Intel Announces Celeron, Pentium Gold And Silver Processors

[Guest]

Intel announced new Pentium and Celeron Silver processors.

Intel Announces Celeron, Pentium Gold And Silver Processors : Read more

 

[bit_user]

The processors also feature 6 lanes of PCie 2.0 connectivity.

Still? C'mon, Intel. These are launching in 2018, which should also see the arrival of PCIe 4.0. On a more practical level, NVMe and dGPUs are two good reasons to have 3.0 in these SoCs. Otherwise, they're not bad for the money.

 

[RememberThe5th]

Now the old generation comes $

 

[cmvrgr]

super for gpd pocket 2

 

[cmvrgr]

nice cpus for gpd pocket 2

 

[wifiburger]

haha that's it ? just wifi on the cpu die ? seriously, !

 

[alextheblue]

The processors also feature 6 lanes of PCie 2.0 connectivity.

Still? C'mon, Intel. These are launching in 2018, which should also see the arrival of PCIe 4.0. On a more practical level, NVMe and dGPUs are two good reasons to have 3.0 in these SoCs. Otherwise, they're not bad for the money.

You're expecting too much out of these Pentium/Celeron-branded Atom SoCs. Only the Gold is Core based. Instead of NVMe think eMMC. Also dGPUs? I don't think that's worth thinking about, not on these superbudget Atoms.

 

[alextheblue]

super for gpd pocket 2

If they build a new one they should at least stick a Pentium Gold or other Core-based "Y" series chip in it. Or drop the price.

 

[frodo29479]

The GNA block is described here https://sigport.org/documents/implementation-efficient-low-power-deep-neural-networks-next-generation-intel-client

 

[bit_user]

The processors also feature 6 lanes of PCie 2.0 connectivity.

Still? C'mon, Intel. These are launching in 2018, which should also see the arrival of PCIe 4.0. On a more practical level, NVMe and dGPUs are two good reasons to have 3.0 in these SoCs. Otherwise, they're not bad for the money.

You're expecting too much out of these Pentium/Celeron-branded Atom SoCs.

I disagree. The CPUs do have enough horsepower to drive entry-level discrete GPUs.

Instead of NVMe think eMMC.

The 6 W variants show up in many low-cost notebooks that probably have a x2 M.2 slot and would therefore benefit from PCIe 3.0.

Also dGPUs? I don't think that's worth thinking about, not on these superbudget Atoms.

They've come a long way, since the original Atom. Think Core2, and you're in the right ballpark.

It's a shame Tom's won't review anything Goldmont-based. I keep asking, but they won't touch it. Here's a nice little HTPC board based on the top-end of the current generation:

http://www.asrock.com/mb/Intel/J4205-ITX/index.us.asp

 

[Olle P]

I wonder if this new naming scheme is intended to add even more confusion in the low end segment...?

 

[Gillerer]

I wonder if this new naming scheme is intended to add even more confusion in the low end segment...?

I have struggled to make sense of the old naming conventions - which parts are high performance, which low power. Now it'll be easy: Gold is Core-based, Silver is Atom-based.

About how the Silver parts could do with PCIe 3? These are low power parts that would be served more than adequately by a SATA or PCIe 2 x2 SSD. Also, Intel are good at product segmentation. If you want a fast NVMe, they'd rather sell you a Gold part, or even better, a Core i3.

 

[bit_user]

About how the Silver parts could do with PCIe 3? These are low power parts that would be served more than adequately by a SATA or PCIe 2 x2 SSD.

Depends on what for. The main argument is graphics. These CPUs are comparable to Core2 quad - more than adequate to power a GTX 1050 or RX 560, in a number of games.

Intel are good at product segmentation. If you want a fast NVMe, they'd rather sell you a Gold part, or even better, a Core i3.

That may be, but these are efficient enough to be passively cooled with a simple heat sink. Passively cooling anything with much higher TDP is going to be much more involved/expensive.

IMO, it's unnecessary. Almost nobody who'd otherwise buy a Pentium Gold is going to instead buy a Silver just because it has PCIe 3.0 instead of its current 2.0. If you really believe what you said about 2.0 not holding it back, then you should agree that 2.0 vs. 3.0 isn't a useful measure to enforce market segmentation.

