pcie lanes, and 8 channel memory.

If the rumors on the speed of the 16C/32T are true, I'd lean toward the two lines. If both are supported on the same chipset, it could be exciting (although the HEDT has less cache and memory channels, you would get extra core speed). It also would mean the X390 chipset would have way more to it. But they may do X399 for all Opteron chips and X390 for the HEDT. Need to read the article again to see if they mentioned chipset.

Edit: it said most likely HEDT and X390. It says it is a desktop chip, not a server chip, but with the 16C being so much faster, it could be those that were not passing binning for 16C, which still doesn't solve the L3 mystery. So much left to find out.

Sent from my SM-G900P using Tapatalk

 

Here, have a look at this:
https://www.techpowerup.com/231473/amd-says-ryzen-1700x-1800x-have-a-temperature-reporting-offset

AMD itself reports that the TJ Max is offset by 20 degrees C on 1700x and 1800x.
Can't say if the same applies to 1700 though.

 

Old news man, we all know 1700X and 1800X have such issue, while 1700 doesn't.

 

This would be nice but I am not sure if I believe mine is off by 20C. It is around 40C at idle which does sound a little high but if it was off by 20C that would put my idle temp around 20C. Seems a little low to me, does 20C idle seem right with a air cooler to you all?

 

Well, you have the same cooler (almost) as me, and my CPU (although 6700K) idles in the high teens in a 20deg C room - PC is located on the floor (colder) + I've heard the temperature sensors on CPUs aren't as accurate at low temperatures - so I reckon you could be idling in the 20's!

 

Inter vs Intra CCX latency doesn't seem to make enough of a difference in gaming tests to explain all of the difference.

Ryzen CCX Performance: 2+2 vs. 4+0

 

AMD Ryzen Quad-Core 2+2 versus 4+0 Core Setups Analyzed - Guru3d.com

"Long story short, your best bet for faster game CPU bound game performance, remains to be overclocking the processor a bit and using fast high frequency memory."

Btw.
Intels Skylake-SP Xeon-processors will get metal based names

Edit:
More news from AMD's competitor Intel Intel Kaby Lake X Core i7-7740K And X299 Platform Details Leak - Hothardware.com

"In a quick performance comparison, it looks like this Core i7-7740K fights a tough battle against AMD's Ryzen 5 1600, almost matching each other in an arithmetic test. The Intel chip does however come ahead in the multi-media tests by quite a large margin, which isn't a surprise, given the clock speed difference between the two, though the Ryzen chip as two more cores at its disposal"

 

I picked my case (Corsair Obsidian 900D) and now I'm waiting for either the Asrock Taichi or Fatal1ty Pro X370 mobos. I'd loved a 10 or 12 core Ryzen CPU, but I don't feel like waiting a few more months unless quad channel memory is confirmed for X390.

One of the reasons why I waited so long is I was in talks with another NBR member about possibly buying their X99 rig but the talks fell apart.

 

Not surprised at all. While synthetics could show up too 10% difference real world is another story . It is about what I would expect and that is where threads are not optimized for dual CCX there is up to a 3% hit but that is about it. This is even more true where the CPU is not being flooded.

This also means stop getting hung up on synthetics. Go by real world results and tests. Once the platform settles in a bit and bugs are sorted out things will change for the better. So as they say best not to be a first adopter some times.

 

I will let you know if I see one become available online. I know they are tough to get ahold of.
Edit: I don't know if you live near a microcenter but don't forget to check them out if you do. They have both those boards listed in Houston but not in stock.

 

So, the X390 was confirmed after the first leak. Go back to the website here that showed schematic type illustrations for both X390 and X399 motherboards. I think it was on there it showed quad channel (which is slightly better than the rumors of the 16C). I would say it most likely does support quad channel.

Also, I recommend the wait. The reason why is you wanted the x99. In August, you'll be able to choose between the AMD X390 and the Intel X299. Both platforms will last until 2020+. Both support future CPUs. X390 should have quad at least, X299 will have hexa-channel. Literally, once I found out about the HEDT AMD setup, and you get side by side comparisons from both companies at the same time, same life span. Just wait...

Sent from my SM-G900P using Tapatalk


Edit: if you must buy something soon, wait until May and buy either 1080 Ti or Vega!

 

You make a great argument. I am/was willing to buy into the X99 platform even knowing it is dead but not at its current prices.

I have a 6900K workstation exclusive for work but for my personal HEDT, I do want something that is going to last a couple years (I am notorious for upgrading to the latest and greatest GPU's every year but I keep the same CPU for two to three years).

I know AMD has stated they are going to support AM4 for awhile, but I am curious if they are going to do the same for X390/X399 since they are different chipsets. At least with Intel, I know I'll have another CPU series to upgrade to if I ever need the extra IPC.

My ideal personal setup is 10c/20t CPU up to 4.0/4.2, 1080 TI SLI and a huge assortment of SSD's. So I shall wait but I am already picking up certain components.

 

First, this uses the socket for the server products. As such, the X399 will be and the X390 likely will be.You cannot do that to the server industry, which is so powerful they force Intel to support two generations on the same chipset. Next, Zen 2 &3 will be supported on AM4, so it is very likely for the larger money maker you would do it.

Finally, I'll wait for Vega. The early silicon, clocked to 1200 (with final clocks at 1500+) already scores around 1080, so with later silicon, it should be a brawl with the Ti. I'm sure the Ti will do some things better, but I'm betting Vega well do well with others. But time will tell....

Sent from my SM-G900P using Tapatalk

 

http://www.tweaktown.com/news/56891/amds-12c-24t-16c-32t-cpus-called-threadripper/index.html

So, more info on the HEDT platform, thread ripper! 64 pcie, up to 32MB cache... No confirmation on memory channels in this article, but 100GB/s on infinity fabric (on Naples with 8 channel memory, they talked about like 176GB/s). Boost of 4.0 (early silicon at 3.7). TDP of 180 (that's a hot chip, water cooling likely needed). 12C with TDP of 140.

ES cinebench R15 of 2500!

Announcement at computex.

Sent from my SM-G900P using Tapatalk

https://www.custompcreview.com/news...works-features-using-multiple-dies-mcm/40188/

"Since combining multiple dies on a MCM will result in a significantly larger chip package, the upcoming 12-core and 16-core processors will utilize a LGA SP3 socket, which is a 4000+ pin socket."

Compare this to the rumors of double the size of Intel's 2133 chips.

https://www.techpowerup.com/231911/amd-16-core-ryzen-a-multi-chip-module-of-two-summit-ridge-dies

So, remember, this socket will be used for YEARS and may have to accommodate more cores in the future.

 

I like it how all of the expectations and disappointments that everyone was having (Low pinned socket, quad channel memory, low PCIe lanes, etc), beside overclocking, are being checked by new leaks and announcements. AMD is delivering AF, and that's what we all want to really see happening!