Posts: 2,046 +234
Highly anticipated: Today Nvidia and Intel are making a joint hardware launch announcement that mostly concerns the gaming laptop market. On the Intel side, we have new 11th-gen Core Tiger Lake H45 processors for gaming laptops and high performance productivity systems. Then from Nvidia, there's new GeForce RTX 3050 and RTX 3050 Ti GPUs for laptops that pack RT features and DLSS.
These new 11th generation Core H45 processors have been long awaited as they will finally bring Intel’s Tiger Lake designs up to 8 cores, suitable for gaming and workstation class of systems and what should hopefully be more competitive with AMD’s Ryzen 5000 line-up.
The new GeForce RTX 3050 Ti and the RTX 3050 laptop GPUs are interesting for a number of reasons. Just like we saw with the RTX 3060 laptop GPU before, Nvidia is announcing the mobile part before they launch the same product on the desktop, which is something they don't do often.
We have to guess that the launch of lower-tier desktop GPUs has been impacted by insane demand for graphics cards, where there's currently no incentive to sell a cheaper product when RTX 3090s are flying off the shelves for $1,500 a pop.
The RTX 3050 series is based on a new GPU die from Nvidia manufactured using Samsung’s 8nm process, so it’s not a cut down version of the existing GA106 being used in the RTX 3060. This is reminiscent of the Pascal era, where Nvidia’s GTX 1060 variants were all GP106, then the GTX 1050 series launched using GP107. Nvidia didn’t provide an official die name for this new GPU but GA107 is likely.
The RTX 3050 Ti features 2560 CUDA cores in 20 SMs, so we see 20 RT cores and 80 Tensor cores as a result. Meanwhile, the RTX 3050 is a cut-down variant with 2048 CUDA cores and 16 SMs, with 16 RT cores and 64 Tensor cores. We don’t know if the 3050 Ti is a using a fully unlocked die, but this die does feature all the RTX features and the latest generation of encoders and PCIe technology, so this is the first time that Nvidia’s mid-to-entry level die includes ray tracing acceleration.
Nvidia is giving OEMs the ability to configure these GPUs from 35 to 80 watts, which is unfortunate as there will be a massive range of performance output under the exact same GPU name. Boost clock speeds for the 3050 Ti range from 1035 MHz to 1695 MHz, suggesting the 80W model could be up to 64% faster than the 35W model which is just insane. Similar margins are seen with the RTX 3050, 1057 MHz at 35W up to 1740 MHz at 80W.
The memory subsystem used for the RTX 3050 series is identical to that of the GTX 1650 series that it’s replacing: 4GB of GDDR6 on a 128-bit bus, though clock rates are currently unknown. 4 GB of memory has been used in Nvidia’s 50 class as far back as Pascal with the GTX 1050 in 2017, so there's been no progress on that front. At the same time, with the RTX 3060 using 6GB, it would be a bit weird for a lower-tier model to have more VRAM.
With this sort of core configuration, 20 SMs, the RTX 3050 Ti Laptop GPU sits between the GTX 1650 Ti with 16 SMs, and GTX 1660 Ti with 24 SMs in terms of layout. However, with Nvidia’s new double-FP32 layout and other Ampere enhancements, we should be seeing more performance from a lower SM count than Turing just like with desktop Ampere. Similarly, the RTX 3050 has the same 16 SMs as the GTX 1650 but double the overall CUDA core count, so performance won’t be double in games but will be faster.
Before looking at Nvidia’s performance claims for their mobile part, what does this mean for the desktop RTX 3050?
Based on what we’ve seen this generation so far, it’s likely the RTX 3050 will feature a different SM count than either laptop GPU. We’re possibly looking at 18 SMs as an example. For reference the RTX 3060 on laptops used 30 SMs versus 28 on the desktop, and the RTX 3070 Laptop GPU used 40 SMs versus 46 on the desktop.
What we do know however is that GA107’s 128-bit GDDR6 memory bus will restrict these GPUs to either 4GB or 8GB of memory, and let’s hope for the desktop cards that figure is 8GB. We also know that 3050-tier GPUs will feature hardware accelerated ray tracing, unlike the previous generation where Nvidia split Turing off into the GTX 16 series without RT or Tensor cores in the lower end.