 

[alextheblue]

I disagree. The CPUs do have enough horsepower to drive entry-level discrete GPUs.

It just doesn't make a sense to buy a chip with performance from a decade ago because it can run many games (at modest settings), and pair it with a discrete GPU. It's a waste of money. Every nickel you spend on the Atom chip and board is FAR better spent towards a modern Core chip, if gaming is important. A CFL i3 or Ryzen 3 is like a rocket ship compared to this, for a relatively small investment (comparing complete build costs). Runs everything, more future proof, upgradeable, etc. But for budget builds even the cheapest 1151 socket 2C4T chips would be substantially better, cost peanuts more (total system), and has more upward mobility as a platform.

The Atoms are fine for low power non-gaming machines. If you must pair it with a GPU, go REALLY cheap, like 1030. You probably don't even need much PCIe bandwidth for that. They should test it. SATA is good enough too, given that an ultra budget system is either going to use a spinner or a really cheap SSD. Entry-level M.2 are often SATA anyway, and even the "affordable" ones that aren't don't have amazing performance above their SATA counterparts in typical RW usage.

 

[bit_user]

I disagree. The CPUs do have enough horsepower to drive entry-level discrete GPUs.

It just doesn't make a sense to buy a chip with performance from a decade ago because it can run many games (at modest settings), and pair it with a discrete GPU.

You just criticized quite a few HTPCs.

Goldmont (and now Goldmont+) is the most power-efficient CPU architecture Intel makes, and also their most cost-effective product. If you want passive cooling and don't want to break the bank, it's the best option.

And a lot of HTPCs still use dGPUs. You can still get a few decent entry-level cards that are passively-cooled.

If you must pair it with a GPU, go REALLY cheap, like 1030. You probably don't even need much PCIe bandwidth for that. They should test it.

If you only use PCIe 2.0, then they would have to add more lanes. Remember that this doesn't have DMI, so the x6 lanes are also used for things like Ethernet and wifi. My Goldmont board has only a single PCIe 2.0 x1 slot - not fast enough for any dGPU. Really, PCIe 3.0 x4 would be fine.

 

[bit_user]

Look, if you really believe it would add zero value to anyone I can accept that's what you think. But you're not about to change my mind that PCIe 3.0 and/or more lanes would help some. So, we should probably call it a draw.

 

[alextheblue]

I really feel that the need for a dGPU is decreasing in HTPC applications (which are themselves a relative rarity amongst consumers given the numerous boxes you can hook up to a TV that handle media). Total power draw of a system with an iGPU will be comparable, and there are potential cost and form factor benefits. Even a 35W Bristol Ridge A12-9800E is worth a look, let alone some future APU based on Zen+Vega. There are also cases where having more CPU punch also comes in handy when decoding certain content, even with GPU offloading, or performing other tasks. But if you still need a fanless system with a fanless dGPU, utilizing a cheap LP Atom with a full-length slot: I believe that PCIe 2.0 x6 supplies enough bandwidth for an entry level card, should they supply most of it to said x16 slot.

https://www.techpowerup.com/reviews/NVIDIA/GTX_980_PCI-Express_Scaling/21.html

Look at the x4 2.0 results. Better than you anticipated, yes? On a 980 no less, and paired with an i7-4770K clocked at 4.2 GHz. I would imagine that setup certainly outclasses a less-demanding true entry-level fanless 1030 or 550/560 paired with a "Pentium Silver". Thus the impact would be lessened further.

Shame no DMI, but network etc would have a very small impact on available bandwidth. I would highly recommend they use native SATA for the SSD. For such a budget build where a more costly chip just isn't appropriate I again believe it will have very little impact in RW use.

Look, if you really believe it would add zero value to anyone I can accept that's what you think. But you're not about to change my mind that PCIe 3.0 and/or more lanes would help some. So, we should probably call it a draw.

