Futuremark announces their new 3DMark Time Spy Extreme benchmark, which will be released to the public on October 11 as a free update for 3DMark Advanced Edition and 3DMark Professional Edition licenses purchased after July 14, 2016.

Time Spy Extreme is a 4K gaming benchmark that raises the rendering resolution to 3840 × 2160. A 4K monitor is not required, but your graphics card must have at least 4 GB of memory.

The CPU test has been redesigned to let processors with 8 or more cores perform to their full potential. Compared with Time Spy, the Extreme CPU test is three times more demanding. It also lets processors use more advanced instructions sets up to AVX2 when supported.

 Personal thought, I have no sources about it. Just pure speculation.

Thanks to der8auer (Source), now we know that the ThreadRipper SKUs have 4 Zeppelin die inside, and even if one of my sources was partially fooled (However the working dies are just two), this info is very good for AMD. Better than expected.

Why do I say that it's a good news? Before to install a Die on the Package, the die has to be binned and tested. This means that in a ThreadRipper CPU are installed four fully functional die. Someone could think that it’s a waste of silicon, but it's partially true.

This also means that our leaks about the crazy high yields of Zeppelin are true. AMD can afford to throw away two fully functional die in order to sell a ThreadRipper CPU.

Lastly, we can have an idea about AMD supply chain. AMD can’t know how many EPYC or ThreadRipper CPUs will sell in the near future, so AMD can produce packages with 4  fully functional dies, as many as it can. AMD can adjust the production of EPYC and ThreadRipper CPUs according to market request.

So, at a later stage, AMD will decide to commercialize these CPUs as ThreadRipper or as EPYC. Especially considering a EPYC/ThreadRipper CPU cost is about 110/120$. It sounds crazy, doesn’t it? Infinity Fabric is really a revolution.

According to latest rumors, next GTX1070Ti will be equipped with GDDR5 @9Gbps memory. Thanks to this choice, this card will be able to deal with the VEGA56, without stepping on GTX1080 toes. GTX1080 is equipped with GDDR5X @11Gbps memory.

Also, the GTX1070Ti will be probably equipped with a tweaked N17E-G3 GPU (GTX1080M). NVIDIA is working very well: it's leaving no room for AMD counter-attacks.

AMD has released a new major software update for all available Radeon graphics cards. Crimson ReLive 17.7.2  brings a number of new features, improvements and bug fixes.

Drivers are available through the Radeon Settings > Notifications or online at the AMD driver download page. 

EVGA has officially announced on its website the new fifth edition of the EPOWER.

The EVGA EPOWER V “The Untouchables” is a stand-alone power board which allows experienced overclockers to provide more juice to various graphics cards, especially physically and power constrained reference designs from Nvidia and AMD.

SPECIFICATIONS

• 12 phase main output (0.6 - 2.0V, 600A)
• 2 phase secondary output (0.6 - 2.3V, 80A)
• IR latest generation digital PWM
• Remote sense option support for vdroop compensation
• Integrated voltage monitor LED display
• Onboard EVbot MKII controls allow you to adjust voltages on the fly!
• Update your EVbot firmware and connect it via the EVbot port!
• Control your EPOWER via software by connecting a USB Type-C connector to your PC
• Dual 3-pin +12V Fan header
• 1 Year, DOA Warranty (no exceptions)

As we said years ago, Zen has a modular (LEGO-like) design. AMD created Zen in order to have a malleable uArch: customers can add or remove SIMDs and instruction sets, customize the different levels of cache, etc.

But this is only the beginning of AMD vision. We know that Apple, like other companies, is using SiP (System-in-Package) technology to produce some of its own products (E.g. Apple S1), but AMD went over.

AMD Chiplet seems to be an evolution of Palo Alto Research Center (PARC) studies.

Chiplet is an evolution of today Interposer employment: the Interposer has a pattern of sites for mounting chiplets.

As we have seen so far, companies put a single GPU or CPU or a FPGA on a Interposer with some memory chips (HMC or HBM). Thanks to Chiplet, we can mix different kinds of chips on the same Interposer: GPU, ASIC, CPU, FPGA, Modem LTE, Ethernet CTRL, different memories (DDR4 or HBM), etc.

Chiplet could be a revolution because we can create a whole mainboard (Ethernet CTRL, USB CTRL, etc.) using a single CPU package!

Some weeks ago we wrote about a possible Revision 2.0 of the Socket 2066. We were wrong … but the truth is closer than you think. The source is the same who told us about Socket FM2, AM4 and TR4.

The mainboards manufacturers will release a new wave of products, not with a new Socket, but with a revised and reinforced VRM section, in order to maintain huge overclocks for long periods with the next 16 core and 18 core Skylake-X SKUs.

Today X299 mainboards will support all Skylake-X line-up without problems, but the manufacturers want to be sure to have the best products to support the Core i9-7980XE launch. The bad publicity made by der8auer and other overclockers has been helpful, by this point of view. The new mainboards will be disclosed during the next weeks.

According to our sources, Intel could release in the near future (4Q17 or 1Q18) the Skylake-X SKUs with soldered IHS.

It’s not a news that Intel has rushed Skylake-X in order to counter-attack incoming AMD Ryzen and  ThreadRipper, and to accomplish this task has made some compromises. One of these is the use of TIM between the Die and the IHS.

It seems that Intel hasn’t had time to test extensively the process of soldering with Skylake-X, so the Santa Clara company has decided to use the TIM to buy time.

However, if Skylake-X will sell well, Intel will avoid to solder the IHS, in order to save money.

Designing a microchip (CPU, GPU, SoC, FPGA, etc.) has proved to be more and more complex and expensive, so the companies are studying different methods to do a faster analysis and enhanced binning of chips produced.

XFR by AMD is one of these methods, as you can read here. In fact we have seen that the Ryzen 1800X is better than the Ryzen 1700X, and the Ryzen 1700X is better than the Ryzen 1700 with our own eyes.

NVIDIA, in Feb. 2016, has filed a patent called: “Performing on-chip partial good die identification”. Thanks to this technology, NVIDIA could make a faster and more precise binning of its GPUs (The GPU V100 is a Behemot! 815 mm²!): “Embodiments of the present invention relate generally to integrated circuit fabrication and testing and, more specifically, to performing on-chip partial good die identification”.

Also, hand-picking good GPUs will be probably easier than ever. Since NVIDIA filed this patent in Feb. 2016, we can say that we could see this technology utilized in consumer version of Volta uArch (2018) if we are lucky.

Let us spread some light on Skylake-E/EP/EX Dies. In several forums there is confusion about the Die utilized by Intel to produce the new Skylake-X SKUs. A lot of users don’t seem to know that Intel – for almost 10 years – use three kind of Die to produce its own Server SKUs in order to maximise profits: LCC (Low Core Count), MCC (Medium Core Count) and HCC (High Core Count).

For this reason we have to wait a little bit longer to see 14, 16 and 18 core variants of Skylake-X SKUs. These CPU are based on MCC Die (Up to 20 Core), while 6/8/10/12 variants are based on LCC Die (Up to 12 Core). Intel has always used only LCC Dies in its HEDT SKUs, until today.

Intel will have hard times to tweak the MCC Die in order to stay inside the TDP supported by LGA2066 Socket (140W, max 160W), and that’s why we could see a new revision of this socket (Intel engineers are still thinking how to solve this situation).