Quote:The overlocking team over at Brazilian publication Teclab has successfully transplanted memory chips from two RTX 2080 Super graphics cards into an RTX 2080 Ti to make what would be the equivalent to an RTX 2080 Ti Super.
...
Teclab noted that the RTX 2080 Ti Super booted up just fine with the RTX 2080 Super's memory. The graphics card's vBIOS accepted the new memory with open arms and ran it at 1,750 MHz, so no BIOS modding was required. In the end, the team was able to push the memory to 2,150 MHz (17,200 MHz effective), which represents a 22.9% and 10.7% increase over the RTX 2080 Ti's and RTX 2080 Super's default memory, respectively.

Unfortunately, Teclab didn't thoroughly test its "RTX 2080 Ti Super" to see how it performs with faster memory. The team only provided a Superposition run, where the graphics card put up a score of 11,460 points with the 1080p Extreme preset. A stock RTX 2080 Ti typically scores between 8,600 and 9,200 points.

Quote:The group detailed the process of removing the memory chips under hot air, giving the extracted chips fresh ball-grids, and placing the chips onto the RTX 2080 Ti PCB. No BIOS modding was required, as the RTX 2080 Ti card's video BIOS was able to auto-detect the chips and run them at 14 Gbps. From here on, manual overclocking easily runs the card at 2000 MHz (16 Gbps) memory, with overclocking headroom to spare. The memory clock could now be dialed all the way up to 2150 MHz (17.2 Gbps), something that's close to impossible with 14 Gbps chips. TecLab is calling their creation the world's first RTX 2080 Ti Super, which could very well be true. Last we heard, the RTX 2080 Ti Super could get more CUDA cores, and not just faster memory. Nevertheless, this mod blew our minds, and provides valuable pointers on how to solder dense BGA components without a multi million-dollar placer. We tip our hats to TecLab.

Quote:While Nvidia is still sitting high on its throne made of Titan RTX cards and hundred-dollar bills, AMD is ramping up the competition with cards like the Radeon RX 5600 XT. That’s forcing Nvidia to make some moves. Namely, Nvidia is dropping the price on its lowest-tier RTX cards, the GeForce RTX 2060, to just $299.

Quote:An EEC (Eurasian Economic Commission) listing, which was recently discovered by VideoCardz, reveals multiple SKUs for GTX 1650 graphics cards from MSI that weren't announced yet. Interestingly, their model names suggest that these will feature GDDR6 memory. If the products listed are actually coming to market, it would appear that the GTX 1650 is getting a performance uplift soon.
...
Although this is an official EEC submission, vendors sometimes submit these things just to claim dibs on a potential product name but never end up releasing said offering.

Quote:EVGA’s RTX 2060 KO Ultra is intended to steal some thunder from AMD’s RX 5600 XT. The end result is an improved RTX 2060 that matches the performance of AMD’s RX 5600 XT graphics cards. The performance difference is so small that In most games, users would be hard-pressed to notice a difference. Select games may show more than a 5% difference, but that goes both ways, leading to an overall tie in performance. The price points for the KO and KO Ultra cards are $299.99 and $319.99, respectively. Other RTX 2060s are priced higher and lose out on the performance per dollar metric the EVGA RTX 2060 KO offers, though there are other factors at play.
...
The EVGA RTX 2060 KO certainly disrupts the market solely based on price. The EVGA KO SKUs are the least expensive RTX models, without any major compromises. While it doesn't score a knock out or even a technical knockout against the AMD competition, it does squeak out a win by a very close judge's decision. It's a good performance option within its own ranks as well as against AMD’s RX 5600 XT. If you're looking for solid 1080p performance at the $300 price point and would like ray tracing capabilities, the EVGA RTX 2060 KO Ultra Gaming should be on the shortlist.

Quote:Multiple aftermarket MSI GeForce GTX 1650 graphics cards with GDDR6 memory appeared in a Eurasian Economic Commission (EEC) submission in February. Today, hardware leaker @momomo_us discovered two listings from German retailer PC Service Point that point to a GTX 1650 refresh en route.