Nvidia’s information about RTX 3050's performance was very light, instead focusing on how awesome their RTX features are, given that the RTX 3050 supports them and the GTX 1650 did not. Below you can see the single performance slide we got, giving us a look at performance in two games without ray tracing, three with ray tracing, and all with DLSS. Nvidia tested these titles at 1080p using medium settings, DLSS Quality and medium ray tracing where applicable.
The sample size is too low to draw any firm conclusions, but basically what Nvidia is claiming here is that with DLSS enabled, the RTX 3050 Ti Laptop GPU is capable of 1080p medium quality gaming with ray tracing at 60 FPS. That’s a fairly typical budget laptop level of performance that you’d want. As for comparisons between the 3050 Ti and the GTX 1650 Ti, the two games shown here highlight 50 to 60 percent better performance.
In our previous performance testing, we’ve seen that the GTX 1660 Ti is about 40 percent faster than the GTX 1650 Ti in laptops, while the RTX 2060 is about 50% faster. We’ve also seen that generally Nvidia with Ampere are giving us in their new products performance equivalent to a last-gen GPU from the tier above, so the RTX 3060 Laptop GPU is roughly on par with an RTX 2070 Super, the RTX 3070 is similar to an RTX 2080 and so on. So it would make sense that the RTX 3050 Ti falls around the performance of an RTX 2060, which is not bad for budget-class laptops. And I should stress here that these are all mobile comparisons: Nvidia’s mobile GPUs are generally several tiers lower than what we get on the desktop in terms of performance.
Also read: Nvidia RTX 3070 Laptop vs Desktop GPU Review
Nvidia didn't provide any performance data on the RTX 3050 Laptop GPU, but if we had to guess, it’s probably close to the GTX 1660 Ti outside of ray tracing.
The new GPUs are expected to show up in laptops starting at $800 with designs from all the major brands. We can also expect to see a new range of Studio-focused products that will use the RTX 3050 and RTX 3050 Ti, like the Dell XPS 15, which isn’t as much of a gaming laptop as it is for creators.
Intel Tiger Lake H45
Intel’s fully fledged H-series processors for gaming laptops and productivity beasts is finally out the door. This is the first time we're getting 8-core Tiger Lake CPUs, and the goal here is to take on AMD’s impressive Ryzen Mobile 5000 series in both gaming and applications. This is also the first time we’re seeing 10nm processors used outside of ultraportable low power classes, in this case Intel’s 10nm SuperFin process technology.
From an architectural standpoint, Intel’s 11th-gen H-series processors are similar to Tiger Lake U-series parts in many ways. The CPU cores are the same Willow Cove designs, so Intel are claiming a 19% IPC improvement compared to the prior generation, which in this case is Comet Lake given 10th-gen H-series didn’t use Ice Lake and was still on 14nm using a Skylake derivative architecture.
We are also seeing an upgrade to PCIe 4.0 connectivity, built-in Thunderbolt 4 support, and Intel Xe integrated graphics – all the major improvements Intel already made with low-power Tiger Lake last year.
To fit in 8 CPU cores into a reasonable size, Intel have reduced the size of the iGPU from 96 execution units down to 32, which makes sense as most H-series laptops will include discrete graphics, making the iGPU only necessary for some hardware acceleration features.
Speaking of discrete graphics, Tiger Lake H45 includes 20 PCIe 4.0 lanes direct from the CPU, enabling x16 access to discrete graphics in addition to x4 direct to an M.2 SSD. The chipset included on the package, Intel’s 500-series mobile PCH, provides additional lanes and connectivity. Like Tiger Lake U-series, the PCH is off-die but on the same package, which is an improvement on 10th-gen where the PCH was a separate package that also needed to be fit onto the motherboard somewhere.
AMD’s approach in contrast has everything integrated into the one die as an APU, so there is no separate PCH. While this is a good approach from a size perspective, Ryzen Mobile 5000 uses last-gen technology, such as PCIe 3.0 instead of PCIe 4.0 that Intel are providing with Tiger Lake. In a sense, this is a significant overhaul on Intel’s side of every aspect to their H-series offerings.
This is how the line-up stacks up: 5 processors, ranging from a Core i5 up to a Core i9. The flagship is the Core i9-11980HK, which brings 8 cores, 16 threads and 24 MB of L3 cache. It has a base clock of 2.6 GHz and can hit 5.0 GHz on up to two cores in ideal situations. You’ll see some other clock speeds listed in the charts for things like all-core turbo, although it should be noted that most of these clock speeds will only be hit briefly in burst applications before dropping down to a lower clock speed to fit within the power limits as set by the OEM. The 11980HK certainly won’t be hitting 4.5 GHz indefinitely like you might expect from a desktop CPU.
