According to Bloomberg sources, Intel Plans to Fire to 20% of employees as part of a major cost-cutting move, the decision is said to be the new CEO's first major move Lip-Bu Tan, who recently took over the company. As of the end of 2024, Intel employed about 108 900 people, which means that more than 100 people could be subject to layoffs. 21 000 employees.
It is noted that similar information had already surfaced in the summer of 2024, when the layoff of 10 thousand people was discussed. However, the number of personnel has not changed significantly since then. The new report confirms: this time we are talking about much more large-scale measures. This coincides with the expectations of analysts, including - CJ Muse of Cantor Fitzgerald, who previously noted “aggressive cost cutting” and “high levels of motivation” under the new CEO.
Previously, at Pat Gelsinger Intel already had a $10 billion cost-cutting plan in place, but it remains unclear whether Tan is continuing on that course or is forming a new strategy.
The massive layoffs could be announced within days, just ahead of Intel's first quarterly earnings call for the first quarter of 2025. It will be the first under Lip-Bu Tan, and the actions suggest management is not planning to delay the reforms.
Analysts see the possible layoffs as a bullish signal for investors. Amid the rumors, Intel shares have already risen 2% in after-hours trading and 4% in the main session. Investors are also awaiting more announcements at the event. Intel Foundry Day, which will take place on April 29.
Resource Igor's Lab published a study according to which GeForce RTX 50 graphics cards can suffer from overheating of components even during normal operation. The reason lies in dense layout power elements, which disrupts the normal distribution of heat on the board.
During testing PNY RTX 5070 и Palit RTX 5080 Gaming Pro OC, the author of the study, Igor Vallossek, recorded dangerous temperatures in the zone between GPU and output ports. The RTX 5080 board heated up to 80,5 ° C, while the compact RTX 5070 has up to 107,3 ° C, while the graphics chip itself remained in the 70 °C range. Such values, especially higher 80 ° C, can speed up degradation of components.
Overclocker notes that chokes, transistors and VRM circuits are located too close together, without a sufficient heat dissipation system. At the same time, most models no thermal pads on the hot zones of the board, and there is no heat transfer to the backplate. After installing the pads and thermal paste, the temperature of the RTX 5080 dropped to 70,3 °C, and the RTX 5070 to 95 °C and below.
In addition to local overheating, Vallossek also criticized NVIDIA official documentation — according to him, many parameters are specified under ideal conditions and do not take into account real operating scenarios. There were no comments from NVIDIA at the time of publication.
Intel announced 200S Boost profile, designed for processors Core Ultra 200 series and motherboards on the Z890 chipset. The new mode automatically increases the clock speeds of the subsystems — without unnecessary risk and maintaining the warranty. The company says overclocking results in noticeable gains in games where performance is critical. low latency.
Among changes — an increase in the die-to-die bus frequency from 2.1 to 3.2 GHz, as well as an increase in the memory frequency from 6400 to 8000 MT/s, provided the power supply and XMP profile are stable. This should provide an increase in scenarios sensitive to bandwidth and latency.
Overclocking profile only works with Core processors Ultra with index K: 285K, 265K/KF, 245K/KF, and also requires one memory module per channel. UDIMM and CAMM2/CDIMM modules are supported, but Intel warns that only some memory kits will achieve the maximum values.
Enabling 200S Boost is easy: update the BIOS, find the desired item in the overclocking section (near XMP), select a profile, save the changes and reboot the PC. Intel recommends using Cinebench and XTU for stability testing.
Important: The company maintains the standard three-year warranty, if only this profile is used. But as always in case of overclocking - the result depends on the quality of the motherboard, cooling and power supply.
Intel has begun preparations for memory implementation, according to a new job posting GDDR7 in future GPUs. The company's website posted an advert for a memory subsystem validation engineer, stating that the candidate will work with GDDR6 and GDDR7 as part of Arc GPU-products. This means that Intel is already testing new memory standards on prototype graphics chips.