Quote:Chinese news outlet Expreview has reviewed Nvidia's refreshed GeForce GTX 1650 gaming graphics card. The new GTX 1650 has GDDR6 memory for a speed boost over the original version with GDDR5.
...
It appears that the GTX 1650 GDDR6 is hiding another surprise under its hood. Expreview noticed that Zotac's GDDR6 model comes with slightly lower clock speeds. The GTX 1650 Destroyer OC has a 1,485 MHz base clock and 1,665 MHz boost clock, while the GTX 1650 GDDR6 Destroyer OC has a 1,410 MHz base clock and 1,620 MHz boost clock. That's equivalent to a 5% reduction on the base clock and 2.7% on the boost clock.
...
The GTX 1650 GDDR6 delivered about 2-12.5% performance gains over the GDDR5 version, depending on the benchmark and game. Collectively, we're looking at an average performance increase of 6.4%, based on Expreview's results.

The power consumption tests show the GDDR6 variant drawing slightly less power than the original. At idle, both models consumed about 9W. Average consumption on the GDDR6 version was 76W, while the GDDR5 version pulled around 83W. The lower power draw can be attributed to the fact that Zotac's GeForce GTX 1650 GDDR6 Destroyer OC has lower clock speeds compared to Gigabyte's heavily overclocked GTX 1650 Gaming OC.

Quote:NVIDIA updated the product page of its GeForce GTX 1650 graphics card to make GDDR6 an official memory option besides the GDDR5 that the SKU launched with, back in Q2-2019. NVIDIA now has two product specs for the SKU, the GTX 1650 (G5), and GTX 1650 (G6). Both feature 896 "Turing" CUDA cores, 56 TMUs, and 32 ROPs; but differ entirely in memory configuration and clock speeds.

The GTX 1650 (G6) features 4 GB of GDDR6 memory clocked at 12 Gbps, across a 128-bit wide memory bus, compared to the original GTX 1650, which uses 4 GB of 8 Gbps GDDR5 across the same bus width. This results in a 50% memory bandwidth gain for the new SKU: 192 GB/s vs. 128 GB/s. On the other hand, the GPU clock speeds are lower than those of the original GTX 1650. The new G6 variant ticks at 1410 MHz base and 1590 MHz GPU Boost, compared to 1485/1665 MHz of the original GTX 1650. This was probably done to ensure that the new SKU fits within the 75 W typical board power envelope of the original, enabling card designs that lack additional power connectors. As for pricing, Newegg recently had an MSI GeForce GTX 1650 GDDR6 Gaming X listed for $159.

Quote:Building on the EVGA GeForce RTX 2060 KO Ultra Gaming's success, EVGA is adding two more KO Edition graphics cards to its Turing-powered arsenal. The company today announced KO Gaming versions of the RTX 2080 Super and RTX 2070 Super.

Quote:It's uncertain if this will be a trend going forward, but MSI has launched the GeForce RTX 2080 Ti Gaming Z Trio that aims to be one of the best graphics cards by introducing a nice memory uplift. MSI's latest offering is the first GeForce RTX 2080 Ti to come with 16 Gbps of GDDR6 memory.
...
In case you've forgotten, the GeForce RTX 2080 Ti launched with 14 Gbps GDDR6 memory. Coupled with a 352-bit memory interface, the graphics card delivers a maximum bandwidth of 616 GBps. The GeForce RTX 2080 Ti Gaming Z Trio, on the other hand, is rocking 16 Gbps and, therefore, can pump out 704 GBps. This amounts to a 14.3% increase in memory bandwidth.

Quote:The EVGA GTX 1650 GDDR6 SC Ultra is in a difficult position right now. It performs much better than a GDDR5-equipped GTX 1650, nominally at the same $160 price as well. But that's just not going to work when the GTX 1650 Super also costs $160. The 1650 has more cores, giving it nearly 20% higher performance — even with a healthy overclock, the 1650 GDDR6 is still 10% slower than a 1650 Super. This can all be fixed with a price cut, but margins on budget hardware are already thin, so that may not happen.
...
Which brings us back to the earlier question: Where have all the good budget GPUs gone? The answer is that $150-$175 is the new definition of 'budget,' unfortunately. If you want to spend less than $150 on a graphics card, you'll inevitably end up making some serious compromises — on performance, efficiency, features, and more. And looking forward to AMD's RDNA 2 and Nvidia's Ampere architectures, it doesn't seem like those will do any better on the low cost front. Maybe Intel's Xe Graphics will be the hidden savior of budget GPUs. More likely, budget graphics will become the domain of AMD Renoir CPUs and Intel's future Xe LP integrated solutions.