There’s another Core i9 part listed here, the Core i9-11900H which is a 100 MHz lower clocked version of the 11980HK, and without the K suffix, so it won’t have the same fully unlocked overclocking capabilities.
Then there is a single Core i7 part, also with 8 cores and 16 threads. It’s lower clocked, with a 2.3 GHz base and 4.6 GHz turbo, so clock speeds on this part are 300 MHz lower than the flagship Core i9 but on the same CPU configuration.
For mainstream laptops we then have two Core i5 processors, the Core i5-11400H and Core i5-11260H. Both are six core designs with 12 threads and 12 MB of L3 cache, half of higher tier products. The difference between these processors is simply 100 MHz of clock speed, with the 11400H coming in at 2.7 GHz base and 4.5 GHz maximum turbo.
All Tiger Lake H45 processors have a default TDP of 45W, but can be configured down to 35W for lower power systems, or in some cases 65W or higher for beastly gaming machines. All support DDR4-3200 as the base memory spec, and all feature the same iGPU layout and clock speeds. Intel are also providing new features in their Extreme Tuning Utility, XTU, including per core voltage controls on some models and overclocking on the 11980HK.
There are some interesting takeaways here in terms of clock speeds: the maximum frequency Intel appears to hit on 10nm SuperFin with this Willow Cove design is 5.0 GHz, down from 5.3 GHz previously with Comet Lake. However Intel expects that 6% clock deficit to be nullified by higher IPC, so overall performance in lightly threaded applications should be higher.
The base clock for the 11980HK is also higher than the 10980HK, indicating we should get the double whammy of higher operating clock speeds and higher IPC, improving performance in both regards outside of the boost state.
With that said, clock speeds aren’t always faster. The Core i5-11400H, Intel’s new fastest six-core CPU in their Tiger Lake line, only goes up to 4.5 GHz compared to the 5.1 GHz that was possible with the Core i7-10850H. That’s a 12% reduction, which paired with the same 2.7 GHz base clock will make for an interesting gen-on-gen comparison. Although I guess it should be noted here that the 10850H was a Core i7 processor, in this new generation the Core i7 is an 8-core model. So comparing the exact same class of CPU, 11400H versus 10400H, we get a much needed boost from 4 cores to 6 cores.
Comparing the same class of CPU, 11400H versus 10400H, we get a much needed boost from 4 cores to 6 cores.
Intel was very light on productivity performance information. They provided just three percentage improvement figures for 11th-gen over 10th-gen, and the same three figures for the 11980HK versus Ryzen 9 5900HX. This tells us virtually nothing until we can put these to the test.
Later in the presentation, we did see a few more performance data points comparing the Core i9-11950H, which is their vPro commercial equivalent of the 11980HK, to the Core i9-10885H, another eight core CPU from the prior generation. Here in SPEC workloads they are showing 11-12% higher single-thread performance and up to 29% faster multi-thread performance.
If these numbers are somewhat accurate and representative of Tiger Lake’s Core i9 productivity performance (remember: manufacturer provided benchmarks) -- then Intel’s new Core i9 may not beat AMD’s Ryzen 9 in multi-threading, while single-threaded performance may be a close call. It's still a big improvement over Comet Lake, but AMD is dominating multi-threaded performance in 45W laptops.
What Intel seems to be much more confident about is gaming performance. Intel showed performance slides comparing the 11980HK to the 10980HK in 7 titles, showing performance improvements between 6 and 21 percent when paired with an RTX 3080 Laptop GPU at up to 155W.
Intel also showed a slide comparing the 11980HK to the Ryzen 9 5900HX, showing a performance lead of 11 to 26 percent with the same RTX 3080 Laptop GPU. This is a much more informative selection of data than we got in the productivity section, and if these results are accurate, Intel will have an impressive lead in gaming. Of course, it will be interesting to see how these stack up under independent testing where we can control all variables like power level configuration.
Intel also compared the Core i5-11400H against the Ryzen 9 5900HS, with the results slightly in favor of Intel using GeForce RTX 3060 graphics in a slim form factor laptop. So that will be yet another interesting thing to explore when we get these systems in for review.