Dell Precision 7560 & Precision 7760 pre-release discussion

Nope it is US site, no such option 2x16 ECC

 

Sadly Dell has not announced a Thunderbolt 4 dock so far. But Lenovo has announced a TB4 dock with good specs, available August, which could be a nice alternative:
ThinkPad Universal Thunderbolt 4 Smart Dock DS (lenovo.com)
Video Port

• 2 x DisplayPort™ 1.4
• 1 x HDMI™ 2.1
• 1 x Thunderbolt™ 4

USB Port

• 4 x USB-A 3.1 Gen 2* (10 Gbps, 1 x always-on charging)
• 1 x USB Type-C™ (10 Gbps, 5V/3A)
• 1 x Thunderbolt™ 4 (40 Gbps, 5V/3A)

Video Resolution Maximum :8K
Number of Displays Maximum :4 x 4K or 1 x 8K
Audio Stereo/Microphone Combo
Max. Charging to Notebook 100W
CONNECTIVITY

• vPro™ Yes
• Ethernet RJ45
• WiFi
• Bluetooth® Yes

 

I think you will mostly notice the difference in gaming. I have the 7720 with the P5000 and the P5000 performance is just about 10-20% lower than the desktop 1070 (measured with a comparable CPU), depending on the game. I don't think the P4000 will can achieve this.

As for the price $1000 is a lot of money. If one does not game or has the VRAM requirement or does workloads that are heavily CUDA dependent, it's probably not worth it.

But I noticed that German prices are weirdly different. For the 7760, the difference between the best and worst CPU is just 58.50€. Difference between A4000 and A5000 is 565.50€. Memory and storage options tend to be cheaper than US, but base price is much higher.

Also the X17 is much more expensive in Germany. If you choose the most powerful options (with basic ssd) and Premium Support, you end up at pretty much the price for a full configured 7760 (with basic ssd). In the US there is about a $1300 difference.

 

I noticed the same thing on CPUs (on the french store), but it is a display mistake and as soon as you select different CPUs the total order amount augment much more than the increase displayed values.
That's just for CPU and for GPU from A4000 to 5000 the displayed increase is correct and much lower than in the US (ca. 565 euros vs 1170$). 565 euros is about 670$, so US customers are severely disadvantaged here. But yup EU base price is much higher...

 

Strange, it's definitely there for me.
Try selecting the base model and not the data science one. Data science has more limited options (and yet the price is the same if you spec them out identically).

Spoiler

 

You're right. I wasn't paying attention to the total. As they say: If it sounds too good to be true, it probably is.

 

I would think that the P5000 being a bit slower than a desktop 1070 is expected. Usually, mobile cards are one step above their desktop equivalents for the same performance. The 10-20% slower was probably partly caused by it being a quadro with slightly slower memory bandwidth due to ECC memory. However, I would also think that the P4000 would be able to keep up pretty good.

Keep in mind that in Blender Open Data CUDA rendering, the 3070 mobile (A4000) is only 5% slower than the 3080 (A5000) mobile. Maybe it's Blender's optimization but it seems like the extra CUDA cores don't do much to help CUDA performance (though it's just one application)
However, with Blender hardware accelerated ray tracing (OPTIX, uses RT cores), the 3080 is around 15% faster. Still not much considering the $1000+ price difference in the US.

However, in short bursts of load, the A5000 is likely to be noticeably faster, such as gaming. This is because the A5000 can take better advantage of the higher power compared to the A4000. Though, even in the best case scenario for the A5000 with short workloads, it will only be around 15% faster (based on the previous gen). For me, that's not worth the money considering that in the US, going from UHD graphics to the A4000 costs the same as going from A4000 to A5000.

 

Comes down to the power limit. Let's go back a generation.

Precision 7740 & 7750 offered the Quadro RTX 4000 and Quadro RTX 5000 GPUs with the same power limit.
The CUDA cores are the bulk of the power draw when running operations on the GPU.
The Quadro RTX 4000 (in Precision 7740/7750) hits the power limit while running a full load on the CUDA cores. (The GPU is forced to power throttle.) RTX 4000 has fewer CUDA cores than the RTX 5000.
The RTX 5000 obviously is also able to hit the power limit with more CUDA cores. There is more parallelization, but they all run at a lower speed because the same amount of power has to be spread out between more cores.
Because of this, performance between the two is similar when using CUDA only.

Normal rasterization ("3D graphics" without new tricks like DLSS or ray tracing) uses the CUDA cores only. RTX 5000 offers only a ≈5% boost or less over RTX 4000 with this type of workload. Benchmarks have borne this out, gaming and otherwise.

(BTW, the situation was also exactly the same with the Precision 7730 and Quadro P4200/P5200 GPUs. This was the first time that the same power limit had been used for the "4000" and "5000" series GPU, and it caused some confusion & frustration on here when we realized what was going on.)

Now on the Precision 7760 with RTX A4000/A5000, we don't know for sure yet but I think it's pretty likely that the situation will be the same once again. If RTX A4000 hits the power limit when running a CUDA-only load, and the power limits between the RTX A4000 and A5000 are the same... Unless you are using specific features of the RTX A5000 that aren't CUDA-related (tensor cores, RT cores, utilizing more vRAM) then you probably aren't going to get much benefit from the RTX A5000. Basically, you should not take the increased CUDA core count into consideration when deciding which of the two to purchase, or when determining relative performance expectations.

"If RTX A4000 hits the power limit when running a CUDA-only load"... this seems likely as the same GPU chip in a desktop is rated to run at well over 200W.

This whole mess is why NVIDIA is requiring that OEMs divulge the GPU power limits in GeForce 3000-based laptops. The power limit matters more for performance than which actual GPU you have, in many cases.

I don't think that this is going to be true for the reasons described above. There's no indication that the A4000 and A5000 will have a different "burst" power limit; both are rated for 140W which is the number that Dell has thrown out.

 

I didn't mean that they would have different burst power limits, I meant that the A5000 can take advantage of more power because it has more cores which can be pushed to higher clocks more efficiently. Because efficiency decreases exponentially with the increase of clock speeds, the A5000's more cores allows it to be more efficient at higher powers. But the difference in "efficiency" is very minor, only around 15%.

 

Yeah, this is basically the reason for the difference in performance between the RTX 4000 & 5000. I think that 15% is stretching it a lot. (Maybe not for desktop GPUs which go to much higher power levels.) 5% or less is what we have observed with past systems (basically unnoticeable on most workloads).