NVIDIA GeForce GTX 1650 Released: TU117, 896 Cores, 4 GB GDDR5, $150

[btarunr]

NVIDIA today rolled out the GeForce GTX 1650 graphics card at USD $149.99. Like its other GeForce GTX 16-series siblings, the GTX 1650 is derived from the "Turing" architecture, but without RTX real-time raytracing hardware, such as RT cores or tensor cores. The GTX 1650 is based on the 12 nm "TU117" silicon, which is the smallest implementation of "Turing." Measuring 200 mm² (die area), the TU117 crams 4.7 billion transistors. It is equipped with 896 CUDA cores, 56 TMUs, 32 ROPs, and a 128-bit wide GDDR5 memory interface, holding 4 GB of memory clocked at 8 Gbps (128 GB/s bandwidth). The GPU is clocked at 1485 MHz, and the GPU Boost at 1665 MHz.

The GeForce GTX 1650 at its given price is positioned competitively with the Radeon RX 570 4 GB from AMD. NVIDIA has been surprisingly low-key about this launch, by not just leaving it up to the partners to drive the launch, but also sample reviewers. There are no pre-launch Reviewer drivers provided by NVIDIA, and hence we don't have a launch-day review for you yet. We do have GTX 1650 graphics cards, namely the Palit GTX 1650 StormX, MSI GTX 1650 Gaming X, and ASUS ROG GTX 1650 Strix OC.

Update: Catch our reviews of the ASUS ROG Strix GTX 1650 OC and MSI GTX 1650 Gaming X



View at TechPowerUp Main Site

 

[W1zzard]

Started benchmarking now

 

[jabbadap]

Started benchmarking now

What an odd release. Did Nvidia give any performance numbers for it in review guide? I'm not even been able to found product page for the card on nvidia.com.

 

[W1zzard]

What an odd release. Did Nvidia give any performance numbers for it in review guide? I'm not even been able to found product page for the card on nvidia.com.

no guide, no presentation, no briefing, no driver, just a short email with some basic specs and marketing speak

and i know why, the performance just isn't there:

 

[64K]

Well, this is embarrassing for Nvidia. MSRP is set at $20 more than a RX 570 and the 570 is faster as well.

 

[Ibotibo01]

no guide, no presentation, no briefing, no driver, just a short email with some basic specs and marketing speak

and i know why, the performance just isn't there:

I think it is just new GPU.

 

[sutyi]

I think it is just new GPU.

Nope it's just bandwidth starved. This should've came with the lowest bin of GDDR6 or at least a 160-bit memorybus with 5GB GDDR5.

 

[Kapone33]

Nope it's just bandwidth starved. This should've came with the lowest bin of GDDR6 or at least a 160-bit memorybus with 5GB GDDR5.

This has the same amount of memory as the RX 570 4GB model so it has to be more than bandwidth.

 

[B-Real]

no guide, no presentation, no briefing, no driver, just a short email with some basic specs and marketing speak

and i know why, the performance just isn't there:

So in an NV supported title, where the GTX 1060 6GB beats the otherwise equal or bit better RX580 8GB, the RX570 4GB is 13% faster than the 1650. So in the end, the RX570 4GB may be 16-17% faster.

 

[sutyi]

This has the same amount of memory as the RX 570 4GB model so it has to be more than bandwidth.

RX 570 4GB:
256-bit, 8x 32bit 4Gbit - 224GB/s

GTX 1650 4GB:
128-bit, 4x 32bit 8Gbit - 128GB/s

43% less bandwidth attributes to most of it, probably having less then half TMUs doesn't help either. This would be a nice 109-119USD card, too bad it has the MSRP of 149...

 

[Kapone33]

RX 570 4GB:
256-bit, 8x 32bit 4Gbit - 224GB/s

GTX 1650 4GB:
128-bit, 4x 32bit 8Gbit - 128GB/s

43% less bandwidth attributes to most of it, probably having less then half TMUs doesn't help either. This would be a nice 109-119USD card, too bad it has the MSRP of 149...

I did not see the 128 bit bus.

 

[jabbadap]

My dear god. Laptop gtx 1650 is full tu117.