Development of new GPU with GDDR7 support is in its early stages. If we focus on standard design times, then The actual "production" appearance of GDDR7 in Intel's lineup may not happen before 2026. GDDR7 is expected to be used in third generation Arc, on the Xe3 Celestial architecture.
The company seems to be in no rush to enter the premium segment and will focus on successful solutions at the B570 and B580 level. This is a smart move: Intel is increasingly gaining a foothold in the niche affordable graphics cards for 1080p gaming, where it is now fighting with AMD.
appeared on the web 16-core AMD Zen 5c chiplet crystal shots, which are used in new server processors EPYC 9005These chiplets are manufactured according to 3nm process technology, and the photographs clearly show that the core structure has been completely reworked.
For the first time in a compact architecture, AMD used single core without division into CCX blocks (Core Complex). In previous generations, including Zen 4c, the cores within a single chiplet were divided into two CCX blocks of 8 cores, each with its own L3 cache. This complicated the interaction between the cores: in order for one core to exchange data with another, it was necessary to go through the Infinity Fabric bus.
Now everything has changed: Zen 5c combines all 16 cores into one common structure, in which they directly use 3MB unified L32 cache. This significantly reduces latency when transferring data between cores and improves performance, especially in latency-sensitive tasks.
Chiplet size is 5,7 x 14,83 mm - it is stretched to accommodate more cores and a centralized cache. In one EPYC 9005 processor, there can be up to 12 such chiplets, which gives a maximum 192 core. At the same time, the thermal package of the flagship model reaches 500 W, and the starting one is from 400 W.
Zen 5c repeats all the possibilities of the Zen 5 architecture, but it is made in on a 25% smaller body. This makes it a compact analogue for servers and workstations, while maintaining the level of performance. Unlike Intel, where the "energy-efficient" cores work on a different architecture, AMD uses same instruction set and IPC.
Chinese company XFX presented new video card RX 6500 4G DUAL ITX, which is not listed in official AMD line. Despite the lack of an announcement from the manufacturer, the card is already available on the Asian market and is simplified version RX 6500 XT.
RX 6500 DUAL ITX made in a compact format with two fans, and its dimensions are 171 × 37 × 110 mmShe doesn't require additional nutrition, consuming all about 55 W, which allows you to feed exclusively from PCIe slots. Video outputs are presented one HDMI и one DisplayPort.
The model is based on a graphics processor Navi 24containing 16 computing units, as with the RX 6500 XT. But the frequencies are reduced: 1728 MHz basic and 2066 MHz maximum against 2310-2815 MHz in the XT version. Also lowered effective memory frequency — 16 Gbit/s instead of 18 Gbit/s.
In essence, this energy efficient analogue of RX 6500 XT, adapted to systems without a powerful power supply. The performance will be lower, but this option is ideal for compact and office PCs.
In addition, the company XFX has previously shown the LP variant RX 6500 s one slot and one fan, positioning it as the most powerful single-board graphics card AMD.
It's important: AMD itself has not officially unveiled the RX 6500, and the model is not available on the company's website. However, Leaks and OEM Sales cards have been around in China for several months now. The new product is aimed at cheap builds, where low energy consumption and compact dimensions are important.
Company NVIDIA released another one emergency hotfix GeForce drivers - version 576.15, designed to solve serious problems with temperature display and monitoring, which arose after the update 576.02. The new version is aimed primarily at owners GeForce RTX 50-series graphics cards, where the failures are particularly noticeable.
Among most disturbing bugs, which 576.15 eliminates, flickering shadows, Shader compilation failures, temperature sensor not working after sleep и black screen when exiting Modern Standby mode on laptops. Also fixed. error with reduced frequencies at idle, which appeared after the previous driver.
List of fixed issues in 576.15: – [RTX 50] Shadow flickering in some games after installing 576.02 – Lumion 2024 crashes when rendering on RTX 50 – After exiting sleep, the temperature stops being displayed GPU – Shader compilation crashes in some games – Black screen on laptops with RTX 50 when waking up from sleep – V-Sync stuttering in SteamVR when using multiple monitors – Reduced idle frequencies after updating to 576.02
Deserves special attention temperature sensor error, due to which the system could not correctly control voltage and frequencies GPU - This could affect performance and stability. The issue has now been resolved.
