Threadripper Leaks to GN, Pt 2
We’ve received some follow-up information on Threadripper. Last week, we published tables from an official AMD internal document showing upcoming Threadripper next-gen preliminary specs. As a note, this is a new source from us, but we have vetted the source. That said, it’s important that we temper expectations for a few reasons: (1) time hasn’t yet proven how accurate the data is that this source has access to, and (2) things could shuffle in a pre-launch product. Let’s just put this firmly in “grain of salt” territory for now, then we can all look back on this at the Threadripper 3 launch and evaluate if data from this source can be trusted in the future!
We previously looked at official AMD documents that detailed an sTRX4 Threadripper HEDT CPU with quad-channel support for memory, mentioned alongside an sWRX8 Threadripper workstation CPU with eight-channel support. Since then, we’ve learned that, allegedly, sTRX4 will be 4 chiplets, while sWRX8 will be 8 chiplets, so that’d be a 32-core HEDT CPU and higher core-count workstation CPU, potentially up to 64 cores, although it’s possible that AMD could disable some of them or bin-out from failed Epyc CPUs.
We’ll see how it turns out!
ASRock RX 5700 XT Taichi X Launching
As we’ve been ramping-up to review all the 5700 series cards, ASRock too has been ramping its 5700 efforts. The new RX 5700 XT Taichi X OC+ will be ASRock’s flagship 5700-series GPU, hosting a triple-axial cooler with the classic Taichi obsession with LEDs dead-center. Photos reveal that the card has dual-BIOS with a silent and OC mode pre-installed, along with an LED on/off hard toggle along the top of the card. If ASRock’s marketing images are realistic, it looks like it’s running a fairly standard aluminum baseplate for the memory, although the image doesn’t show coverage on the MOSFETs -- that might be a separate pulled-away piece. ASRock is advertising the card as a 10+1 design, using probably the world’s stupidest marketing chart to do so: The chart, as we’ll put on the screen, literally just says 10+1 versus 7+1, with the X-axis labeled as Power Phase qty.
That’s not really how this works. More phases doesn’t necessarily mean better, and even if it did, this is still an embarrassing way to advertise a product.
Anyway, marketing aside, the card will be among the most expensive 5700 XT units we’ve seen so far, so we’ll try to get our hands on it.
Rumor: GTX 1660 Super & GTX 1650 Ti
Rumors have surfaced this past week indicating that Nvidia could be readying a GTX 1660 Super in conjunction with a GTX 1650 Ti, both in a presumable response to AMD’s entry level Navi cards, supposedly based on Navi 14.
The alleged GTX 1660 Super, for its part, would see the same TU116-300 die from the original GTX 1660 extended to the Super counterpart. However, the card would come with a memory upgrade, in the form of 6GB of 14Gbps GDDR6 memory. The current GTX 1660 boasts 6GB of 8Gbps GDDR5. The memory would still operate over a 192-bit interface. Although the rumors don’t explicitly mention it, we could also expect a bump in both core clock and boost clock, especially given that the Turing silicon has matured at this point.
The same rumor also points to the GTX 1650 Ti coming to market with up to 1152 CUDA cores, up from the 896 cores found on the current GTX 1650. As always, take the rumor with a grain of salt. Although, Nvidia trying to round out its low end Turing offerings in fortification against AMD’s looming, cheaper Navi-based cards would make sense.
AMD’s RDNA 2 & Zen 3 Architectures Set for 2020 on 7nm+
New slides from AMD made their way onto the web recently, and they more or less confirm what we already knew. The slides appear to be part of a bigger corporate deck from a recent presentation.
Regarding Zen, the slides show the 7nm Zen 2 as “shipping” and the Zen 3 7nm+ as “design complete.” This reiterates the speculation that Zen 3 will be locked into 7nm+, which isn’t a huge surprise. The slides also show Zen 4 as “in design” with a 2022 launch window.
Moving over to RDNA, the slides indicate that AMD is moving further away from its GCN architecture, forging ahead with RDNA 2. Like Zen 3, RDNA 2 will arrive on the back of TSMC’s 7nm+ process node with EUV.
AMD Releases AGESA 1003ABBA Helping Boost Clocks
In our last HW News episode, we mentioned that AMD had responded to the enthusiast outcry regarding Ryzen 3000 processors not achieving proper boost speeds. Earlier this week, AMD delivered on its promise to update the community regarding its fix for what it identified as a firmware problem on its end. We already posted our testing results from the ABBA update, if you’re curious on how performance is affected.
In addition to a more aggressive boosting algorithm, AMD also overhauled idle states with the latest AGESA microcode. AMD says this should net lower voltage states while the CPU is idle, and a new activity filter will help the processors filter out “intermittent OS and application background noise.” This should help prevent the boost algorithm from overzealously responding to workloads that don’t truly need it.
Lastly, AMD will roll out a new Monitoring SDK kit, which will allow for better monitoring and reporting of CPU behavior for Ryzen 3000 chips. AMD notes that final BIOSes should arrive in about three weeks’ time.
6GHz Memory Overclock Achieved
If previous records were any indication, the 6GHz frequency threshold was bound to be crossed sooner or later. As of now, G.SKILL can lay claim to that distinction. The new record was achieved by no other than accomplished overclocker Toppc, and validated by HWBot.
Toppc used an MSI MPG Z390I Gaming Edge AC motherboard, paired with an i9-9900K. The DDR4 used to claim the 6,016.8MHz world record was G.SKILL’s own 8GB Trident Z Royal. And of course, it goes without saying that a generous amount of LN2 was used. The previous record was a 5.9GHz overclock obtained by MSI and overclocker Kovan Yang back in August.
Gigabyte's Aorus Brand Extended to Liquid Coolers
If you thought the CLC market was already crowded, than Gigabyte is here to disagree with you. The company recently announced that its ever expanding Aorus brand will now include all-in-one liquid coolers. Gigabyte has already dabbled in CPU cooling, with its Aorus ATC800 air cooler.
The Aorus Liquid Cooler 240 will use a -- wait for it -- 240mm radiator, cooled by a pair of RGB illuminated 120mm fans. The cooler will also feature a pump-block design similar to that of the ASUS ROG Ryujin, in that it will have an LCD screen atop the pump-block housing that can display temperature, among other things.
Supported sockets include AM4, LGA115x, and LGA2066. The Aorus Liquid Cooler 240 is also compatible with the Asetek AMD TR4 Retention Kit. Pricing or availability wasn’t disclosed.
Rumor: Navi Could Be Headed to Workstations
Further fueling Navi 14 speculation is the discovery of new Navi 14 PCI IDs in a Linux patch. The code within the patch adds the Navi 14 PCI IDs 0x7341 and 0x7347, respectively. In addition, the patch references a “workstation SKU” in the patch message.
We’ve also seen signs from past Linux drivers pointing to AMD revising Vega silicon for the rumored Arcturus compute cards.
Editorial: Eric Hamilton
Host: Steve Burke
Video: Josh Svoboda, Andrew Coleman