Spoiler: gtx 1650 Mobile specs

 

[A.Stables]

This has the same amount of memory as the RX 570 4GB model so it has to be more than bandwidth.

Memory amount doesnt = bandwidth

570 spec
Memory Size 4 GB
Memory Type GDDR5
Memory Bus 256 bit
Bandwidth 224.0 GB/s

The GTX 1650 is based on the 12 nm "TU117" silicon, which is the smallest implementation of "Turing." It is equipped with 896 CUDA cores, 56 TMUs, 32 ROPs, and a 128-bit wide GDDR5 memory interface, holding 4 GB of memory clocked at 8 Gbps (128 GB/s bandwidth). T

almost double bandwidth, anyone who buys this card is a fool

 

[sutyi]

My dear god. Laptop gtx 1650 is full tu117.

Spoiler: gtx 1650 Mobile specs

View attachment 121535

Guess there is a GTX 1650 Ti inbound with 1024CU-s and either GDDR5X or GDDR6 around 10Gpbs.

 

[THANATOS]

Did anyone expect a RX570 killer or something? From the specs It was clear It will be slower than RX570 4GB or GTX1060 3GB!
Yes, RX570 has better absolute performance and price/performance. Anyone who is not limited by space or wants a super energy efficient gpu should choose It over this new GPU.
On the other hand the absolute performance of this gpu is ~30% higher than 1050Ti and ~55% higher than GTX1050 while costing a bit more than GTX1050Ti and consuming within 75W, that's not bad for a successor.

 

[A.Stables]

Well, this is embarrassing for Nvidia. MSRP is set at $20 more than a RX 570 and the 570 is faster as well.

Not embarrassing, Nvidia are just releasing this knowing that people will buy it, No press drivers??? would they have done this with RTX 2080ti on launch? People they see the brand, see turning, see the bullshit marketing (faster then the rip off 1050ti (what about the actual competition in the market)) and if your not upto speed in the GPU sector and generally most peple really arent & the way the press have reacted tells the story. Remember that NDA people? GPP? 680? 580? 5900? 970? bumpgate. shudder Geforce 5200. 1040? DDR4?
Rant over/ AMD are no saints but come one press, are you scared to call Nvidia for what they are?


bad day? Yes

Did anyone expect a RX570 killer or something? From the specs It was clear It will be slower than RX570 4GB or GTX1060 3GB!
Yes, RX570 has better absolute performance and price/performance. Anyone who is not limited by space or wants a super energy efficient gpu should choose It over this new GPU.
On the other hand the absolute performance of this gpu is ~30% higher than 1050Ti and ~55% higher than GTX1050 while costing a bit more than GTX1050Ti and consuming within 75W, that's not bad for a successor.

No but the way its been put on market and marketed and launched makes my blood boil ; enable chill and limit fps on the 570 and lets see how much it pulls ;

 

[THANATOS]

My dear god. Laptop gtx 1650 is full tu117.

Spoiler: gtx 1650 Mobile specs

View attachment 121535

Being called 1650 but having 1024 cuda instead of 896? I would think 1650Ti has more cuda than just 1024. With 1152 It could be close to 1060 6GB If the bandwidth is high enough.

 

[sutyi]

Did anyone expect a RX570 killer or something? From the specs It was clear It will be slower than RX570 4GB or GTX1060 3GB!
Yes, RX570 has better absolute performance and price/performance. Anyone who is not limited by space or wants a super energy efficient gpu should choose It over this new GPU.
On the other hand the absolute performance of this gpu is ~30% higher than 1050Ti and ~55% higher than GTX1050 while costing a bit more than GTX1050Ti and consuming within 75W, that's not bad for a successor.

Not really, we expected it to be around 10% slower but with a competitive price. This ain't any of that by the looks of it.

 

[THANATOS]

No but the way its been put on market and marketed and launched makes my blood boil ; enable chill and limit fps on the 570 and lets see how much it pulls ;

RX570 is 13% faster in this Witcher 3 game, so If you limit FPS and enable chill for RX570 It could end up with worse performance than GTX 1650 while consuming more power.
This card should be very good for budget gaming in notebooks.