I will agree that some will want it, and it's possible someone would really benefit, though that's probably an edge case. I agree it will have some impact that could be measured in a benchmark. I don't think it's worth fussing at Intel over. I'm not attempting to "win" anything. I do not care about that. If I valued that, I would never go against the Opinion of the Masses, knowing that I would be excoriated for daring to question Doctrine.

Anyway, I respect your opinion, even if I do not agree. If after reviewing the above benchmarks and commentary you still strongly feel this thing needs PCIe 3.0, then yes... agree to disagree.

 

[bit_user]

I believe that PCIe 2.0 x6 supplies enough bandwidth for an entry level card, should they supply most of it to said x16 slot.

https://www.techpowerup.com/reviews/NVIDIA/GTX_980_PCI-Express_Scaling/21.html

Look at the x4 2.0 results.

Again, the SoC has x6 total. So, you don't get x6 for a GPU or probably even x4. We might be talking about a GPU slot with only x1 or x2.

Another consideration is that certain bus bottlenecks might be even more apparent, with a slower CPU. When rendering a frame, the CPU has to do a certain amount of preparation before sending commands and data to the GPU. The slower their link, the longer it'll take before the GPU can start working on the frame. So, the latency from a slow CPU and slow GPU connection just gets added.

Shame no DMI, but network etc would have a very small impact on available bandwidth.

You're missing the point. This has no South Bridge (i.e. no external PCIe switch), so each peripheral is going to eat at least an entire lane. Like I said, my J4205 mini-ITX board has only a single x1 PCIe 2.0 slot. I think if they had the lanes to spare, they'd have made it x2 or x4.

 

[bit_user]

As for performance, here's one comparison between a Sandybridge i3-2350M (2 cores, 4 threads @ 2.3 GHz; 35 W) w/ dual-channel memory vs. Gemini Lake Pentium Silver N5000 (4 cores, 4 threads @ 1.1/2.7 GHz base/turbo; 6 W) using only single-channel memory. Both machines are laptops.

http://browser.geekbench.com/v4/cpu/compare/5409790?baseline=5315949

If it had dual-channel memory, I bet it'd be an across-the-board sweep.

 

[alextheblue]

You're missing the point. This has no South Bridge (i.e. no external PCIe switch), so each peripheral is going to eat at least an entire lane. Like I said, my J4205 mini-ITX board has only a single x1 PCIe 2.0 slot. I think if they had the lanes to spare, they'd have made it x2 or x4.

Any board manufacturer eccentric enough to couple this Atom with a full size x16 slot could use a switch themselves.

As for performance, here's one comparison between a Sandybridge i3-2350M (2 cores, 4 threads @ 2.3 GHz; 35 W) w/ dual-channel memory vs. Gemini Lake Pentium Silver N5000 (4 cores, 4 threads @ 1.1/2.7 GHz base/turbo; 6 W) using only single-channel memory. Both machines are laptops.

http://browser.geekbench.com/v4/cpu/compare/5409790?baseline=5315949

If it had dual-channel memory, I bet it'd be an across-the-board sweep.

Yeesh, N5000 is slower than I anticipated. Though it is Geekbench which is... well... it's Geekbench. So who knows. Maybe it's actually not that bad.

Either way, there is literally nothing I could say that would matter. So while I feel that anyone would be better served with a non-Atom solution for an HTPC (such as an APU), you feel that this Atom plus a ton of PCIe bandwidth and a graphics card would be the best solution. That's an insurmountable disagreement, so I'll just stop here. Agree to disagree.

 

[bit_user]

Yeesh, N5000 is slower than I anticipated. Though it is Geekbench which is... well... it's Geekbench. So who knows. Maybe it's actually not that bad.

I was impressed at how well it did against Sandybridge, when I was still thinking Core 2 ballpark. How fast did you think it'd be?

you feel that this Atom plus a ton of PCIe bandwidth and a graphics card would be the best solution. That's an insurmountable disagreement, so I'll just stop here. Agree to disagree.

You're exaggerating my position. I just think that some applications of it could benefit from more and/or faster PCIe lanes.

Non-gaming HTPCs would do fine with this chip (or many of its predecessors), alone, and no dGPU. This can decode 10-bit HEVC 4k to HDMI 2.0. That's nothing to sneeze at.