This is something we haven’t seen before. NVidia has taken a relatively successful card, the GT 1030, and has implanted DDR4 in place of GDDR5. It’s actually getting system memory on it, which is a tremendous downgrade. The memory bandwidth reduction is several-fold, dropping from 48GB/s to about 16GB/s with DDR4, but the part that’s truly wrong is that they used the same product name.
The GT 1030 has always been an interesting product, and that’s only true because of the mining boom and GPU scarcity issues of earlier this year. Typically, the GT 1030 – or similarly ultra-low-end cards – would not get our recommendation, as a GTX 1050 or RX 550 would make more sense and be close in price. Earlier this year, even GTX 1050s and RX 550s had evaporated, leaving only overpriced GT 1030 GDDR5 cards (that we were somewhat OK with recommending). Fortunately, performance was decent. Was. Before the DDR4 surgery.
It’s time to benchmark the GT 1030 versus the GT 1030 Bad Edition, which ships with DDR4 instead of GDDR5, but has the same name as the original product. In a previous rant, we railed against these choices because it misleads consumers – whether intentionally or unintentionally – into purchasing a product that doesn’t reflect the benchmarks. If someone looks up GT 1030 benchmarks, they’ll find our GDDR5 version tests, and those results are wildly different from the similarly priced GT 1030 DDR4 card’s performance. On average, particularly on Newegg, there is about a $10 difference between the two cards.
The GT 1030 with DDR4 is one of the most egregious missteps we’ve seen when it comes to product marketing. NVidia has made a lot of great products in the past year – and we’ve even recommended the GT 1030 GDDR5 card in some instances, which is rare for us – but the DDR4 version under the same name was a mistake.
There's been some online internet outrage about a leaked nVidia NDA, as published by Heise.de previously. Some of the online comments got a little out of hand and were severely misguided, so we decided to get on a call with a US-based, licensed lawyer, rather than continue to watch as armchair "experts" tried to extract any nefarious meaning they could from the document.
We'll leave this one to audio format. Our call with our legal correspondent goes through the document line-by-line and should answer any remaining questions. Overall, we think this was a mountain made of a molehill, and that little language in the document is abnormal or 'dangerous.' Note also that parties may terminate at any time (despite what some commenters will tell you), and that such an NDA doesn't somehow magically "prevent GPP2" from getting out because, remember, that wouldn't be covered under "Confidential Information." GPP was never disclosed to press by NV -- that was all third-party sources, and so that information is not covered under the agreement.
Audio/video below -- you can just tab away from this one, it's primarily audio:
Hardware news hasn’t slowed since Computex; in fact, this week has been among the busiest in months, with several news items out of the “Big Three” manufacturers. NVidia has seemingly purchased too many GPUs, according to GamersNexus sources (and verifying other stories), GPU shipments overall are trending downward, Intel’s CEO “resigned,” and AMD is working on Vega 20 and V340 products.
Other news for the week includes smaller items, like Be Quiet! opening a US service center and expanding US operations. Learn more in the video, or find the show notes below:
Some new rumors have indicated an nVidia GPU launch in “late July,” which correspond with our previous GDDR6 timelines putting us in July-September for a launch date. Our long-standing estimate has been August to September for the most probable launch window. We’d still plant it in August, but Tom’s seems to be reporting late July.
The hardware world has been busy for the past week. This week's news recap covers one rumor -- speculation that Intel "might" show a GPU in 2019 -- and then covers major news stories. One of those is Intel's Z390 chipset, whose block diagram has been detailed against existing Z370 block diagrams. We'll talk those chipset differences in the show notes (and the video) below. NVidia's earnings report also showed remarkably strong performance for the company, with mining revenue marking a new category of earnings at $289 million. What's unclear is how that's tracked -- we don't know if that's direct-to-miner sales, e.g. selling to large mining operations, or if that's also counting users who buy 10 GPUs at a time on Newegg. The latter might appear like a normal "gaming" purchase, depending on how it's all tracked and broken-down.
A handful of other news items are also present, including net neutrality discussion, Corsair's Obsidian 1000D and Spec-Omega, and a couple of other items. Learn more in the video below or, if you prefer written text, the show notes below that.