 

[jabbadap]

Being called 1650 but having 1024 cuda instead of 896? I would think 1650Ti has more cuda than just 1024. With 1152 It could be close to 1060 6GB If the bandwidth is high enough.

Well it's kind of the case gtx1070s: more cuda cores for mobile to get performance parity with desktop counter part for lower power consumption. 1152ccs would mean using tu116 gpu, tu117 has 1024ccs thus 1152 is out of question.

 

[THANATOS]

Not really, we expected it to be around 10% slower but with a competitive price. This ain't any of that by the looks of it.

Sorry, but not a single card from Nvidia has a competitive price/performance to RX570 so expecting this to have was too optimistic. If you compare It to GTX1660 then the price/performance is a bit better and compared to the predecessor much better.

Well it's kind of the case gtx1070s: more cuda cores for mobile to get performance parity with desktop counter part for lower power consumption. 1152ccs would mean using tu116 gpu, tu117 has 1024ccs thus 1152 is out of question.

Wow, I didn't even notice that mobile GTX1070 has more cuda than the desktop part.
Is It confirmed that full tu117 has only 1024 cuda? 1152 cuda would be 1/2 of 2070, that's why I wonder.

 

[Vayra86]

Not embarrassing, Nvidia are just releasing this knowing that people will buy it, No press drivers??? would they have done this with RTX 2080ti on launch? People they see the brand, see turning, see the bullshit marketing (faster then the rip off 1050ti (what about the actual competition in the market)) and if your not upto speed in the GPU sector and generally most peple really arent & the way the press have reacted tells the story. Remember that NDA people? GPP? 680? 580? 5900? 970? bumpgate. shudder Geforce 5200. 1040? DDR4?
Rant over/ AMD are no saints but come one press, are you scared to call Nvidia for what they are?


bad day? Yes


No but the way its been put on market and marketed and launched makes my blood boil ; enable chill and limit fps on the 570 and lets see how much it pulls ;

I guess there will always be that special someone telling people they should severely handicap performance on an AMD card to make it perform like its competitor, alongside the argument that Nvidia has it easy and AMD is fighting a difficult battle. News flash: people are smarter than this. Another one: perf/watt numbers don't lie and AMD still loses at those in a big way, even despite radical undervolting and other measures.

If you're really invested in this, point the finger at AMD for getting Polaris to the sorry state it's in today and zero development to fix it. Its no surprise that the company that dóes actually make things more efficient every round is getting sales. Yes, even just because its got a more sexy product stack.

 

[Casecutter]

What did you expect... they released Turing with slight bumps in performance over an equivalent price point (aka 2070 for $500 while 13% better than a GTX 1080) and went according to plan with little to no push back. Here's the same thing it just seen to have more effect/merit in the entry level. Here Nvidia provides perhaps a 25% improvement for like 7% price jump. So, at this point this is probably the least offensive of the Turing reveal and this is where there's the Freak-Out.

Another one: perf/watt numbers don't lie and AMD still loses at those in a big way, even despite radical undervolting and other measures.

And always from the guy who's running some OEM PSU that's more often hardly Bronze+ rated. If worrying about efficiency and gaming for more than weekly/causally presence, a good modern Gold PSU and a $130 RX 570 has more merit to those really wanting to enjoy a truly immersive gaming experience.

 

[jabbadap]

Sorry, but not a single card from Nvidia has a competitive price/performance to RX570 so expecting this to have was too optimistic. If you compare It to GTX1660 then the price/performance is a bit better and compared to the predecessor much better.

Wow, I didn't even notice that mobile GTX1070 has more cuda than the desktop part.
Is It confirmed that full tu117 has only 1024 cuda? 1152 cuda would be 1/2 of 2070, that's why I wonder.

No but it would be quite odd to having more than that when full tu116 has 3*512cc GPCs thus 1536ccs.

 

[Ruru]

Nope it's just bandwidth starved. This should've came with the lowest bin of GDDR6 or at least a 160-bit memorybus with 5GB GDDR5.

I doubt the GPU supports any wider than 128-bit. I'm also surprised if it supports GDDR6.