Quote:Nvidia just released its latest 451.48 drivers for Windows PCs. These are the first fully-certified DirectX 12 Ultimate drivers, but they also add support for WDDM 2.7—that's Windows Display Driver Model 2.7. New to Windows 10 with the May 2020 update, and now supported with Nvidia's drivers, is hardware scheduling. This new feature shows up in the Windows display settings, at the bottom under the Graphics Settings, provided you have a Pascal or later generation Nvidia GPU. Could this help the best graphics cards perform even better and maybe shake up the GPU hierarchy? Probably not, but we decided to find out with empirical testing.

First, it's important to note that this is not a new hardware feature but rather a new API feature. GPUs, at least as far as we understand things, have been able to support GPU hardware scheduling for some time. The description from Microsoft is vague as well.
...
What to make of all of this, then? Nvidia now supports a feature that can potentially improve its performance in some games. Except, it seems just as likely to hurt performance as well. This is a new API and driver feature, however, so perhaps it will prove more beneficial over time. Or perhaps I should have dug out a slower CPU or disabled some cores and threads. I'll leave that testing for someone else for now.

At present, across five tested games using multiple APIs, on average (looking at all nine or ten tests), the change in performance is basically nothing. The 9900K with RTX 2080 Ti performance is 0.03% slower, and the GTX 1050 with the 9900K performed 0.73% slower. The 3900X with RTX 2080 Ti did benefit, but only to the tune of 0.06%. In other words, the one or two cases where performance did improve are cancelled out by performance losses in other games.

If you're serious about squeezing out every last bit of performance possible, maybe for a benchmark record, you can try enabling or disabling the feature to see which performs best for the specific test(s) you're running. For most people, however, it appears to be a wash. Your time will be better spent playing games than trying to figure out when you should enable or disable hardware scheduling—and rebooting your PC between changes.

Quote:Chinese publication IThome reported today that Nvidia is ostensibly discontinuing the GeForce RTX 2070, RTX 2070 Super, RTX 2080 Super, and RTX 2080 Ti. We've reached out to Nvidia, but the chipmaker has told us that it doesn't comment on rumors or speculation.

If we look at it objectively, it makes sense to start phasing out Turing-powered models to dedicate resources for the new and upcoming Ampere graphics cards. If Nvidia's continues to use the same development cycle for Ampere, the chipmaker should already have working samples of Ampere. And if things progress as they are, mass production would likely commence in August.

IThome's next claim is a bit controversial, though. The media alleges that Nvidia has purportedly suggested that its partners raise the prices on the aforementioned Turing graphics cards at the beginning of this month. The rumored reasons behind the chipmaker's recommendation are the rising interest in cryptocurrency mining and insufficient supply on TSMC's part. As a reminder, TSMC manufactures the Turing silicon for Nvidia on the foundry's 12nm FinFET process node. However, TSMC also has many other big-name clients, including Apple, AMD, and Qualcomm, so the foundry's services are always in high demand.

Nvidia might start gradually withdrawing the GeForce RTX 2080 and RTX 2080 Ti graphics card from the market to make way for the next best thing. If the information is accurate, the chipmaker could announce the GeForce RTX 3080 and RTX 3080 Ti in September, which is the rumored month for Ampere's launch. IThome thinks that the announcement will take place on September 17, but the new outlet wouldn't swear by it though.

Quote:Wouldn't it be cool to upgrade your graphics card with more VRAM? While the average person might not do it, YouTuber VIK-on published a video showcasing his journey to upgrading a Palit-branded GeForce RTX 2070 from 8GB of Micron GDDR6 to 16GB of Samsung GDDR6 memory.