Intel’s Hades Canyon NUC is well-named: It’s either a reference to hell freezing over, as AMD and Intel worked together on a product, or a reference to the combined heat of Vega and an i7 in a box that’s 8.5” x 5.5” in size. Our review of Hades Canyon looks at overclocking potential, preempting something bigger, and benchmarks the combined i7 CPU and Vega M GPU for gaming and production performance. We’re also looking at thermal performance and noise, as usual. As a unit, it’s one of the smallest, most-powerful systems on the consumer market get right now. We’ll see if it’s worth it.
There are two primary SKUs for the Intel NUC on Newegg, both coming out on April 30th. The unit which most closely resembles ours is $1000, and includes the Intel i7-8809G with 8MB of cache and a limited-core Turbo up to 4.2GHz. The CPU is unlocked for overclocking. It’s coupled with an AMD Vega M GH GPU with 4GB of high-bandwidth memory, also overclockable, but does not include memory or an SSD. You’re on your own for those, as it’s effectively a barebones kit. If you buy straight from Intel’s SimplyNUC website, the NUC8i7HVK that we reviewed comes fully-configured for $1200, including 8GB of DDR4 and a 128GB SSD with Windows 10. Not unreasonable, really.
The past week of hardware news primarily centers around nVidia and AMD, both of whom are launching new GPUs under similar names to existing lines. This struck a chord with us, because the new GT 1030 silently launched by nVidia follows the exact same patterns AMD has taken with its rebranded RX 460s as “RX 560s,” despite having significant hardware changes underneath.
To be very clear, we strongly disagree with creating a new, worse product under the same product name and badging as previously. It is entirely irrelevant how close that product is in performance to the original - it’s not the same product, and that’s all that matters. It deserves a different name.
We spend most of the news video ranting about GPU naming by both companies, but also include a couple of other industry topics. Find the show notes below, or check the video for the more detailed story.
Find the show notes below, or watch the video:
Revealed to press under embargo at last week’s GTC, the nVidia-hosted GPU Technology Conference, nVidia CEO Jensen Huang showcased the new TITAN W graphics card. The Titan W is nVidia’s first dual-GPU card in many years, and comes after the compute-focused Titan V GPU from 2017.
The nVidia Titan W graphics card hosts two V100 GPUs and 32GB of HBM2 memory, claiming a TDP of 500W and a price of $8,000.
“I’m really just proving to shareholders that I’m healthy,” Huang laughed after his fifth consecutive hour of talking about machine learning. “I could do this all day – and I will,” the CEO said, with a nod to PR, who immediately locked the doors to the room.
At GTC 2018, we learned that SK Hynix’s GDDR6 memory is bound for mass production in 3 months, and will be featured on several upcoming nVidia products. Some of these include autonomous vehicle components, but we also learned that we should expect GDDR6 on most, if not all, of nVidia’s upcoming gaming architecture cards.
Given a mass production timeline of June-July for GDDR6 from SK Hynix, assuming Hynix is a launch-day memory provider, we can expect next-generation GPUs to become available after this timeframe. There still needs to be enough time to mount the memory to the boards, after all. We don’t have a hard date for when the next-generation GPU lineup will ship, but from this information, we can assume it’s at least 3 months away -- possibly more. Basically, what we know is that, assuming Hynix is a launch vendor, new GPUs are nebulously >3 months away.
NVidia today announced what it calls “the world’s largest GPU,” the gold-painted and reflective GV100, undoubtedly a call to its ray-tracing target market. The Quadro GV100 combines 2x V100 GPUs via NVLink2, running 32GB of HBM2 per GPU and 10,240 CUDA cores. NVidia advertises 236 TFLOPS Tensor Cores in addition to the power afforded by the 10,240 CUDA cores.
Additionally, nVidia has upgraded its Tesla V100 products to 32GB, adding to the HBM2 stacks on the interposer. The V100 is nVidia’s accelerator card, primarily meant for scientific and machine learning workloads, and later gave way to the Titan V(olta). The V100 was the first GPU to use nVidia’s Volta architecture, shipping initially at 16GB – just like the Titan V – but with more targeted use cases. NVidia's first big announcement for GTC was to add 16GB VRAM to the V100, further adding a new “NV Switch” (no, not that one) to increase the coupling capabilities of Tesla V100 accelerators. Now, the V100 can be bridged with a 2-billion transistor switch, offering 18 ports to scale-up the GPU count per system.
We moderate comments on a ~24~48 hour cycle. There will be some delay after submitting a comment.