Company G.Skill presented first 256GB DDR5 UDIMM memory kit capable of running at frequency of 7000 MT/s. It non-server solution — the kit is tested on a regular AM5 platform with processors Ryzen 7 9800X3D and Ryzen 9 9900X.
The set consists of four modules of 64 GB each (64GB x4), making it a direct competitor to the recently released 256GB kit from V-Color. However, while V-Color operates at 6000 MT/s with CL38 latencies, G.Skill reaches the same 6000 MT/s with low latency CL32, albeit with higher secondary timings.
The main difference is platform. While V-Color was tested with a HEDT system based on TRX50 and AMD Threadripper, G.Skill works on mainstream boards: ASUS ROG Crosshair X870E Hero and MSI MPG X870E Carbon WiFi. Moreover, on the MSI MEG X870E Godlike motherboard, using Ryzen 7 9800X3D, succeeded overclock memory to 7000 MT/s, with delays amounting to CL38-50-50-126.
Additionally, G.Skill demonstrated on the budget MSI MAG B850M Mortar WiFi 6400 MT/s on the same modules - an indicator previously unavailable for such volumes of RAM on AM5. This proves that high memory capacity is no longer the prerogative of server solutions and workstations.
Despite the impressive frequencies, it is worth remembering that increasing the timings can reduce the performance gain to zero. However, such volume and speed are especially relevant for tasks related to AI, modeling and data processing. G.Skill plans to make kits available not only to enthusiasts, but also to the mass consumer, simplifying access to high-speed memory for workloads.
After the release of NVIDIA graphics driver 576.02, which was supposed to fix the problems with stability and black screens, users encountered new bug: the video card temperature sensor freezes at 21°C, most often after exiting sleep mode or a cold start.
On Reddit, the owners GeForce RTX 5070 Ti, 4070 and other RTX 30/40 series cards they are reporting en masse that in Task Manager and Open Hardware Monitor temperature GPU does not change — it freezes at 21°C, regardless of the load. At the same time, the fans begin to work more actively, which indicates a real increase in temperature, just data is not being transmitted correctly.
The reason, judging by everything, is the interface failure NvAPI_GPU_GetThermalSettings, which is responsible for transmitting temperature data. If a failure occurs, the video card cannot correctly regulate frequencies and voltage, which in the long term is capable of reduce performance.
The only reliable solution is to restart your PC. MSI Afterburner author Alexey "Unwinder" Nikolaychuk has already confirmed that the problem is not in third-party software, but in the NVIDIA driver. A temporary solution is to avoid sleep mode and not use version 576.02. You can also roll back to version 572.xx or 566.xx, if you encounter this bug.
Company SPARKLE, which is a partner Intel for the release of video cards, denied previously published information about the development Arc Battlemage GPU with 24 GB of video memory. Let us recall that a month ago, in the Chinese social network bilibili A SPARKLE representative directly pointed out the existence of such a model, which caused a wave of discussions on the Internet.
According to the initial report, the graphics card was supposed to be aimed at professional segment, and not the games market. At the same time, it would be based on BMG-G21 chip, as already confirmed Arc B580 with 12GB of memory. Double the memory capacity could make the 24GB model useful in AI and rendering tasks.
However, soon after the surge of interest, SPARKLE deleted the post and issued a refutation, stating that such a video card was not in the plans. However, the page with the refutation also disappeared soon after - which hints at a possible NDA violation or premature leaking of information.
At the moment Intel has not commented on the situation., and there was no official announcement of a model with 24 GB of VRAM. According to rumors, if the card does come out, it could be introduced in May or June under the name Arc Battlemage Pro 24GB.
Despite official denial, possibility of existence such a model is maintained. SPARKLE's refusal to confirm may be due to an attempt limit leakage until the embargo is lifted.