VIK-on first got the idea of upgrading the RTX 2070 to 16GB when he was sent a leaked diagram for the RTX 2070 with a 16GB VRAM option. The leaked diagram shows that the RTX 2070 can support both 1GB and 2GB GDDR6 chips from Samsung (the 2GB chips are used to make 16GB of total VRAM capacity)

Remember those leaks on a potential 16GB RTX 2070 and RTX 2080 a few years ago? Well, this leaked diagram, if real, suggests that Nvidia was at least working on the idea behind closed doors.
...
Either way, it was cool to see that upgrade an RTX 2070 to 16GB does actually work. Hopefully one day we'll see an RTX 2070 that is fully functional with 16GB of GDDR6 to see if adding an additional 8GB of VRAM would have been worth it for Nvidia.

Quote:Nvidia announced its RTX 3060 during CES last week, but according to one report, the company has actually restarted production of its RTX 2060 and RTX 2060 Super. If true, it would mean Nvidia doesn’t think it can alleviate the graphics card shortage quickly enough if it relies solely on 7nm GPUs.

The rumor comes from French site Overclocking.com, which claims to gotten confirmation from several brands. Reportedly, Nvidia shipped out a new set of RTX 2060 and 2060 Super GPUs to re-enable the manufacture of these cards. If true, Nvidia could potentially alleviate the GPU shortage by relying on TSMC’s older (and presumably, less-stressed) 12nm product line.
...
Frankly, it’d be nice to see the RTX 2060 and 2060 Super back in-market, if only to bring a little stability to it. Here are Newegg’s current top-selling GPUs as of 1/20/2021:

Newegg’s best-selling GPUs are bottom-end Pascal cards. The last-gen RX 580 and the GTX 1660 Super are the only two consumer cards selling for under $500. Both of them are terrible deals at this price point.

There’s always a bunch of low-end garbage stuffed into the GPU market. Typically, these parts live below the $100 price point, where you’ll find a smorgasbord of ancient AGP cards, long-vanished GPU micro-architectures, and rock-bottom performance that almost always costs too much. Today, the garbage has flooded into much higher price points. Want a GTX 960? That’ll be $150. How about a GTX 460 for $145 or an HD 7750 for $155? There’s a GTX 1050 Ti for $170, which is only $40 more than the GPU cost when new, over four years ago.

Right now, it’s impossible to buy any GPU for anything like MSRP. If bringing the RTX 2060 and RTX 2060 Super back to market actually provides some stability and some kind of modern GPU to purchase, I’m in favor of it. At this point, it wouldn’t be the worst thing in the world if AMD threw the old Polaris family back into market, either. While they wouldn’t be a great value at this point ordinarily, the cheapest RX 5500 XT at Newegg is $397. Under these circumstances, any midrange GPU manufactured in the last four years that can ship for less than $300 would be an improvement.

The past five years have been the worst sustained market for GPUs in the past two decades. Currently, GPU prices have been well above MSRP for 24 out of the past 56 months, dating back to the launch of Pascal in late May, 2016. This isn’t expected to change until March or April at the earliest. When cards aren’t available at MSRP for nearly half the time they’ve been on the market over five years and two full process node deployments, it raises serious issues about whether we can trust MSRPs when making GPU recommendations. Right now, the best price/performance ratio you can get in the retail market might be an RX 550 for $122.

The GPU market in its current form is fundamentally broken. Manufacturer MSRPs have the same authority as any random number you might pick out of a hat. There are a lot of factors playing a part in the current situation, including manufacturing yields and COVID-19, but this problem started four years before the pandemic.

AMD and Nvidia need to find a better way to ensure that customers are able to buy the cards they actually want to purchase, or they need to delay their launches for a sufficient length of time as to build up a meaningful stockpile of hardware, sufficient to supply launch demand for a matter of days, not seconds. Alternately, they may need to delay launches until yield percentages and availability are high enough to ensure a constant stream of equipment to buyers.

