AMD's Ryzen CPU's (Ryzen/TR/Epyc) & Vega/Polaris GPU's

Discussion in 'Hardware Components and Aftermarket Upgrades' started by Rage Set, Dec 14, 2016.

  1. hmscott

    hmscott Notebook Nobel Laureate

    Reputations:
    5,001
    Messages:
    17,582
    Likes Received:
    21,599
    Trophy Points:
    931
    Dell goes on Epyc server journey with AMD
    Rackmount server competition for PowerEdge 14G Xeon SP rackers
    By Chris Mellor 6 Feb 2018 at 15:49
    https://www.theregister.co.uk/2018/02/06/dell_goes_on_epyc_server_journey_with_amd/

    "Dell is producing one- and two-socket rackmount servers using AMD Epyc processors alongside its Xeon SP server family.

    The Epyc processor is said to be faster than equivalent Xeons.

    There are three PowerEdge AMD-powered servers: the R6415, R7415, and R7425. These accompany the PowerEdge 14G R640 and R740 Xeon SP servers in the Round Rock company's server portfolio, and they inherit general PowerEdge management and feature goodness.

    As context, the Xeon-powered R640 is a 1U, two-socket system while the R740 is a 2U, two-socket one.

    R6415
    The R6415 (tech specs PDF here) is pitched as an edge computing server and is a 1U system supporting one AMD Naples socket SP3 compatible processor with up to 32 cores. It has 8 x 2.5-inch SAS or SATA drives, or 4 x 3.5-inch ones, or 10 x 2.5-inch NVMe (SSD) drives. These latter drives are not hot-swappable, as is true with the R7415 and R7425 servers.

    According to the reference guide, it can have up to 1TB of DRAM with a single processor. A slide we have seen says it can have 2TB:
    [​IMG]
    We have asked Dell which number is the correct one.

    R7415
    The R7415 (tech specs PDF here) is a mid-range single-socket server. It's a 2U system supporting one AMD Naples socket SP3 compatible processor. The box can hold 8 x 3.5-inch, 12 x 3.5-inch, 12 x 3.5-inch + 2 x 3.5-inch (rear), or 24 x 2.5-inch SAS, SATA, or NVMe drives (up to 12 SAS/SATA/NVMe + 12 NVMe drives.)

    Memory tops out at 1TB again (as per tech specs) or 2TB (slide deck), and Dell has been asked for clarification.

    [​IMG]
    Dell R7425 schematic. Click to enlarge.

    R7425
    This system is pitched at high-performance computing workloads. It's a 2U, 2-socket R7425 (ref PDF guide here) can have a pair of Epyc CPUs, up to 32 x SAS, SATA, or nearline SAS hard drives or SSDs, and up to 24 NVMe flash drives. This is the big box of the three, with a maximum of 4TB of memory.

    Performance
    Dell quotes AMD numbers comparing Epyc to Xeon Skylake systems at equivalent cost points:
    [​IMG]
    The numbers show Epyc processor systems delivering better price-performance than Skylake ones.

    HPE on Epyc trail too
    Dell competitor HPE produced a Gen 10 DL385 server using 1 or 2 Epyc CPUs in November last year. That meant 32 or 64 cores and a maximum 4TB of memory.

    At the time HPE said an Epyc dual-CPU 7601-based DL385 system scored 257 on SPECrate2017_fp_base (throughput) and 1980 on the SPECfp_rate2006, both higher than any other two-socket system score then published by SPEC.

    A Xeon DL380 with dual Xeon Gold 6152 22-core CPUs scored 197 on the SPECrate2017_fp_base in comparison.

    Presumably the Dell Epycs will do likewise compared to the equivalent Xeon ones.

    Prices: The R6415 starts at $2,179.00, the R7415 at at $2,349.00, and the R7425 at at $3,819.00. Systems are available and shipping worldwide."
     
    Last edited: Feb 6, 2018
    Raiderman and ajc9988 like this.
  2. ajc9988

    ajc9988 Death by a thousand paper cuts

    Reputations:
    1,233
    Messages:
    5,083
    Likes Received:
    7,117
    Trophy Points:
    681
    I was reading through to catch up but had to address this directly. The leaked bench shows 15% while only having 200MHz increase. That means the extra performance came from other changes. Period. Stop being so ignorant when you have been corrected already!
     
  3. Deks

    Deks Notebook Prophet

    Reputations:
    1,001
    Messages:
    4,174
    Likes Received:
    1,400
    Trophy Points:
    231
    I was going off the technical spec sheets provided for the manuf. process.
    And I also mentioned that the 'leaks' we got are just that... leaks that could be engineering samples running on smaller bump ups.
    Did AMD confirm only 200 MhZ increase for the final product?
    Did AMD also say anything about an IPC increase for refreshes or other changes? Other changes like what specifically? Better implementation of Infinity Fabric, bigger caches, what?
    What kind of changes did AMD actually put into the refresh?
    And is that SiSandra benchmark even the real deal?