Right now, we have launch days that sell out instantly and interminable delays between new shipments. If these rumors are true, and we hope they are, Nvidia bringing back the RTX 2060 and 2060 Super will help a little in the short term, but what we obviously need is for AMD and Nvidia to take a fundamentally different approach to product inventory management. As things stand, these aren’t product launches. They’re product teases.

Quote:In the video, Testing Games ran an assortment of gaming benchmarks, including Cyberpunk 2077, Battlefield V, and Forza Horizon 4. On average, the used mining 2080 Ti was about 10% slower than the brand new 2080 Ti. One outlier was Forza Horizon 4, which showed the mining card as 20% slower than the new card.

The main culprits for the reduced performance are GPU clock speed and temperatures. On average, the heavily used RTX 2080 Ti mining card was 16C hotter than the brand new RTX 2080 Ti. This caused over a 100MHz drop in boost frequency for the mining card, creating the performance losses. This is totally normal as Nvidia's GPU Boost 4.0 algorithm (equipped on Turing and Ampere-based cards) is tuned to be very sensitive to GPU temperature.

Unfortunately, Testing Games didn't benchmark the used mining card with replacement thermal pads and a fresh new application of thermal paste. Oh, and cleaning — dust buildup in the heat sink fins can also greatly hinder cooling performance. Theoretically, this should be all that it takes to bring GPU temperatures back down to normal and gain all that performance back.

Even if you aren't a miner, this test shows why it's a great idea to dust out your computer every now and then, and even apply a fresh coat of thermal paste to your CPU and GPU after several years. Thermal paste is known to get dry (especially on laptops) after years of use, which will reduce its thermal performance. Typically the hotter the CPU or GPU is, the faster the TIM will dry out. Besides, after three years, the warranty period on the card has likely expired so you've got nothing to lose.

This also applies to gamers buying used mining cards on eBay or from some other retailer. If you find the used graphics card underperforms, all it might need is a bit of maintenance to bring the card back up to full speed. Just be sure to factor in the costs (time and thermal pads) when bidding.

Quote:In an attempt to relieve some of its Ampere stock, Nvidia recently introduced the Cryptocurrency Mining Processor (CMP) product line that specifically targets Ethereum mining. While the chipmaker was reluctant to share more details, the truth has come to light in the latest GeForce Game Ready 461.72 WHQL driver.

When Nvidia launched CMP, the chipmaker was firm on the fact that it wouldn't impact the supply of its GeForce gaming graphics cards. Since CMP was designed to solve the cryptocurrency mining situation with Ampere, it would make sense that the graphics cards were based on older silicon. VideoCardz dug into the 461.72 driver and discovered that the device IDs for the 30HX and 40HX are closely related to Turing.
...
On the other hand, the 50HX and 90HX won't be available until the second quarter of this year so these two models aren't in Nvidia's driver yet. However, we have some suspicions that that the 50HX may be using the TU102 silicon, which is the one that powers the GeForce RTX 2080 Ti. The 50HX appears to be a lower clocked version of the past flagship with 1GB less memory.

The 90HX seems to take after the GeForce RTX 3080. If so, it would be the only CMP device to leverage the Ampere silicon, more specifically the GA102. Our take is that Nvidia is probably recycling defective GA102 dies that don't meet the specifications for the GeForce RTX 3080 or RTX 3090 and sticking them into the 90HX. It's a great way to maximize yields after all.

Reutilizing Turing to cushion Ampere's shortages is a good idea given that Turing is still on TSMC's older 12nm process node. Additionally, it'd allow Nvidia to dump leftover Turing silicon. Since Nvidia tapped into Samsung's 8nm process node for Ampere, CMP production has no impact on its latest GeForce gaming graphics cards. But whether CMP can convince cryptocurrency miners to lay off Ampere is another question.

There is still a lot that we don't know about Nvidia's CMP offerings though. Since CMP is based on Turing, there should be headroom for hash rate optimization. It will be interesting to see whether Nvidia will put a lock on CMP's performance like it did with the GeForce RTX 3060. More importantly, we'll need to look at the pricing for CMP because if it isn't priced right, cryptocurrency miners would just continue to scavenge GeForce products instead.