    I'm being ignorant for asking questions because the information doesn't exactly mesh well?
    We got 1 leak... nothing that's been substantiated by AMD thus far (correct me if I'm wrong though), and which could be an indication of an engineering sample (or it might not).
     
    Vistar Shook likes this.
  4. ajc9988

    ajc9988 Death by a thousand paper cuts

    Reputations:
    1,233
    Messages:
    5,083
    Likes Received:
    7,117
    Trophy Points:
    681
    I detailed it damn near a month ago. They did tweaks to lower memory latency, which can indirectly boost IPC, but I don't consider that an IPC boost, per se. They also worked on memory speed and capacity, but don't know the effect. I said on January 19th "AMD said they made changes to SenseMI, to the IMC and memory latency, die shrink, power efficiency, and clockspeed and overclocking. We have seen 15% on early silicon, which 6% is only attributed to the speed bump, with the other 9% to all other changes to the CPU, which would include IPC." This was responding to your post responding to my showing the sisoft performance. So, unless we are discussing another leak I do not know about, the statement stands and you need to listen when others speak.

    So, just looking at raw data, that 200MHz gave 15% and more is due to tweaks than the speed boost, which part is the throttle mechanism changes to hit higher for longer if using less than full MHz on all cores, some is the memory tweaks which are indirect IPC improvements, and the die shrink, clock speed, and OC is process changes. Power efficiency comes from both process used/die shrink and from the senseMI changes.

    So we do not need to know whether engineering sample or not, because if it is an ES, then the ffinal will likely be better. Also, 200 seems right for an "official" clock change, especially without being an X series. Look at Intel, on their CPUs for official clocks, they have small increases like that on official speeds also. It means nothing on OC, like I told you two and a half weeks ago. As to IPC, you said there were likely none. It is too early to say, but if no changes occurred, you couldn't get 15% increase over the prior gen on a 200MHz boost in speed which is only a 6% increase in speed. It is mathematically impossible.

    Now, whether the bench is real is up for debate. We'll find out. But don't wait 2.5 weeks then try saying the same stuff that was already addressed. Makes you look like a troll (not saying you are, saying it is unbecoming behavior).
     
    hmscott and Raiderman like this.
  5. Deks

    Deks Notebook Prophet

    Reputations:
    1,001
    Messages:
    4,174
    Likes Received:
    1,400
    Trophy Points:
    231
    I might have missed your reply before... or simply forgotten about it.
    6% clock bump aside, could AMD have realistically done enough changes to Ryzen refresh to get extra 9% without touching the clocks beyond a minor bump?
    Though they mentioned that Ryzen 1 was their 'worst case scenario'.
    And, you do realize you are basing those numbers on a sandra bechmark that may or may not be real in the first place and makes it as speculative as anything else (given that we received 0 formal confirmation from AMD).

    Do we have any official information from AMD itself on Ryzen+ new stock and boost clocks though?
     
    ajc9988 and Vistar Shook like this.
  6. ajc9988

    ajc9988 Death by a thousand paper cuts

    Reputations:
    1,233
    Messages:
    5,083
    Likes Received:
    7,117
    Trophy Points:
    681
    Now this is the conversation I can engage in! So, memory latency on Ryzen versus the mainstream chips still using ring bus was easily 2.5x what Intel's was if you had to jump CCX. Compared to the mesh on Skylake-X, the latency was negligible, literally like 25%. So, to fix the ram latency, we could see a good boost (based on 3200MHz ram). Now, that could account for a couple percent (look at the speed boost going from 2400 to 3200 on the platform, which was not just ram speed, but also faster IF and lowering latency by 25 percent).

    So, that is part of the answer, but not all of it, unless they really did a number on lowering that latency. Now, it is on a new process, which requires some tweaks. Those tweaks could account for another couple percent, unless they really tried to improve the IPC. So, assuming they didn't, we would only get incidental boosts in IPC, which is possible. Added with the latency, that could account to a 5-7% boost without it being unreasonable. How they got 9%, I don't know.

    But, remember, this could be a higher boost only at certain workloads, so the 15% only applying to the type of workloads shown in this sisoft leak, which isn't showing even the full plethora of benches available on sisoft. So, it could be a high watermark, so to speak. Or someone could have found a way to spoof it. Or any number of things.

    Either way, there isn't enough here to fully speculate, yet.

    I haven't seen official anything yet, but we are two months from launch now. So they are finalizing some things and MBs should be getting finalized soon or in the process.

    Sent from my SM-G900P using Tapatalk
     
  7. TANWare

    TANWare Just This Side of Senile, I think. Super Moderator

    Reputations:
    2,320
    Messages:
    9,043
    Likes Received:
    4,203
    Trophy Points:
    431
    You have to remember that they are basing on stock speeds. This means that improvements to the turbo boost could be the major part of the difference. If this is the case then the turbo boost will mean little to nothing for those overclocking. Also if this is the case then just a 200 MHz boost to base will not be enough for me.

    On the other note for those that do not overclock the overall boost to bring the CPU's closer to performance of stock Intel offerings is very welcome.
     
    Last edited: Feb 8, 2018
    Raiderman and hmscott like this.
  8. Raiderman

    Raiderman Notebook Deity

    Reputations:
    462
    Messages:
    746
    Likes Received:
    1,726
    Trophy Points:
    156
    If it's 200mhz for base and boost for instance, my 1700x/2700x would be 3.6ghz, and 4ghz. I would want a bit more than 200mhz on the overclock to be interested. More like 500 to 600 over the boost clock for me to get excited. If the 2700x or 2800x can hit 4.5ish than I will throw down for one.

    Sent from my SM-G935T using Tapatalk
     
  9. hmscott

    hmscott Notebook Nobel Laureate

    Reputations:
    5,001
    Messages:
    17,582
    Likes Received:
    21,599
    Trophy Points:
    931
    AMD Ryzen 3 2200G - Ryzen 5 2400G APU Preview


    AMD Ryzen 3 2200G and Ryzen 5 2400G APU Preview (3 pages)
    Stock Cooler and Xbox One Comparison
    Published: 8th February 2018 | Source: AMD | Price: | Author: Mark Campbell
    https://overclock3d.net/reviews/cpu_mainboard/amd_ryzen_3_2200g_and_ryzen_5_2400g_apu_preview/1

    "AMD's Ryzen 3 2200G and Ryzen 5 2400G fills me with a lot of excitement, as while they are part of the "low-end" of AMD's Ryzen CPU lineup they offer a compelling option for basic gaming and deliver a boatload of interesting design changes.

    What excites me most is the 2400G, as its CPU and GPU component both offer what we would describe as specification advantages over the Xbox One S, making the chip look like a great option for budget gaming given its literal console-grade specifications.

    The products boost in clock speeds, and use of Precision Boost 2 also showcases a clear advantage over AMD's older Zen processors, giving us a glimpse of exactly what to expect from AMD's upcoming Zen+ series of processors. AMD's boost in supported memory clock speeds is an incredibly encouraging thing to see, especially if the company plans to expand on this further with Zen+, which will be based on a newer 12nm production process from Globalfoundries.

    Sadly, we cannot get into more in-depth detail regarding the performance of this product, as that will have to wait for the official release of these CPUs on February 12th."
    Giggyolly
    I thought these APUs were too good to be true TBH, but I can see there are now some drawbacks with the new quad core design.
    That being said, they seem minimal, and look to offer some really great low-end value parts.
    I occasionally do office PCs for friends and family, and these could replace the i3s I'd usually opt for!

    AlienALX
    Expectation. People have very high ones, then you find yourself let down all of the time. The trick is to have none, and see the product for what it is. IE - a cheap, dirty little APU that can actually game. Meaning you can have a tiny rig that doesn't bust your bank account and run games on low settings.

    This idea was enormously popular with the original AMD APUs. Even though the on die GPU was crap (6670 or something) it enabled people to just stick one in and game without the added expense of a GPU.

    And yes, the GPU area on this should piddle on Intel's best tbh.

    Over the coming months I fancy building a Sega AM3 only box. Running Supermodel. This would be absolutely ideal for me [​IMG]

    RobM
    Speaking from experience as having gamed AAA titles on an APU, ie my last setup was an A10 and I was able to play with reasonable setting @ 1080p without any real problems except when things got really busy on screen such as late game HOI or Arma3 but for the most things were pretty fine.
    This new generation I bet will be far superior and akin to an ryzen 3 with a 750ti or better
    AMD RAVEN RIDGE Ryzen 5 And Ryzen 3 With Vega Graphics UNBOXING

    AMD Raven Ridge Unboxing, R5 2400G & R3 2200G Ready for Benchmarking!

    AMD Raven Ridge UNBOXING Reviewer's Kit

    AMD Raven Ridge Ryzen + Vega Unboxing the presskit

    AMD Ryzen APU with Radeon Vega unboxing and install

    The AMD Ryzen 5 2400G with Radeon RX Vega 11 Graphics Up Close!

    World's First Ryzen APU Unboxing!
     
    Last edited: Feb 8, 2018
    Raiderman and ajc9988 like this.
  10. hmscott

    hmscott Notebook Nobel Laureate

    Reputations:
    5,001
    Messages:
    17,582
    Likes Received:
    21,599
    Trophy Points:
    931
    Probably the 7th AMD Ryzen APU Unboxing on YouTube
     
    ajc9988 likes this.
Loading...

Share This Page